Better systems. Less chasing. Room to dream bigger.

The parse fileAll documents normalized into one indexed matter file; later additions land in the same place

File intake memoDocuments reordered into sense

Medical chronologyRecords sequenced in time; body parts, treatment gaps, and prior conditions surfaced for counsel

5 more steps

Evaluate now or cure the gaps?Counsel decides whether the record supports evaluation or discovery must fill it first

Valuation & defense postureReserve range, exposure, and strategy are counsel's alone

11 steps3 branchesFree, permanently

Open the law boards →

See the law boards → The law membership

A clause read against your playbook, and a line you may take or leave.

The contract desk reads one clause against the firm’s own negotiation positions and comes back with a verdict, the playbook line it rests on, and a suggested markup. Counsel decides whether any of it is proposed. Paste nothing; the clauses below are invented.

Contract desk: one clause, one playbook Simulated demo

The clause, invented

Playbook position

The playbook, invented

Governing law is the customer’s home state, or a state the playbook names as neutral.
Venue, in the conservative position only: no exclusive venue in the vendor’s home state.
Liability cap floor: twelve months of fees in the standard position, twenty-four in the conservative position.
Indemnity and confidentiality breaches sit outside any cap.
Assignment on a change of control needs notice in the standard position; notice and a written assumption of obligations in the conservative position.
Any suggested markup is counsel’s to propose or drop. The desk proposes nothing to the other side.

Verdict

Why, with the playbook line it rests on

Counsel decides whether to propose it. The desk never sends a redline.

Illustrative. Invented clause, invented playbook. A suggestion is never an acceptance, and nothing here is signed, filed, paid, submitted or sent.

A rule changed overnight. Here is the sentence that moved.

The regulatory monitor watches the sources a practice answers to and reports a diff, not a guess: the exact span that changed, the matters it touches, and a memo drafted for counsel and never sent by anything but a person. Both rule texts below are invented.

Regulatory monitor: what changed, and where it lands Simulated demo

Before

After

Counsel reads, edits and decides whether anything goes to a client.

Illustrative. Invented rule text, fictional matters. Nothing here is signed, filed, paid, submitted or sent.

Intake and conflicts

A lead becomes a structured record, the conflicts pass runs against a fictional client book, and the attorney gets a packet, not a verdict. It runs in the playground on the home page, as the client-intake scenario.

Open the intake scenario in the playground →

Docket and deadline watcher

A new entry appears on a docket the firm follows; the deadline set is prepared with every date citing its rule, and a question goes to the attorney when two rules disagree. It runs on the workflow demo, as the legal docket watcher.

Open the docket watcher →

A curriculum written for lawyers from the ground up.

Four quarters: The Boundary, Data Hygiene, Confidentiality and Privilege, and Verification Discipline, anchored to your own matters and taught in the shape a jurisdiction can count. How credit runs is in the note at the foot of this page.

What defense firms actually worry about, answered one by one.

“It trains on our clients.” We do not train on your data and we do not submit it for training: contractually, in any tier, as an entry in a prohibited-actions register and a term of the engagement letter. Any third-party tool in the build is named to you in writing, with its own terms, before it is used.
“Our data leaves the building.” The work runs on your firm’s own subscription and your own storage, with no server of ours in the loop. The party holding the data is the firm, as it always was.
“The logs are a second copy of the file.” Run records name sources as identifiers, never their contents. The fields that would carry prompt and output text exist in the standard and are left empty by rule.
“We’re locked in once we start.” The confidentiality answer is also an exit clause: when an engagement ends there is nothing to migrate and nothing to ransom.

ABA Formal Opinion 512, mapped to architecture.

The opinion binds nobody; your jurisdiction’s rules govern. What it is good for is structure: its duties are the committee’s checklist, and each one lands on something checkable here, not a policy PDF.

Confidentiality. Per-client isolation of credentials, storage, and memory; we do not train on your data and do not submit it for training, contractually; client-owned instances, so privileged material never sits on our infrastructure.
Competence. A tier ladder assigns each task an autonomy ceiling by risk, and the training curriculum teaches the boundary: what may be assembled, what must be judged.
Supervision. One named seat that can stop any workflow, every agent action logged, and nothing unreviewed ever reaches a court, regulator, or counterparty.
Communication. The engagement letter discloses what the systems do and who reviews it, and where client information would enter a tool, consent is specific to the tool and the risk; the opinion is plain that boilerplate is not enough.
Candor toward the tribunal. Sourced answers or a refusal: every citation traces to a real record, and nothing unreviewed is filed. A citation appears only when it resolves to a record someone can open; anything below that confidence is a refusal, and the refusal is enforced by tests.
Fees. Scope and a written ceiling agreed before work begins; billed time reflects work actually done. Efficiency the systems create belongs to the client relationship.

The AI Committee Briefing.

Fifty minutes for the whole committee on the fears, the boundary, and what to do Monday. It books directly, no diagnosis required.

Book the briefing →

The law curriculum is structured to support CLE where your jurisdiction permits: we are not an accredited CLE provider in any state, and credit runs through your jurisdiction’s own process.

Makes Eleven gives no legal advice. Which ownership or structure a firm may use is a question for the firm’s own counsel and its jurisdiction.

Every financial deliverable is inspected by a licensed CPA before it reaches you, disclosed in every engagement letter; this is advisory work, and no financial-statement audit, review or other attest report is issued.

The three desks on this page are simulations on invented records. Nothing here is signed, filed, paid, submitted or sent.

Built for firms whose work
becomes a public record.

A policy firm carries more than its clients’ issues. It carries the disclosure calendar, the record of every contact and what it was about, the memory of each client’s matters, and the gift-rule question that arrives at five in the evening. We build the systems that keep all four current, for the people who sign and file.

Talk through the work → See the disclosure calendar

A gray facade of identical square windows, one of them standing open

What we build for a policy firm.

Each piece assembles a record your people decide from. Each piece is built to run on your firm’s own AI subscription and reads only the sources you approve.

The disclosure calendar. Every registration, quarterly activity report and semiannual contribution report on one calendar, each date computed from the statute line it comes from and shown beside it. A date that lands on a weekend or a holiday moves the way the law says it moves.
The activity register. Each contact logged when it happens: who made it, which chamber or agency, which issue. The quarterly report is assembled from the register line by line, so a line with nothing behind it shows as a blank rather than a guess.
Regulatory monitoring. A bill or rule text changes; the sentence that moved is shown, and the clients it touches are listed with the reason. It is the same monitor we build for law firms: see it run on the legal practice page.
The client report. A weekly report for each client, assembled from the register and the monitor, every line traced to the entry behind it. The principal edits it and decides whether it goes out.

A curving timber stair under a warm line of evening light

The gift-rule question that arrives at five in the evening.

May the client host this staff member, at this event, on these terms? The desk assembles the facts and the text of the rule that governs, from the House and Senate ethics committees’ own published guidance, and puts both in front of the person who answers.

Build confidence and trust.

Six mechanisms hold on every build, and for a policy firm each one answers a question a client will ask about its own confidential strategy.

Per-client isolation. Credentials, storage and memory are separate for every client. Agents read only the sources you approve.
No training on your data, contractually. We do not train on your data and we do not submit it for training. That is a term in the agreement, not a checkbox in a settings panel.
The examiner is never the builder. Every system is checked by someone who didn’t build it, against test sets with known answers.
Sourced answers or a refusal. Every date and every report line traces to a record someone can open. Below that confidence the answer is a question to a person, never a guess.
Nothing filed or sent by an agent. No agent signs, files or pays. Nothing goes to an agency, a congressional office or a client without a person’s instruction to send that specific communication.
Every agent action is logged. The log is part of what you keep, and it survives the end of the engagement along with everything else.

Every deadline, and the line it comes from.

A fictional firm with three clients and one just retained. Its due dates are computed from the statute as this page opens, for this year and for anything from last year still ahead. Choose any row to see the rule and the arithmetic behind its date. Every name here is invented, nothing is stored, and nothing is signed, filed, paid, submitted or sent.

Disclosure calendar Simulated demo Computing this year’s dates…

This year’s lobbying disclosure filings for a fictional firm, one row per filing
		Period
The calendar is computed in your browser when this page opens.

The registrant files. Nothing here files.

Nothing here is signed, filed, paid, submitted or sent.

Every date and every line comes from a source you can read.

The calendar reads the statute rather than a list someone typed, and it reproduces every due date listed on the Clerk of the House’s lobbying disclosure page. The report below is built to the line items the same statute names. These are the lines both read, quoted from the source and linked.

