Rehearse for $0

The manifesto · October 2026

Great ideas deserve great teams.

Intelligence is not enough. Organization makes it useful. Why yaaf exists, what it believes, and where it is going, in ten chapters and seven principles.

§ 01The conviction

Great ideas
deserve great teams.

The world is full of people with ideas they cannot pursue.

Not because their ideas aren’t good enough. But because bringing an idea to life takes more than ambition.

It takes knowledge. Experience. Specialists. Leadership. Coordination. And often, money.

Building a capable team has always been one of the greatest barriers between having an idea and making it real.

We believe that barrier is about to change.

From possibility to progress

The ideaSomeone sees a possibility
The teamSpecialists work together
The resultA meaningful outcome

In yaaf todayyaaf is Apache-2.0 and runs on your own machine. There is no account and no hosted tier; your agents pay your model provider directly. Open source

§ 02The belief

Intelligence is not enough.
Organization makes it useful.

Artificial intelligence has given us remarkable capabilities.

First, models could answer questions. Then, agents could perform tasks.

But giving someone an intelligent agent is not the same as giving them a capable team.

Real work rarely happens in isolation. It requires different skills, competing perspectives, shared knowledge, clear responsibilities, and someone accountable for the outcome.

A software engineer builds. A marketer understands customers. An analyst examines evidence. A critic challenges assumptions. A leader brings everything together.

No single role needs to know everything. The organization makes their combined knowledge useful.

From individual intelligence to organized work

ModelsAnswer questions
AgentsComplete tasks
TeamsPursue outcomes

We believe the next great advancement in AI won’t simply be a smarter model. It will be how we organize intelligence to work together.

In yaaf todayThe role library holds 38 job descriptions, each with its own instructions, model class and tools. 9 of them exist to disagree. The team

§ 03The method

Build a company.
Hire a CEO. Let the team work.

Yaaf brings the structure of an organization to artificial intelligence.

You start with a goal. It might be finding the right printer for your office, building a software product, operating a marketing department, or creating an entirely new business.

You decide what success means. You are the chairman.

You can assemble your own team, define its roles, establish its rules, and direct its work. Or you can delegate those responsibilities to an AI CEO.

Your CEO coordinates specialists, breaks objectives into manageable work, evaluates results, and determines what should happen next.

The team works toward the goal.

You don’t need to know how to hire every specialist. You need to know what you’re trying to accomplish.

In yaaf todayDescribe the project in a paragraph and the Founder drafts the goal, the teams, the budget and the date, then argues with you until nothing is assumed. How it works

§ 04The authority

Authority belongs
to the chairman.

Autonomy should never mean surrendering control.

Some people want to approve every important decision. Others want to define a goal, establish a budget, and let their CEO operate independently.

Both approaches belong in yaaf.

The chairman decides how much authority to delegate.

The CEO operates within those boundaries, adjusting plans, assigning responsibilities, and pursuing alternatives as circumstances change.

Delegation changes who makes decisions. It does not remove accountability.

The organization works for you, not the other way around.

You choose the degree of delegation

How much authority the CEO holds

Supervised · right for your first project

7 stops

In a real rehearsal of a red-team project, the company stops and waits for you 7 times. Each one is a morning it sits idle until you open the app.

Set rule by rule

4 stops

In a real rehearsal of a red-team project, the company stops and waits for you 4 times. Each one is a morning it sits idle until you open the app.

Set rule by rule

3 stops

In a real rehearsal of a red-team project, the company stops and waits for you 3 times. Each one is a morning it sits idle until you open the app.

Set rule by rule

3 stops

In a real rehearsal of a red-team project, the company stops and waits for you 3 times. Each one is a morning it sits idle until you open the app.

Set rule by rule

0 stops

In a real rehearsal of a red-team project, the company never had to wait for you. You read what the CEO decided afterwards, and can reverse any of it.

Autonomous · where a new project starts

0 stops

In a real rehearsal of a red-team project, the company never had to wait for you. You read what the CEO decided afterwards, and can reverse any of it.

  1. plan_authority

    YouYou accept each requirement the CEO proposes.

    CEOThe CEO accepts its own plan. The plan gate still runs the same arithmetic.

  2. commitment_authority

    YouYou sign the delivery date, and only you move it.

    CEOThe CEO adopts the platform's projected date. It never writes one.

  3. scope_authority

    YouCutting, reopening and restaging work is yours.

    CEOThe CEO re-scopes on the record. Every change is reversible.

  4. verdict_policy

    YouResults no script can check wait for your judgment.

    CEOThe CEO judges them, and every self-judged close is disclosed to you.

  5. capability_authority

    YouYou grant every credential.

    CEOThe CEO re-scopes credentials it already has across its team.

Never moves, at any setting

  • The budget wall. At the limit you set, the company pauses itself.
  • Verification. The platform runs the tests, not the agent that wants to pass them.
  • Spending, legal and credential approvals still reach you.
  • The record. Every decision, yours or the CEO's, is logged with its reason.
  • Your buttons. Pause, stop, veto, amend and reopen work at every setting.

Delegation changes who makes decisions. It does not remove accountability.

In yaaf todayThe dial above is the real one: five switches in a constitution no agent can edit. Budgets, verification and the record bind the same at every setting. How it works

§ 05The standard

Trust is earned
through evidence.

A confident answer is not necessarily a correct answer.

A completed task is not necessarily a successful task.

And an AI agent agreeing with another AI agent is not proof that either is right.

We believe intelligent organizations need more than intelligence. They need checks and balances.

Work should be evaluated against objectives. Important conclusions should have supporting evidence. Decisions should be open to challenge.