The three forms: The LDA reports include lobbying Registrations (LD-1), Quarterly Activity Reports (LD-2) and Contributions Reports (LD-203). LDA.gov, the House and Senate filing site
Registration, LD-1: No later than 45 days after a lobbyist first makes a lobbying contact or is employed or retained to make a lobbying contact, whichever is earlier, or on the first business day after such 45th day if the 45th day is not a business day… 2 U.S.C. 1603(a)(1)
Quarterly activity, LD-2: No later than 20 days after the end of the quarterly period beginning on the first day of January, April, July, and October of each year… or on the first business day after such 20th day if the 20th day is not a business day… 2 U.S.C. 1604(a)
Contributions, LD-203: Not later than 30 days after the end of the semiannual period beginning on the first day of January and July of each year, or on the first business day after such 30th day if the 30th day is not a business day… Each filer also certifies that it has read and is familiar with those provisions of the Standing Rules of the Senate and the Rules of the House of Representatives relating to the provision of gifts and travel. 2 U.S.C. 1604(d)(1)
The gift rules: Senate Rule 35 (the Gifts Rule) is a rule of prohibition. Senate Select Committee on Ethics. In order for a House Member, officer, or employee to accept any gift from any source, it must meet one of these exceptions. House Committee on Ethics
Foreign principals: FARA requires certain agents of foreign principals who are engaged in political activities or other activities specified under the statute to make periodic public disclosure of their relationship with the foreign principal, as well as activities, receipts and disbursements in support of those activities. U.S. Department of Justice, FARA.
The states: State lobbying and gift rules vary by state. Where your firm answers to them, the calendar carries each one with its own source line, in the same shape as the federal dates above.

Concentric timber rings rising in a warm, curved ceiling

Every line in a report began as a conversation.

For your firm we build the record of your relationships: who met whom, on whose behalf, about which issue, and what followed. The quarterly report is assembled from it.

The quarterly report, assembled from the register.

One month of a fictional firm’s contacts for one client, logged as they happened. Assemble the report and every line of it points back to the entries it came from; where the register holds nothing, the line says so.

Activity register to quarterly report Simulated demo Loading the register…

The activity register: one row per contact
	Date				Specific issue	Foreign entity note
The register loads in your browser when this page opens.

The lobbyist reviews and the registrant files.

Nothing here is signed, filed, paid, submitted or sent.

Talk through the work.

Bring the calendar you keep today and the last report a client asked for. The conversation starts from those, and the diagnosis decides what, if anything, is worth building.

Talk through the work →

We give no legal advice.

Built for campaigns that
run on a deadline.

Campaigns run on a calendar nobody can move. We build the systems behind the staff and consultants you already have: research assembled from public sources with every figure sourced, outreach a staffer reviews and sends, and drafts pre-checked against the disclosure rules we cite, with anything unclear routed to your counsel. Nationwide.

Talk through the work → See the pre-check

A white wall with a narrow opening onto a staircase that climbs out of view

Build with clients, not for them.

Your consultants set the strategy; we build the tools that carry it, and a person on your staff owns every step that leaves the building.

Research from public sources. Briefs assembled from the public record, every figure linked to the page it came from. A figure with no source stays blank and is flagged for a person.
Outreach drafts. Emails and scripts drafted from your campaign’s own message, each line traced to it, for a staffer to edit and send.
Strategy and message sessions. Working sessions with your consultants, starting from a diagnosis of how the campaign runs today.
Contribution records, reconciled for the treasurer. The committee’s own records matched against its deposits, every figure recomputed by a second, independent calculation.
The disclosure pre-check. Every draft read against the published rules before a person sees it, each finding labeled Suggested, Needs a source or For counsel.
Public sentiment digests. Themes read from public posts, each with the posts behind it and a confidence word from a printed rule. A reading of public posts. Not a poll, not a sample of voters.

An empty room with a wooden table, white chairs and daylight through tall windows

Two seats hold a campaign together.

The treasurer answers for every dollar and every filing; the campaign manager decides every send. What we build puts the record in front of each of them, sourced, before they decide.

Accelerate impact and reduce risk.

Every draft is checked before a person sees it, and every step that leaves the building is a person’s, so the work moves faster and the risk stays with the seat that answers for it.

Per-client isolation. Credentials, storage and memory are separate for every client. Campaign committees are private clients, engaged separately from our work for public agencies.
Your data stays yours, contractually. We do not train on your data and we do not submit it for training. That is a term in the agreement.
The examiner is never the builder. Every system is checked by someone who did not build it, against test sets with known answers.
Sourced or a refusal. Every figure and every quote traces to a source someone can open. Where there is none, the answer is a question to a person.
A person sends. A staffer sends or posts everything that reaches a voter, a donor, a reporter or a platform, on an instruction to send that specific message. The treasurer authorizes spending and files every report.
Every agent action is logged. The log is part of what you keep, and it survives the end of the engagement.

Every draft, read against the rules before a person sees it.

A fictional fundraising email from a fictional committee. Check it and each finding lands as a row with the rule behind it; choose a row to see the rule’s own words and where the draft falls short. Every name here is invented, nothing is stored, and nothing is signed, filed, paid, posted or sent.

Compliance pre-check Simulated demo Loading the draft…

From: The fictional authorized committee of a candidate for federal office
To: 1,200 recipients on the committee’s own list
Subject: Before the quarter closes

Friend, the quarter closes in a few days, and every contribution before then shows the strength of this campaign.

The opponent said last week that “road repair can wait another year.” The candidate thinks it cannot.

Image: a photorealistic image of a road crew at work, generated with an AI tool

Chip in before midnight to keep the office open through the fall.

The campaign team

End of message

Findings from the pre-check of the fictional draft, one row per finding
	Rule
Nothing checked yet. Each finding lands as a row once you check the draft against the rules.

The campaign manager decides and a staffer sends. Nothing here is sent.

Nothing here is signed, filed, paid, posted or sent.

The rules the pre-check reads.

The pre-check quotes its sources rather than paraphrasing them. These are the lines it reads, in the sources’ own words, each linked.

Disclaimers, 11 CFR 110.11: (a)(1) (a) Scope. The following communications must include disclaimers, as specified in this section: (1) All public communications, as defined in 11 CFR 100.26, made by a political committee; electronic mail of more than 500 substantially similar communications when sent by a political committee; and all websites and internet applications of political committees available to the general public. (b)(1) (b) General content requirements. A disclaimer required by paragraph (a) of this section must contain the following information: (1) If the communication, including any solicitation, is paid for and authorized by a candidate, an authorized committee of a candidate, or an agent of either of the foregoing, the disclaimer must clearly state that the communication has been paid for by the authorized political committee; 11 CFR 110.11(a)(1) and (b)(1)
AI and fraudulent misrepresentation: This fraud may be accomplished using AI-assisted media, forged signatures, physically altered documents or media, false statements, or any other means. The statute, and the Commission’s implementing regulation, is technology neutral. Federal Election Commission, interpretive rule, 89 FR 78785
Google election ads: Google believes that users should have information to make informed decisions when viewing election ads that contain synthetic or digitally altered content, specifically images, video, or audio that inauthentically depict real or realistic-looking people or events. Advertisers must disclose all election ads that contain synthetic or digitally altered content by selecting the checkbox in the “Altered or synthetic content” section in their campaign settings. The disclosure must be clear and conspicuous, and must be placed in a location where it is likely to be noticed by users. Google advertising policies, political content
Meta election ads: Advertisers must include a verified “Paid for by” disclaimer on these ads to show the entity or person responsible for running the ad across Meta technologies. Advertisers must also disclose when a social issue, elections, or political ad contains a photorealistic image or video, or realistic sounding audio, that was created or edited using third-party generative AI tools to: Depict a real person as saying or doing something they did not say or do; or Depict a realistic-looking person that does not exist or a realistic-looking event that did not happen, or alter footage of a real event that happened; or Depict a realistic event that allegedly occurred, but that is not a true image, video, or audio recording of the event. Meta, ads about social issues, elections or politics
The states: State rules on AI in political ads vary by state; your counsel confirms which apply.

A mountain lake with pines on its shore and snow peaks behind

This week’s record, not last month’s.

Research briefs and sentiment digests are rebuilt as often as the campaign needs them, each with its sources and its date, so the person deciding reads the current record.

What the public record says this week, and how sure to be.

Twelve fictional public posts on local issues. Read them and the digest groups them into themes, each with a confidence word from the rule printed below and the posts behind it. Every post here is invented, nothing is stored, and nothing is signed, filed, paid, posted or sent.

Public sentiment digest Simulated demo Loading the posts…

A reading of public posts. Not a poll, not a sample of voters.

The confidence rule. High: in 5 or more posts from 3 or more distinct sources. Medium: in 3 or more posts from 2 or more distinct sources. Low: in 2 or more posts. Fewer than that: Too few posts to say.

Themes read from twelve fictional public posts, one row per theme
		Distinct sources	Confidence
Nothing read yet. Each theme lands as a row once you read the posts.

The communications director decides what it means. Nothing here is a poll.

Nothing here is signed, filed, paid, posted or sent.

Talk through the work.

Bring the calendar you keep today and the last message that went out. The conversation starts from those, and the diagnosis decides what, if anything, is worth building.

Talk through the work →

Makes Eleven gives no legal advice.

A licensed CPA inspects every financial deliverable, including reconciled contribution records, before it reaches you, disclosed in every engagement letter; this is advisory work, and no financial-statement audit, review or other attest report is issued.

The Ledger Suite

Six desks. One engine.
Your subscription, your storage.

Productized desks for the recurring work a practice must do perfectly and does manually. Each runs on your own AI subscription and your own Drive, with no server of ours in the loop. Every desk comes with a weekly research robot, a client desk for calls and notes, a market watch, and a packaging studio.

Colleagues working across desks in a warm-lit office

Market evidence

~650K

Actively licensed US CPAs — down from about 672,000 in 2024, a bench that is now shrinking while the work is not. Agents do volume; CPAs inspect and judge, as advisory work, with no financial-statement audit, review or other attest report issued.
Source: NASBA Accountancy Licensee Database, August 2026

The suite

One engine, six verticals.

A new vertical costs one research-and-generate cycle on the same engine, which is why the suite covers this much ground.

Ledger OS

The flagship, and the engine the other five desks run on. Built for firms whose vertical isn’t listed yet.

Ground

The real-estate desk. It holds the deal file, the diligence trail, and the recurring closes.

Counsel

For legal practices: matters, deadlines, and the accounting a trust examiner checks.

Close

Reconciliation and close discipline, for accounting firms that sell it and firms that need it.

Stewardship

Wealth management. The review cadence, the meeting file, and the evidence of care.

Practice

Healthcare delivery. It runs the operational spine behind the clinical work, and never the clinical work itself.

A single small sailboat on a wide expanse of open water, seen from above

Why that architecture matters

Nothing of yours sits on our servers.

It is the shortest answer a regulated buyer can give their own committee, and it is architecture rather than a promise. There is no copy of your file on our side to lose, to disclose, or to train on.

The flagship example

Counsel × Close: the trust-accounting desk.

Three-way trust reconciliation recomputed independently every cycle, pre-bill hygiene, settlement-disbursement preparation, and an exception watch. Agents assemble and reconcile, licensed CPAs inspect (advisory work, with no financial-statement audit, review or other attest report issued), and a human moves every dollar. Every desk in the suite delivers that same pattern for its own vertical: a recurring run, sourced figures, human checkpoints. See it in the legal practice →

What monthly buys

The maintained retainer, itemized.

The weekly run. The desk’s recurring work, delivered on cadence, every figure sourced or refused.
Model and capability currency. A nightly scout watches AI releases and capability changes; anything material to your system is flagged to your administrator with cost and benefit attached. You decide, and nothing switches itself.
Maintenance and repair. When a connected system or model changes underneath the desk, we fix the desk — that is what the retainer is for.
Everything stays yours. Your subscription, your Drive, your instance. End the retainer and the desk keeps working as-built. There is nothing of yours on our side to move or to hold.

Two ways to buy

One-time build, or desk + retainer.

Sold as a one-time productized deliverable, or as desk + customization + the maintained weekly retainer. Priced to the firm and the scope of work, and quoted after a conversation rather than from a template.

Start the conversation →

Automations · Productized, monthly

Software teammates,
run to a standard.

An agent engagement puts a software teammate on one repeatable task, end to end. We start where your workflow map ends, so nothing ships that the diagnosis didn’t support.

Start a conversation → Train your team →

A bright open-plan office in a timber-framed loft

What an automation is, and what it runs

What they run

The recurring work a firm must do perfectly and does manually: the close, eligibility checks, document intake, deadline monitoring, follow-up that never lapses. One bounded task per agent, chosen by the diagnosis.

What ships with every one

A Guardian agent that watches the others, autonomy tiers assigned per task by risk, a licensed CPA inspecting every financial deliverable (advisory work, with no financial-statement audit, review or other attest report issued), and a log of every action. Every engagement includes all of it. You don’t have to ask.

How they’re measured

Your baseline is recorded on day one; day 30 brings a one-page before-and-after against it, then the delta monthly. If it proves out we scale it, and if it doesn’t we stop.

The engine underneath is the same five stages every time — watch it run on the method page. The boundaries are contractual, in writing on the assurance page.

Behind every agent is a network: when a solution needs a capability beyond the tools we run in-house, such as a voice model, a vision pipeline or custom software, we assemble specialists for it per engagement, under the same written standard. And every maintained engagement carries 24/7 support from a team on retainer across time zones.

Four small rowboats seen from above on open teal water

The shape of it

What the fleet looks like from above.

Each agent is a crew with one bounded job, run end to end. A Guardian watches every one of them, and the fleet moves in the same direction because the diagnosis set the course.

The first ten minutes

You bring the question. We draw the machine behind it.

You bring the process that hurts. We draw it step by step and tag each step honestly: what a machine can carry, what needs a person, and what nobody should have been doing at all.

Workflow Mapper Simulated demo Mapping

Illustrative maps built from the Workflow Atlas. Your diagnostic maps your own firm’s workflows. Nothing here is signed, filed, paid, submitted or sent. Open the Workflow Atlas →

The systems you already run

We build into your stack. We don’t ask you to leave it.

We integrate by API with the systems you already run — ledgers, portals, practice management, the drive, the inbox. We don’t migrate your ledger or your CRM; the source of truth stays where it is. Which systems connect, and how, is settled in the diagnosis.

All logos, product names and company names are trademarks of their respective owners, shown solely to identify systems we integrate with. No affiliation, partnership, sponsorship, or endorsement is implied.

Related services

Where this connects.

The boundaries an agent runs inside, delivered as your firm’s written standard.

Learn more →

The same firm data your agents work from, queryable in plain English.

Learn more →

The assessment that clears the base before an agent is proposed.

Learn more →

The diagnosis that decides which task deserves an agent at all.

Learn more →

Next step

Name the task that eats the most hours.

A thirty-minute call, online. You’ll leave knowing whether the task is agent work, including when the honest answer is a checklist and one hire.

Start a conversation →

Dashboards · Recurring

Your own data,
answered or refused.

Ask your business a question in plain English and get an answer with its sources shown — or an honest refusal when the data can’t carry one. Never a guess dressed as an answer.

Start a conversation → Train your team →

How a dashboard is built

On top of what you run

The dashboard connects by API to the systems you already trust — ledger, practice management, CRM, drive. We never migrate a source of truth; the answers cite the system they came from.

Sourced or refused

Every answer names its records. Below the confidence threshold the response is a question back to you — the same house rule that governs everything we ship.

Yours, isolated

Per-client credentials, storage and memory; we do not train on your data and we do not submit it for training, contractually. Where the zero-infrastructure pattern fits, it runs on your own subscription and your own storage.

Ask in plain English

Numbers that answer for themselves.

The board question that used to take a week of assembly becomes a question you type. Every answer arrives with the records behind it — or as an honest refusal, which is its own kind of answer.

A presenter walking a small audience through a performance-analysis slide

What a dashboard actually is

Most people say “dashboard” and mean a screen they can type into.

A chart reports. An application is where the work happens: where an intake gets filed, a review gets routed, an exception gets escalated and somebody signs off. Both get called a dashboard, and confusing them is how firms buy a set of graphs when what they needed was a place to do the job. We ask which one you meant before anything is built, because the answer changes everything underneath.

Read it from the bottom. Your records stay in the storage your firm already owns, which is what lets the whole thing run inside a subscription you are already paying for, with no server of ours in the loop. The layer above does the work. The layer above that is what people actually open. The top layer is the report a partner reads on a phone. Where your firm already runs Microsoft 365 or Google Workspace, that bottom layer exists today and the build starts one floor up.

Which layers you need is a finding, not an assumption. Plenty of firms that ask for a dashboard need the second layer and nothing else, and being told that is worth more than being sold four.

Ask it something hard

The answer arrives with its receipts. Or it refuses.

Four questions a managing partner might put to their own records. Watch the answer assemble, every figure cited back to where it came from, and watch the last one get refused, because the records could not carry it.

Dashboards Simulated demo Answering

Illustrative. Your dashboard answers from your own systems, and names them. Nothing here is signed, filed, paid, submitted or sent. Start a conversation →

Ask it yourself. Eight questions, two refusals.

Pick a question. The answer assembles from invented records, every figure opens to the lines behind it, and where the records cannot carry an answer the desk says so and names what to start recording instead.

Ask the records Simulated demo

Ready

Question (simulated)

Illustrative. Eight questions, fixed on purpose: nothing you type exists here to leave the browser. Every name, record and figure is invented for this page.

Illustrative. Nothing here is signed, filed, paid, submitted or sent. Your desk reads your own systems, and names them. Start a conversation →

The view that just reports, on a phone.