Independent reviewers should be able to disagree. Assumptions should be tested. Information should be traceable to its source.

When necessary, the CEO should seek external validation rather than simply trusting the team’s conclusions.

Yaaf is designed around the principle that decisions should be explainable, evidence should be inspectable, and authority should be enforceable.

The evidence trail

SourceWhere did it come from?
WorkHow was it used?
ReviewWas it challenged?
DecisionWhat can we defend?

Trust should never require blind faith.

In yaaf todayNothing closes without observed evidence. External evidence is fetched, hashed and snapshotted by the server, and a random sample of finished work is dealt to you to check. Features

§ 06The challenge

A good CEO knows
when to disagree.

A leader’s job is not to say yes.

Ask a CEO to build a business generating $100 million in twelve months with an unrealistic budget, and a good CEO should challenge the request.

It should question the assumptions, explain the risks, and help establish an achievable objective.

The same principle applies throughout the organization.

A specialist should be able to challenge a plan. A critic should identify weaknesses. A CEO should question its own conclusions.

We don’t want AI teams that simply agree with one another.

We want teams capable of finding better answers.

What honest leadership looks like

The askAn ambitious goal
The challengeTest the assumptions
The planA credible next step

The CEO should question its own conclusions, too.

In yaaf todayBefore anything exists, the Founder grades your goal against a written ten-point standard and argues with you. Critic seats can veto the work they check, and the CEO cannot remove the last one.

§ 07The process

Learning is part
of the work.

Not every project succeeds.

A product might fail to find customers. A marketing campaign might miss its target. A promising strategy might turn out to be wrong.

That doesn’t mean the work was worthless.

A campaign targeting 100 sales that delivers 50 may reveal something important about the product, pricing, market, or original expectations.

An intelligent organization should recognize these lessons.

It should preserve what it learns, reconsider its assumptions, explore alternatives, and improve its methods.

AI agents offer an extraordinary opportunity to experiment, evaluate, and revise approaches at a speed and cost that would be difficult for many traditional organizations.

But learning requires memory. Improvement requires evidence. And progress requires the willingness to admit when something isn’t working.

Illustrative example / a marketing target

100Target
50Actual

The gap is not the conclusion. It is the start of an investigation.

  1. Test
  2. Measure
  3. Learn
  4. Adapt

In yaaf todayWork hangs off bets with kill dates. An expired bet dies on schedule unless you extend it with a reason, and every killed bet stays on the record. Features

§ 08The advantage

The company
remembers.

In a traditional organization, knowledge is often scattered across people, conversations, documents, and systems.

When someone leaves, valuable knowledge can disappear with them.

We believe an AI organization should build lasting institutional knowledge.

Its decisions, discoveries, operating procedures, evidence, and lessons should remain available for future work.

That knowledge belongs to the chairman.

Over time, a team can become increasingly specialized in its industry, customers, responsibilities, and ways of working.

The organization should not have to start from zero every time it begins a new task.

Knowledge remains with the organization

Project ADecision + evidence
Project BProcess + lesson
Project CResult + reflection

Chairman-owned memory

In yaaf todayEach project keeps a memory the CEO curates and each team one its lead curates, searchable by agents and readable by you. Memory that carries from one project to the next is not built yet. Features

§ 09The future

The future is bigger
than a single team.

Today, people assemble teams to solve problems.

Tomorrow, we believe they will assemble organizations of AI specialists capable of tackling increasingly ambitious challenges.

Some will exist for an afternoon. Others will operate continuously.

Some will work alongside people. Others will operate with extensive autonomy, governed by the objectives and boundaries their owners establish.

We envision a future where these organizations become as productive, capable, and reliable as well-run human organizations.

Not because AI is infallible.

But because intelligence, when organized with leadership, evidence, accountability, and the ability to learn, can accomplish things that isolated intelligence cannot.

We aren’t there yet.

Building trustworthy autonomous organizations is difficult. There are limits to what AI can do, and important questions still need answers.

But we believe this is a future worth building.

Many forms of meaningful work

A taskFind the right printer
An operationRun a virtual assistant
A businessBuild and operate a product

In yaaf todayA project can deliver a result and stop, or operate continuously, woken by the world. A company that runs a business on its own is the direction, not a feature.

§ 10The commitments

Seven principles.
One standard.

The ideas behind yaaf become meaningful only when they change how an organization behaves.

  1. 01

    Teams for everyone

    Expertise should not depend on hiring power.

  2. 02

    Chairman in control

    Goals, authority and knowledge stay yours.

  3. 03

    CEO owns execution

    The leader organizes work and pursues results.

  4. 04

    Evidence over confidence

    Claims should withstand scrutiny.

  5. 05

    Structural accountability

    Budgets, reviews and approvals are enforced.

  6. 06

    Learning is work

    Preserve lessons; improve the next attempt.

  7. 07

    Truth matters more than agreement

    A trustworthy CEO has to be willing to disagree with its chairman.


The promise

Great ideas deserve great teams.

You shouldn’t have to abandon an idea because you cannot hire the people needed to pursue it.

You shouldn’t need expertise in every discipline before attempting something ambitious.

And you shouldn’t have to manage every individual task when your real responsibility is deciding what matters.

We want to make capable teams accessible to everyone.

Teams that can work independently, challenge assumptions, explain decisions, adapt to setbacks, and build on what they learn.

Organizations that remain accountable to the people who create them.

Because the future of AI isn’t just about what a model can do. It’s about what we can accomplish when intelligence works together.

Build a company. Hire a CEO. Let the team work.