Three tiles a partner reads before the first meeting. Each carries the moment it was refreshed and how many records stand behind it, because a number that hides its own staleness is the failure mode.

Morning report Simulated demo

Refreshed

Illustrative. Nothing here is signed, filed, paid, submitted or sent.

Related services

Where this connects.

When the answer to a question should become an action, run to a standard.

Learn more →

Who may ask what, and what the answers may touch — in writing.

Learn more →

Measure what your current tools actually return before adding another.

Learn more →

What a citation is worth depends on the layer it cites.

Learn more →

Next step

Bring the question you can’t answer today.

If your systems hold the data, we’ll show you what it takes to make it answerable. If they don’t, we’ll say that instead.

Start a conversation →

Governance

Adopt fast.
Answer for it.

The board wants AI moving this quarter; the regulator, the carrier and the client want proof of control. One unreviewed output in front of the wrong reader costs more than the program saved — so we deliver the controls as a service, and we start where your data governance ends, so you’re not building from scratch.

Start a conversation → Train your team →

What Governance delivers

Your written standard

Autonomy tiers per task, a prohibited-actions register, examiner-never-builder review, and logging. We draft them in your firm’s own language and sign them into your engagements. The architecture we run ourselves, published here, becomes yours.

Framework-aligned

NIST AI RMF alignment, disparate-impact testing as an engineering step, records-retention and FOIA awareness where the public is the client. When the committee asks what governs this, you hand them a document.

For the seat that answers

Built for the managing partner, compliance officer or administrator whose name is on the outcome: one accountable seat named in writing on day one, and every boundary printed before work begins.

This is the written standard. The assessment that produces the findings it governs, on both the exposure side and the people side, is AI risk and change impact.

City towers with their upper floors disappearing into fog

The altitude

What control looks like here.

When the regulator, the client or the board asks how AI is controlled here, the answer is your own written standard: autonomy tiers, a prohibited-actions register, and logs a reviewer can read.

Related services

Where this connects.

The systems these controls govern, shipped with a Guardian on every one.

Learn more →

Access and audit questions answered from your own records.

Learn more →

Where ungoverned tools are already running, find them first.

Learn more →

Governance decisions made at board level, before the build.

Learn more →

Next step

Bring the question your committee keeps asking.

Thirty minutes, online, and you leave with the shape of the answer. The assurance page above is the standard we’d hold your build to.

Start a conversation →

Strategy Consulting

The diagnosis
is the strategy.

Bespoke, judgment-heavy work done with you: how the firm runs, where the margin leaks, what to fix first and what to leave alone — written down, priced to scope, with a ceiling that holds.

Start a conversation → Train your team →

Looking up the glass and metal flank of a tower, two faces meeting at a ridge against an open sky

Market evidence

46%

Of broker-led sale engagements ended without a closing even in the strongest market the survey had recorded; most years the rate runs higher. The deal-killer we see most is messy financials, and we fix that before it costs you.
Source: IBBA & M&A Source Market Pulse, Q4 2021

What Strategy Consulting covers

Diagnosis & redesign

Intake and diagnosis, workflow redesign, GAAP readiness and compliance workflow design. The map of how work actually flows is a deliverable you keep, whether or not you continue.

Exit readiness

The Books Ladder from clean records to diligence-ready, add-backs evidenced, a data room that survives the request list. Engaged by the owner, welcomed by their broker — no referral fees in either direction.

Board-level advisory

Led by a founder with an economics degree, an MBA in finance and entrepreneurship, a banking innovation award won at a top-ten US bank, and national leadership roles across AAREP, ULI and HAND.

A pale tiled facade curving away against an open sky

How it is scoped

Priced to the firm, held to a ceiling.

There is no rate card. The engagement is priced to your firm and to the scope of the work, named in writing with a ceiling that holds for the term.

This is the engagement we call the Audit, and it is for a firm deciding what to do about AI in the first place. If AI is already in your building and nothing has moved on the P&L, the sibling engagement is The Efficiency Audit, which measures what you are already running before anything new is proposed.

And when the diagnosis recommends a capability we haven’t delivered before, sourcing it through the delivery network is our job. The recommendation is never limited to what we already have on the shelf.

The posture

Diagnosis before prescription.

The proposal contains only what the diagnosis supports — including when the honest answer is a checklist and one hire, not agents. Strategy work here starts at that altitude: what the firm is for, what the numbers say, and what to leave alone.

Some of that redesign happens in front of a system nobody is allowed to change. Every practice runs at least one: the practice-management platform, the records portal, the legacy ledger front end. The vendor owns the back end, replacing it is a migration nobody will approve, and so the daily workflow stays broken for years. We study that system read-only, then build the layer your people actually need on top of it: search that ranks, an intake screen that writes back through the vendor’s own interfaces, the report the vendor never shipped. The contract is untouched, the vendor’s data stays the source of truth, and taking our layer away restores exactly what you had.

Related services

Where this connects.

Where the diagnosis finds repeatable work, an agent carries it.

Learn more →

The layer the strategy stands on, measured before anything is proposed.

Learn more →

The control questions a board should settle before the build.

Learn more →

When AI is already in the building, start by measuring it.

Learn more →

Next step

One process. One accountable seat. One month, measured.

That’s the engagement shape, in writing on day one. The conversation that starts it books directly.

Start a conversation →

The Efficiency Audit

Already running AI?
Find the missing margin.

The tell is everywhere: the tool got adopted, everyone says it helps, and no line on the P&L moved. The Efficiency Audit finds the gap between what your firm spends on AI and what the numbers show for it.

Start a conversation → Train your team →

What the Efficiency Audit measures

What gets measured

Every AI tool in the building against the workflow it was bought for: what it costs, who touches it, what changed on the P&L, and where the work re-routed around it instead of through it.

Why tools stall

Powerful engines bolted onto processes shaped by human limits — the eight-hour shift, the sequential handoff, the approval that waits for Monday. The fix is a redesign of the road, not another engine. The full argument is here.

What you keep

Written findings, a ranked fix list with what each is worth, and an honest verdict on every tool: fix the process around it, renegotiate it, or drop it. Credited toward whatever engagement follows — if any does.

The Efficiency Audit assumes the tools are already bought. If nothing is running yet and the real question is where to start, the engagement you want is the Audit, under Strategy Consulting, which maps how the work moves before any tool is chosen.

When the fix list names a capability new to us, sourcing it is part of the job, through the delivery network the firm keeps for exactly that.

If the question your committee actually asks is what this exposes the firm to, and what it does to the people who have to work differently, that is the same engagement seen from the other side: AI risk and change impact.

What you keep

Findings ranked by what they’re worth.

Every tool in the building gets an honest verdict — fix the process around it, renegotiate it, or drop it — with the value of each fix attached, so Monday’s priority list writes itself.

A workshop leader smiling beside a whiteboard of diagrams

Related services

Where this connects.

When tools underperform, the base layer is the first suspect.

Learn more →

Keep the measurement running after the audit ends.

Learn more →

The controls that keep the fixed version fixed.

Learn more →

When the finding is bigger than a tool, the diagnosis takes over.

Learn more →

Next step

Name the tool you’re least sure about.

Thirty minutes, online. If the honest answer is “keep it and change nothing,” you’ll hear that too.

Start a conversation →

AI Risk & Change Impact

Two questions
before the technology.

What could this cost us if it goes wrong, and what happens to the people who have to work differently because of it. Most AI conversations skip both and start with the tool. These are the questions a managing partner, a compliance officer or a board actually has, and they are answerable before anything is built.

Start a conversation → Read the standard →

The two assessments

AI risk management

Where AI already touches your work, what it is allowed to do, and what happens when it is wrong. Autonomy assigned one task at a time by risk, a written register of actions no agent may ever take, a named person accountable for anything reaching a regulator, a court or a counterparty, and a log of every action taken. The answer is a document your committee can read, not a reassurance.

AI change impact

Who does this differently on Monday, what they lose, what they gain, and who has to agree before any of it holds. This is the discipline that gets cut first when budgets tighten, which is why so many working tools sit unused. A redesign nobody adopted is not a technology failure. It is a change nobody was paid to manage.

What you get

Findings ranked high, medium and low. Each one carries what it would take to fix, what the fix is worth, and what it costs. You choose which ones we do. An honest finding of leave this alone is a valid outcome and appears on the list like any other.

These are not new engagements. They are what the Audit and The Efficiency Audit already produce, named the way you would ask for them, with the written standard from Governance behind both. If your firm has AI running and nothing has moved on the numbers, start with the Efficiency Audit. If nothing is running yet and the question is whether it should be, start with the Audit.

Why this order

The technology is the easy half.

A model that drafts a document is a solved problem. Deciding what that document is allowed to be used for, who checks it, what happens when it is wrong, and how the people around it change their week: that is the half that decides whether any of it survives the first quarter.

Related services

Where this connects.

When AI is already running, this is where both questions get measured.

Learn more →

The written standard the risk answer is drawn from.

Learn more →

When the finding is a decision rather than a fix.

Learn more →

Assurance

Every mechanism named here, published and checkable.

Learn more →

Next step

Bring the question your committee keeps asking.

Thirty minutes, online. If the honest answer is that you have less exposure than you think, you will hear that too.

Start a conversation →

Skills Enablement

Free to install.
Not free to trust.

Your team can add AI skills to ChatGPT, Copilot, Claude or Gemini in a few clicks, and most of them are someone’s weekend project. When Snyk scanned nearly four thousand published skills in February 2026, more than a third carried a security flaw and seventy-six were outright malicious. We tell you which ones to install on the platform you already run — and, more usefully, which ones not to.

Start a conversation → See the controls →

How skills are screened and deployed

Screened, not collected

Every recommendation is read before it reaches you: the license from the license file rather than the badge, what the thing is allowed to execute, and where your data goes when it runs. The most recent screen ruled out a popular pack whose terms forbid earning consulting revenue from it, and a vendor-badged plugin that turned out to be proprietary. Neither would fail a virus scan.

On the platform you own

Broken out by what you actually run — ChatGPT, Microsoft Copilot, Claude, Gemini — because the answer differs. Each entry says what it is for, how someone non-technical turns it on, and which switch in your own admin console holds back everything you haven’t approved.

Nothing that signs or pays

Several popular integrations can send a contract for signature or move money. Ours ship read-only or not at all, because no agent of ours signs, files, or pays — the same boundary printed in every engagement, enforced in the setup rather than in a policy nobody reads.

Glass facade reflecting the building opposite

Why it recurs

A list goes stale. A standard doesn’t.

These tools can be set to update themselves, which sounds like a feature until you realize it means unreviewed changes to something shaping your firm’s written work. We keep yours held at a version somebody read; nothing updates it without a review, and we re-read the whole set every month. What you are buying is not the list. It is that someone competent looked at it this month.

Related services

Where this connects.

The written standard these choices are made against, in your firm’s own language.

Learn more →

What your team already installed, and what it is actually returning.

Learn more →

Membership

The training quarters that teach a team to use what was installed.

Learn more →

Where a repeated task deserves more than a skill and gets built properly.

Learn more →

Next step

Tell us what your team is already running.

Thirty minutes, online. Bring the platform you pay for and whatever people have installed on it, and you leave knowing what to keep, what to switch off, and what to lock down.

Start a conversation →

Data hygiene

Your stack is only as smart
as its dirtiest data.

An agent system is layers: your records at the bottom, the assembled context in the middle, the agents on top. Every layer inherits the one beneath it — which means you are only ever as ready as your weakest foundational layer.

Start a conversation → Train your team →

See it happen

One dirty field. Watch it climb.

~~Agents · act~~ Acts within its gates, correctly.

~~Context · assemble~~ Assembles the right file, every time.

~~Data · the base~~ Reconciled, deduplicated, dated, owned.

Everything above inherits the base. That is the whole argument for starting there.

Before any of it runs

The part nobody wants to start with.

An agent inherits whatever the records already are. So the first work in any build is the records themselves: where they live, who owns each field, and which of them a system may read at all. Nothing runs on data the diagnosis has not seen.

The deliverable

The Data Foundation Assessment.

A stage of the Audit, run before any agent is proposed: data hygiene measured, not assumed.

What it checks

Duplicates, unreconciled balances, undated and unowned records, fields that disagree between systems, and who can touch what. The base layer, examined the way an auditor examines it.

What you keep

A written findings list ranked by what each fix opens up — because a deduplicated client file is not housekeeping, it is what makes every layer above it trustworthy.

Why it comes first

No agent is proposed onto a base the assessment hasn’t cleared. Fixing the base costs less than fixing everything it corrupts — and it is the least glamorous work in the stack, which is why it gets skipped.

Related services

Where this connects.

What a clean base makes possible — and a dirty one sabotages.

Learn more →

Sourced answers need a base worth citing.

Learn more →

Where underperforming tools get measured against what they cost.

Learn more →

The five stages in full →

The diagnosis that decides which layer to fix first.

Learn more →

Next step

Ask us what your base layer would score.

Thirty minutes, online. Bring one report you don’t fully trust and we’ll trace the doubt to its layer.

Start a conversation →

Try a workflow

Watch the work move.
Then see what moved it.

A worked example you can drive yourself. One item arrives, four steps run on their own, and a named person takes the decision a person has to take. Every name, record and figure below is invented for this page: nothing here is your data, nothing is stored, and nothing is signed, filed, paid or sent.

Map your own process → Talk it through →

A stone arcade of identical bays running toward a lit doorway

Workflow demo Simulated demo Ready

The attorneyNamed accountable seat

What stage one watches

What stage five may do

Illustrative workflow. No company data required. Every record here is sample text.

The run

Nothing is signed, filed, paid or sent from this demo. Nothing you do here is recorded.

Read the method behind it →

What just happened

The same five stages carried it.

Every agent this firm builds runs one engine: Signal, Enrich, Score, Compose, Act. The panel below reads the run you just watched back against those five stages, and it moves as the run moves.

A figure walking away down a colonnade of square stone piers

Where a person sits

Four stages run on their own. The fifth is a person, by design.

Autonomy is assigned per task by risk, so it is never one switch for a whole workflow. The stage that can change something carries the tightest tier, and the seat that owns it is named in writing on day one.

Autonomy tiers, per task

Each task carries an explicit tier, from fully checked to fully autonomous, assigned by risk. In the run above, pulling records is not the same risk as acting on them, so the two do not carry the same tier. A Guardian agent watches the others and every action is logged, whatever the tier.

Read the standard →

A named seat, from day one

One accountable seat is named in writing before any work begins. In the demo it is the attorney or the controller. In your firm it is a person with a name, and the stage-five decision is theirs. It is part of the workflow rather than a setting somebody can switch off later.

See how it is written →

Below the threshold, a question

Choose an uncertain item in the demo and the run stops short of acting. Below the confidence threshold the answer is a question to a human, never a guess, and the question arrives with the records that made it uncertain already attached.

Next step

Now do it with your own process.

The Workflow Builder turns the process that hurts into a map in about ten minutes. No account, and nothing is stored until you choose to send it. It becomes the pre-call assessment, so the free conversation starts from your work rather than from a blank page.

Open the Workflow Builder →

Assurance

Careful by promise,
and by architecture.

Autonomous can’t mean unaccountable. So every workflow here has autonomy assigned by risk, a named person who can stop it, a Guardian checking the builder’s work, a register of actions no agent may take, and a log of everything done. Most firms answer trust questions with adjectives; those five are what we answer with instead.

A white ribbed facade curving against an overcast sky, its structure fully visible

Ethical AI · Human-centered AI · Augmented intelligence

Three phrases, defined as things you can check.

All three are usually offered as adjectives, which is why they have stopped meaning much. We use them because here each one resolves to a mechanism. Ethical AI is a written register of thirty-seven classes of action our agents may never take, which grows and never shrinks; contractual no-training on your data; a licensed CPA inspecting every financial deliverable before it reaches you, as advisory work, with no financial-statement audit, review or other attest report issued; and a rule that whoever checks the work is never whoever built it. Human-centered AI is autonomy granted one task at a time by risk, a named person accountable for anything that reaches a regulator, a court or a counterparty, and a diagnosis that is allowed to end in leave this alone. The industry’s term for that shape is human in the loop, or HITL, and it is a fair description of what we build: a person sits inside the run, not after it.

Augmented intelligence is the third, and it is the one we would put first. Artificial intelligence is the technology: a model that drafts, extracts, classifies and computes. We use it, and saying otherwise would be dishonest. Augmented intelligence is a statement about where the judgment sits. The machine carries the retrieval, the assembly and the arithmetic; the decision that someone answers for stays with a person, by design rather than by habit. The difference between the two is not which software is running. It is whether anyone could tell you, afterward, who decided.

On overuse

Half the value of a diagnosis is the list of things to leave alone. The proposal contains only what the diagnosis supports, including when the honest answer is that you need a checklist and one hire rather than agents. We have written that proposal, and we will write it again.

On your people

What a redesign takes off a week is retrieval and re-keying. Judgment stays, and so do the people holding it. Capacity is the point, not subtraction. The employment evidence we rely on, and its sources, are on the perspectives page.

On slop

Nothing here is allowed to guess. Every answer names its sources or says plainly that it does not know, and where the work is arithmetic a second independent computation checks the first. We publish nothing we cannot reproduce.

Ownership

Everything runs on your side.

Everything we build for you runs on your own AI subscription, in your own Drive, and, where structured memory earns its keep, in your own database instance, under your keys. No server of ours holds your client data. That is the confidentiality answer, the lock-in answer, and the exit answer at once: when an engagement ends there is nothing to migrate and nothing to ransom. (This website’s own membership accounts are the one thing we do hold. The privacy policy lists exactly what and where.)

Your data, isolated

Per-client credentials, storage, and memory. Agents read only sources you approve. We do not train on your data and we do not submit it for training, contractually. Every agent action is logged.

Compliance built in

A HIPAA Business Associate Agreement per engagement, where patient data is in scope, signed before any of it moves. Confidentiality-first architecture for legal clients. Records-retention and FOIA awareness for public-sector work.

Licensed humans check the work

No financial deliverable leaves until a licensed CPA has inspected it; this is advisory work, and no financial-statement audit, review or other attest report is issued. The agents assemble the work and show their sources line by line. Where the work is arithmetic, a second independent computation checks the first, and a disagreement is a question for the CPA rather than something the system settles on its own. The CPA who inspects is never the builder. Tax positions and valuation opinions stay with the licensed professional, never the agent; an attest report is for an independent licensed firm engaged for that purpose, never an agent and never us. Your engagement letter states the arrangement in writing.

Sourced answers or a refusal

Our dashboards and research agents cite what they know and say so when they don't. No guesses dressed as answers, in your business or in ours.

Security architecture

Security architecture.

What we offer a security team is an architecture with far less to certify.

No system of ours holds your client data. What we build for you runs on your own AI subscription, in your own Drive, under your keys, so the usual question, how well does the vendor secure our data, mostly does not arise. What we do hold is this website: your membership account, and anything you send us through a form or the Help Desk. The privacy policy lists exactly what and where.
Per-client isolation. Credentials, storage and memory are separate for every client. Agents read only the sources you approve.
No training on your data, contractually. We do not train on your data and we do not submit it for training. That is a term in the agreement, not a checkbox in a settings panel. Any third-party tool in the build is named to you in writing, with its own terms, before it is used.
Every agent action is logged. The log is part of what you keep, and it survives the end of the engagement along with everything else.
A data processing agreement, whenever your counsel wants one. Say so and we will put one in front of you. If they would rather work from your paper, send it and we will review it.
Bespoke access, scoped case by case. Where an engagement requires us to hold access as part of a bespoke build, the security arrangement is scoped case by case under Sid Darden’s advisory before the build starts, and written into that engagement’s letter.

You can check part of this from where you are sitting. Open your browser’s network panel on any page of this site: no analytics, no advertising pixels, no third-party trackers, no session recording. What is there belongs to Cloudflare, our host, and the privacy policy names all of it, alongside every other processor we use and why.

For your reviewer

Every answer your committee asks for, in one place.

Four questions decide most vendor reviews. Here are ours, each traced to where it is stated in full, and each dated. Print this page and the whole set comes with you.

What we hold, and who else touches it

The short answer is that the work runs on your side and this website is the only thing we hold.

What to confirm with us

Everything on the left is published. These are not, because no firm answer exists yet — ask and you get a straight one.

If you think something has gone wrong with data, write to [email protected]. A person reads it. We publish no response target and no breach-notification window, because we have committed to neither; what we commit to for your engagement is in your engagement letter.

Reviewed 27 September 2026. Every statement above is published elsewhere on this site and linked to its source; if one of those pages changes, this one changes with it.

On film

Your firm’s immune system, in 55 seconds.

How an agent system stays safe, told the way biology tells it: a watcher, a set of specialists, and a memory that only grows.

Makes Eleven · Film 02 The Immune System — assurance, animated 2026 A 55-second safety story

The immune
system.

An agent system stays safe the way a body does.
Something separate watches everything the rest of it does.

THE FAILURE MODE

The plausible
guess.

Confident, fluent, unsourced. Every AI system eventually produces one. Ours is built to expect it.

×No source

×No doubt

✓Fluent

FIRST RESPONDER · ON EVERY ENGAGEMENT

The Guardian.

A separate agent that watches every action the other agents take, and flags anything out of pattern before it leaves the building. It monitors, and it never builds.

Then the specialists arrive.

The Examiner~~Never the builder~~

Checks every system against test sets with known answers, and never checks their own work.

The Second Sum~~For recomputable output~~

Where output is arithmetic, an independent computation re-derives it. The two answers must agree.

The Licensed Inspector~~For financial work~~

A licensed CPA inspects every financial deliverable before release, disclosed in every engagement letter (advisory; no attest report).

The refusal register only grows.

Classes of action our agents are built to refuse. The list ratchets up. It never shrinks.

And it never forgets.

~~LOGS~~

~~REGISTER~~

~~BASELINE~~

~~BOUNDARIES~~

When the same failure tries again, it is refused automatically, and the refusal is logged.

Everything you just saw
is written down.

The full standard is on this page, below.

00:00 / 00:55

Read the film as text

An agent system stays safe the way a body does. Something separate watches everything the rest of it does. The failure mode is the plausible guess: confident, fluent, unsourced. Every AI system eventually produces one; ours is built to expect it. First responder: the Guardian, a separate agent shipped on every engagement that watches every action the other agents take and flags anything out of pattern before it leaves the building. Then the specialists arrive: the Examiner, who is never the builder and checks every system against test sets with known answers; the Second Sum, an independent computation that re-derives every recomputable number; and the Licensed Inspector, the CPAs who inspect every financial deliverable before it leaves, disclosed in every engagement letter, as advisory work, with no financial-statement audit, review or other attest report issued. The refusal register only grows: thirty-seven classes of refused action today, and the list ratchets up and never shrinks. And it never forgets: logs, register, baseline, boundaries. When the same failure tries again, it is refused automatically, and the refusal is logged. Everything you just saw is written down. The full standard is on this page, below.

The written standard

Every agent ships against a reliability standard you can read.

An engineering document that governs every build.

Autonomy tiers, per task. Each task carries an explicit tier, from fully checked to fully autonomous, assigned by risk.
A prohibited-actions register. Thirty-seven classes of action our agents are built to refuse, from signing filings to touching payments. It grows; it never shrinks.
The examiner is never the builder. Every system is checked by someone who didn't build it, against test sets with known answers.
Drift is measured, not assumed. Each examination records which model actually served, so when the model underneath changes, the next examination shows whether the agent's behavior moved.
Measured against day one. Your baseline is recorded at the start. At day 30 you receive a one-page before-and-after against it, with every number sourced or absent and estimates labeled, and the delta monthly after that.
Recomputable output. Where the work is arithmetic, a second, independent computation checks the first.

Governance

Boundaries you get in writing.

Filings always carry a human signature. Tax returns, government filings and legal documents are assembled by agents and signed by licensed people.
Money moves only on a human’s button. Payment initiation, transfers and payroll runs keep a person on approval, every time.
Judgment stays with licensed professionals. Tax positions and valuation opinions are theirs, never an agent’s; an attest report, meaning an audit or a review of financial statements, is for an independent licensed firm engaged for that purpose, never an agent and never us.
External release is a human decision. Nothing unreviewed leaves for a regulator, court or counterparty.
Below the confidence threshold, the answer is a question. A sourced answer or an honest ask — never a plausible-sounding guess.
Every number we publish carries its record type, and we say what has not yet been published. No claims we can’t reproduce, in your business or in ours.
If the baseline has not moved by day 30, we say so and you can stop. The day-30 report states the delta plainly, and you may end the engagement at the end of that month. The term goes into your engagement letter before any work begins.

Next step

Put this architecture in front of your reviewer.

Bring your security team or outside counsel to the first call. Everything on this page is theirs to test, and if you go on to a diagnostic, it is a deliverable you keep either way.

Start a conversation →

Guides

What we would hand you
before you hire anyone.

Three documents, free. Read any of them right here, with no address at all. Leave an address and we will send you a copy; no sales sequence follows, and a person answers if you reply. Everything in them is what we actually run, which is why they are specific enough to be useful against a firm that is not us.

The guides

For the managing partner

The managing partner’s guide to AI in a law firm

Seventeen pages for the person whose name is on the answer when the committee, the client or the carrier asks how AI is controlled here. ABA Formal Opinion 512 duty by duty, paired with the artifact each duty actually demands; the three failure modes that produce discipline and what prevents each; and a checklist your committee can run against any vendor, this one included.

Read it now →

For the committee

The prohibited-actions register

All thirty-seven classes of action our agents are built to refuse, with the basis for each and the scope it binds. It is built from the file our own tests check the published count against. The hash of that file is written into the PDF and recorded beside it, and our tests fail if the PDF, the record and the register ever disagree.

Read it now →

For the first conversation

The A.G.E.N.T. method, a short read

Audit, Gauge, Engineer, Navigate, Track: the five stages, the artifact each one leaves with the client, the four questions that come before any of it, and the roughly eight-week shape a first engagement takes.

Read it now →

A long corridor of warm timber fins receding toward a lit doorway

What a guide is

Written down before you need it.

Each one is the part of a diagnosis that does not depend on your firm: what to ask, what to check, and where the honest answer is to leave a process alone. They are addressed to the person who has to defend the decision inside the firm.

In progress

The State of AI in Regulated Work

Our annual report, published by Makes Eleven Consulting, assembled from the monthly ecosystem reviews we already run. It has not been published yet and there is no earlier edition to read. The release is announced in the Letter, so this is the Letter: subscribe once and the report reaches you the same way everything else does.

Nothing here is legal, tax or accounting advice, and no document on this page argues that you should hire us. If they are useful and you never call, that is a fine outcome. The Workflow Mapper and one board in every vertical of the Workflow Atlas are open the same way.

Perspectives

The myths, examined.
The evidence, cited.

The concerns we hear most from operators and public officials, answered with primary sources. AI is already trusted where the stakes are highest: medicine, law, government, finance.

“AI will replace my people.”

What the evidence says

The Yale Budget Lab has tracked the US labor market monthly since ChatGPT's release and finds no discernible disruption to overall employment. RAND reaches the same conclusion, with more businesses reporting jobs added from AI than cut. The pattern is augmentation: AI absorbs volume; people keep judgment, relationships, and accountability.
Sources: The Budget Lab at Yale, updated monthly · RAND, 2025

How we handle it

Our engagements are designed around your team, not instead of it. People are the other half of eleven.

“AI makes things up too much to trust with real work.”

What the evidence says

Unsupervised AI does invent, which is why serious deployments never run unsupervised. The FDA has authorized over 1,400 AI-enabled medical devices under formal validation. Court systems that permit AI-assisted work now require human verification of its output. The fix is structure, not hope.
Sources: FDA AI-enabled device list, updated periodically · California Rule of Court 10.430, 2025 · South Carolina judiciary interim AI policy, 2025

How we handle it

The same structure: agents do volume, licensed professionals inspect financial deliverables, every answer is sourced or refused, and an agent sends nothing without an explicit instruction to send that particular thing — never anything signed, filed, paid, or bound for a regulator, court or counterparty unreviewed.

“Our data will end up training some model, or leaking.”

What the evidence says

That risk is real with consumer tools and absent with properly configured enterprise deployments: contractual no-training terms, tenant isolation, encryption, and audit logs are standard. California's court system draws exactly this line: confidential data prohibited in public AI systems, permitted in managed internal ones.
Source: California Rule of Court 10.430, effective September 2025

How we handle it

Per-client isolation, allowlisted sources only, a written security posture you can hand to your counsel, and this on training: we do not train on your data and we do not submit it for training. Where a third-party tool is part of what we build, we name it and its terms in writing before it is used, so you are reading that vendor’s promise rather than ours.

“AI is too legally risky for government.”

What the evidence says

Government frameworks don't prohibit AI; they govern it. NIST's AI Risk Management Framework, federal agency guidance, South Carolina's state AI strategy, and the SC judiciary's generative-AI policy all chart responsible adoption. Cities are already deploying: permitting prescreens, multilingual constituent services, records automation.
Sources: NIST AI Risk Management Framework · SC judiciary interim AI policy, 2025

How we handle it

Framework-aligned deployments with human approval, audit trails, records-retention awareness, and bias testing as standard, pilot-sized to fit procurement realities.

“It’s a fad.”

What the evidence says

FDA authorizations of AI medical devices grew from a handful a year to more than three hundred in 2025 alone. Nearly a third of practicing lawyers now use generative AI. Regulated professions are the slowest adopters by design, and they have adopted.
Sources: FDA AI-enabled device list · ABA Legal Technology Survey Report, 2024

How we handle it

We don't sell momentum; we sell diagnosed fit. If AI isn't the answer to your problem, that's the report you'll get.

“We’re too small, or not technical enough, for this.”

What we have seen

This one is a judgment from our own work rather than a finding, so it cites no study: fit to the work has mattered more than in-house technical depth. The published engagement shape runs roughly eight weeks.

How we handle it

You don't need in-house engineers. The diagnosis tells us what fits your size and budget; a written ceiling means no surprises; and our enablement work leaves your team able to run what we build.

“AI in public systems invites bias and civil-rights problems.”

What the evidence says

The risk is documented: in 2025 the Massachusetts Attorney General reached a $2.5 million settlement with student-loan lender Earnest over underwriting models alleged to disadvantage Black, Hispanic, and non-citizen applicants. The same case defines the fix. Test for disparate impact, keep humans on decisions, document everything.
Source: Massachusetts Attorney General, July 2025

How we handle it

Disparate-impact testing, human approval on consequential decisions, and full audit trails are standard in our public-sector and lending-adjacent work, not add-ons.

“AI consultants are just reselling tool tutorials.”

What the evidence says

Many are, which is why most enterprise AI pilots show no measurable P&L impact. Deployments fail when nobody maps the workflow first.
Source: MIT Project NANDA, The GenAI Divide: State of AI in Business 2025

How we handle it

We’re a boutique advisory firm that uses AI, not an AI vendor with a pitch. The diagnostic is a real deliverable you keep either way, the proposal contains only what the diagnosis supports, and every claim we make in public carries its evidence or says what is still missing.

Why not just use ChatGPT ourselves?

Often you should, and we will say so

Often you should, and a diagnosis that ends in buying the subscriptions, writing a one-page acceptable-use policy and training the team is a real outcome here rather than a failure. What a subscription cannot do is map the workflow, decide which steps a machine should carry and which a person must keep, or name who is accountable when the output is wrong.

For drafting, summarizing and thinking out loud, a good assistant and an hour of training is often the whole answer, and no engagement is needed to get it. Sometimes the honest recommendation is a checklist and one hire rather than agents at all.

What a subscription cannot do

It cannot map the workflow, decide which steps a machine should carry and which a person must keep, or name who is accountable when the output is wrong. It cannot inspect a financial deliverable with a licensed CPA (advisory work, with no financial-statement audit, review or other attest report issued), keep a register of actions no agent may take, or measure the result against a day-one baseline. The tool is the cheap part; the work is deciding what to point it at and who answers for the answer.

What happens when the AI model changes?

It will, and that is assumed

Models change constantly, which is why this firm watches the field every month rather than picking one and hoping. What we build is engineered to your workflow rather than to a particular model, and repairing a desk when a connected system or model changes beneath it is what a maintained engagement is for.

The map, the decision points, the accountable seat and the register outlive any provider’s release notes.

What we do about it

The Guardian re-checks the running process rather than trusting that it still behaves, the same recomputation and inspection steps run whatever is underneath, and repairing a desk when a connected system or model changes beneath it is what a maintained engagement is for. If a change makes something we built worse, you hear it from us first.

Written by Leon Fields II. Reviewed 27 September 2026. Every position that makes a factual claim cites a primary source; where the evidence changes, this page changes with it.

An oval skylight opening onto cloud, seen from inside a curved concrete shell

Why these are written down

A claim we cannot show you is a claim we drop.

Every piece here is argued from something you can go and check. Where the evidence runs out the piece says so, rather than filling the gap with a plausible sentence. That is the rule the agents work under too.

Next step

Test the thinking against your own workflow.

The positions on this page were formed in operating work, not in theory. Map one workflow in ten minutes, free and unstored, and see what the same lens says about your firm.

Open the Workflow Mapper →

Our team

An operator who kept finding
the same gaps.

Meet the team where one and one makes eleven.

Our world-class people

Founder-led, with two strategic advisors closest to the core, one on cybersecurity and one on automations, and a network of licensed CPAs who inspect every financial deliverable before it reaches you, disclosed in every engagement letter, as advisory work, with no financial-statement audit, review or other attest report issued. The rest of the bench is assembled for the engagement, never carried as overhead. Open a card for the full account.

Leon Fields II Founder & Managing Partner

An economist by training and an operator by habit, who kept finding the same gaps in every process he touched.

Wylie “Sid” Darden Strategic Advisor, Cybersecurity

A former U.S. Air Force cyber operations officer who holds the CISSP.

Joshua House Strategic Advisor, Automations

How a manual process becomes a governed system. Six years at EY leading technology delivery in its government and public sector practice.

Three lanes

Some problems are operational. Some are technical. Some move only when the right people are in the room.

Operations

Workflow mapping, process redesign, the diagnosis itself. The lane everything else depends on.

Automation

The agents, dashboards, and systems, built to the written standard and measured against day one.

Relationships

Policy rooms, procurement processes, the introductions that move initiatives. Compounding requires every gear turning.

One habit runs under all three lanes: a monthly review of what the AI ecosystem now makes possible, so the advice never ages quietly.

In the room

Leon has spent the better part of a decade building relationships in Washington.

At a conference, in front of a sponsor backdrop

Range, not a fence

The verticals are evidence, not a boundary.

The industries that carry named playbooks carry them because the work happened there first. They are proof of range, not a fence: no AI ask is out of scope, and the delivery network of advisors and specialists is how the firm answers one it has not met before.

The firm behind the founder

We’re just some humans who think computers should do more work.

One accountable firm, and a network built to deliver anything AI. A boutique advisory shop by design: a small accountable core that stays close to the work, and a bench assembled per engagement rather than carried as overhead you end up paying for.

The delivery network

Two strategic advisors sit closest to the core: Sid Darden on cybersecurity and Josh House on automations, both named above with their credentials. Around them, specialist advisors, consultants and ad-hoc expert hires are assembled for the engagement that needs them, and a network of licensed CPAs inspects every financial deliverable before it reaches you, as advisory work, with no financial-statement audit, review or other attest report issued.

No AI ask out of scope

Voice AI and AI calling, computer vision, sales and marketing automation, custom software, managed AI subscriptions for a whole team — the tools we run in-house carry the core, and the network carries the rest. Finding the right capability is what you hire us for.

Support that keeps running

Every maintained engagement and membership carries 24/7 support from a retainer team across time zones. The Help Desk is where a report becomes a ticket, and every fix lands in a shared repository.

AI has leveled the playing field in a generational way. We exist so small and middle-market firms can strategize, compete and scale in ways they never could — capturing market share and influence like never before.

How the work is governed

Autonomous can’t mean unaccountable.

Every agent we build runs inside a written standard: autonomy assigned per task by risk, every action logged, a Guardian agent on every engagement watching the others, and a named person accountable for anything that reaches a regulator, court or counterparty. Below the confidence threshold, the answer is a question to a human.

What governs the work is written into the contract, not implied. A network of licensed CPAs inspects every financial deliverable before it reaches you, disclosed in every engagement letter, as advisory work with no financial-statement audit, review or other attest report issued. The support behind an engagement is built the same way: a report becomes a ticket, a fix lands in the repository, and a person answers for it.

Written into every engagement

A person signs, files, and pays.
Tax positions, valuation opinions, and attest reports stay with the licensed professional who answers for them.
Every recommendation rests on the diagnosis.
Every claim we make can be reproduced.
Every proposal names its price, and the ceiling holds.

The full standard, and the register of thirty-seven classes that only grows, is on the assurance page.

Next step

Meet the operator behind the pages.

A thirty-minute conversation about the task that eats the most hours at your firm. No deck, no pitch, and an honest answer even when the answer is a checklist and one hire.

Book a conversation →

Solutions · by industry

Built for the way your
industry actually works.

The same five-stage engine, tuned to the documents, deadlines, and regulators of your world. Every engagement starts with the diagnosis. These are the worlds where our playbooks run deepest.

A butcher in a blue apron smiling at his phone behind the counter

Each card names the seat we usually sit in when we work in that world. If your industry is here and your seat is not, or your seat is here and your industry is not, the diagnosis works exactly the same way.

The industries we work in

Real Estate & Affordable Housing

Run the operational spine of a portfolio on time, every month.

Usually our seatThe person answerable when the monthly close on a portfolio is late, and the team working the queue underneath them.

Rent rolls and ledgers reconciled monthly, variances named
Delinquency queues worked with escalation built in
Development draw packages assembled from source documents
Lease-up pipelines and unit turns tracked to the day
Vendor invoices checked against the contracts that govern them
Computer-vision reads of inspection and unit-turn photo sets
Resident communications drafted for a person to send

See the results →

A person writing notes at an office desk

LIHTC Compliance

Keep the credit by keeping the file — before the auditor asks.

Usually our seatThe compliance officer whose name is on the tenant file when the agency asks to see it.

Income certifications assembled and second-checked
Recertification calendars worked to the anniversary, not to the reminder
Tenant-file completeness sweeps ahead of agency audits
Placed-in-service and compliance-period deadlines watched
Agency reporting packages drafted for review
Year-15 files built as the years happen

See the results →

A meeting board covered in planning notes

General Compliance, Any Industry

Track every obligation from the source document to the evidence.

Usually our seatWhoever holds the obligation register and has to prove, later, that it was kept.

Obligation registers extracted from regulations and contracts
Evidence collected continuously, all year
License and filing renewals on a calendar with margins
Policy-to-practice gap reviews with findings ranked
Regulator correspondence drafted for sign-off
Incident logs kept audit-ready as they happen

See the results →

A dental treatment room with the chair ready

Healthcare & Adjacent Practices

Recover the front-office hours that eligibility and claims consume.

Usually our seatThe practice administrator running the front office patients actually experience.

Eligibility verified before every visit
Claims assembled, scrubbed and chased to payment
Denials worked by root cause
Recall and reactivation outreach that actually goes out
Voice AI on the front-desk phones — reminders, scheduling, overflow
Intake digitized before the patient arrives
Credentialing and license files kept current

See the results →

Legal

Protect the hours you bill and the deadlines you cannot miss.

Usually our seatThe managing partner, or the partner who chairs the AI committee and answers for what it decides.

Billable time captured from the file as the work happens, for the attorney to confirm
Intake done in days: conflicts, engagement paper, matter setup
Deadline calendars read from the documents themselves, every date confirmed by a person
Work routed between attorney and assistant at the right rate
Demand and discovery drafts assembled for attorney review
Conflicts checks run before the first call back

Lobbying

Know what moved in the record before your client asks.

Usually our seatThe principal who signs the client report and the disclosure that follows it.

Legislative and regulatory monitoring fused with the client book
Hearing and comment-deadline calendars kept ahead of the docket
Position summaries drafted from the actual bill text
Stakeholder and committee maps kept current
Client reports assembled weekly, sourced line by line
Disclosure filings prepared for a human signature

A restaurant point-of-sale terminal at the bar

Restaurants & Hospitality

Defend the margin line by line, week by week.

Usually our seatThe owner-operator, or the general manager who owns the prime-cost line every week.

Supplier invoices checked against agreed prices, every line
Inventory variance flagged weekly with the receipts attached
Prime-cost reporting from the POS you already run
Licenses, permits and inspections renewed with time to spare
Labor schedules checked against forecasted covers
AI phone answering for reservations and orders, tuned to your menu
Delivery-platform and chargeback reconciliation

See the results →

Government

Move the queues residents feel — pilot-sized and procurement-aware.

Usually our seatThe department head answerable for a queue residents can see from outside.

Permitting prescreens that cut review queues from months to days
Program intake and eligibility with a human on every consequential decision
FOIA and records requests processed on the statutory clock
Grants managed and reported without the scramble
311 and constituent-service routing with context attached
Legacy paper intake digitized form by form

Government Contractors

Keep the bid, the submission and the signature with a person, and the record with the file.

Usually our seatThe capture lead or the contracts manager who answers for the compliance matrix, the clause list and the flow-downs.

Every pursuit a record with a stage and a gate a person answers
Compliance matrices built from the solicitation’s own text, every shall traced to a row
FAR clauses tracked with who read what, the prescribing text governing
Prime clauses paired against each subcontract, gaps flagged, language drafted for counsel
Past-performance volumes assembled from sourced records, a missing field left blank

A conveyor sortation floor inside a plant, cartons moving along the lines beneath steel gantries

Manufacturing & Supply Chain

See the shortage coming while there is still time to do something about it.

Usually our seatThe operations or supply-chain manager who answers for on-time delivery, and for the cost of a line that stopped.

Purchase orders and acknowledgments matched against what actually arrived
Supplier lead times and late deliveries tracked into a pattern you can quote
Inbound shortages flagged against the build schedule before the floor feels them
Non-conformance and quality records assembled from the shop-floor file
Outbound freight tracked to proof of delivery, detention documented as it happens
Certificates of analysis, safety data sheets and supplier documents kept current
Customs and country-of-origin paperwork prepared for a person to sign

Map the process that hurts →

A carpenter shaping timber at a workbench

Owner-Led Firms Preparing To Sell

Arrive at diligence with numbers a buyer can trust.

Usually our seatThe owner, ideally well before the broker is engaged.

A recast P&L with every add-back evidenced
A data room assembled early and kept current
Working-capital and revenue-quality analyses run in advance
Customer-concentration exposure mapped honestly
Sales and CRM automation that documents revenue quality as it runs
Diligence request lists answered from the file, fast
Engaged by the owner, welcomed by their broker

See the results →

Political Campaigns

Run the calendar nobody can move, inside the rules that change by state.

Usually our seatThe campaign manager who owns every send, and the treasurer who owns every dollar and every filing.

Research from public sources, every figure sourced
Outreach drafts a staffer reviews and sends
Strategy and message sessions with your consultants
Contribution records reconciled for the treasurer
Messages checked against published disclosure rules
Public sentiment digests, labeled as not a poll