Founders are not short on data. They are short on unedited reality.
The modern executive stack is full of softeners: dashboards, summaries, meetings, AI briefs. Each promises reality without the bruise.
Vendors sell them the oldest executive fantasy with a new AI budget: central planning that finally works. One room, one map, one AI chief of staff stitched across Slack, docs, CRM, tickets, metrics, and people data. A business that finally stops arguing back. That is why the pitch feels like relief, and why it is the trap. Most AI management pitches are CEO dashboards with better autocomplete. They promise a business that can be understood from one seat, then quietly make it changeable only from there.
Central planning fails at local knowledge, not effort or intelligence. The center can set direction, fund the commons, and enforce standards. It cannot know enough, soon enough, to make every useful move from above.
Use AI to strengthen central planning and the company will look modern while getting slower. Use it to make local work compete and the company compounds.
Inside a company, a free market has to be engineered. It needs memory: where the pain started, what someone built, who adopted it, and whether the company changed. Otherwise local work becomes another pile of demos and anecdotes.
The center still chooses the game. It just stops pretending it can price every move from above.
A better company does not make headquarters omniscient. It turns the thing someone built in frustration into work others can pull: a draft, a script, a path, a workflow. The useful work travels. The rest has to earn another day. That is the point: not to make the company easier to command, but harder to fool.
AI Makes Agency Auditable
AI does not just give the center better sight. It changes who gets to claim helplessness.
The person closest to the problem used to have to gather the context, find the policy, write the pitch, wait for the meeting, and hope the idea survived translation.
Now the first move is cheaper. A complaint can become a draft. A recurring failure can become a workflow someone else can steal.
The old alibi was coordination: missing tools, unclear mandate, no time, permission in another room. Real constraints still exist: access, time, incentives, authority. AI does not erase them. It makes the boundary auditable. Show the draft. Show the attempt. Show where it stopped.
Not perfectly. Not forever. Enough to separate the alibi from the constraint.
Inside a company, a bid is not a complaint, a ticket, or a strategy doc. It is a first version with evidence attached: here is what changed, here is what broke, here is the decision I am forcing.
A bid also needs an owner. AI can draft the move. It cannot stand behind it. Attempts get cheap; ownership keeps them real.
That is the bargain. People owe the company more than explanations. Leaders owe the conditions that make agency honest: access, time, authority, and real stakes. Ask for agency while withholding those conditions, and you are laundering extra work through ambition. Show up empty-handed, and the org chart is not the first thing on trial.
Auditable agency makes central planning more expensive, not less. Managers who only route permission become latency. ICs who only diagnose run out of cover. Companies that cannot absorb local building turn AI into theater.
Free Markets Find the Bruise
Markets beat central planning at one job: catching the bruise while it is still fresh.
Leadership sees across the system. People closest to the work feel the customer papercut, the broken handoff, the stupid weekly ritual, and the SQL query quietly holding the business hostage.
If that truth has to climb the ladder before it can matter, the company has already chosen latency. By the time the center understands the problem, the person who felt it has learned to route around it, complain about it, or stop caring.
Useful work usually starts there: selfish, specific, and irritated. Someone in support knows which logs matter, which project to open, which tenant context changes the answer, how to rerun a failed audience export without making the failure worse. That judgment is company memory trapped in a person. A free market’s first job is to turn it into something others can pull before the bruise becomes folklore.
Before Gaia, that support memory usually became a Slack tip or a document: useful once, hard to reuse, easy to lose. Slack is still where the bruise announces itself: loud, public, messy, and weirdly efficient. But a company cannot inherit a Slack thread.
Gaia is our bet that company memory should run, not sit in docs. It is not just a monorepo; it is an executable map of the company, where living systems, apps, agent tools, history, rationale, and runnable knowledge sit close enough to co-evolve. Teams and agents can inspect the path, pull it, change it, and make the company more legible to itself.
Work has to travel before it is blessed, and that movement has to leave marks: who pulled it, what changed, who owns it now, and what budget or attention it consumed. Without that memory, the market collapses back into Slack: visible heat, noisy demand, no price.
Free Markets Need Institutions
Free markets inside companies are dangerous. Without institutions, they become markets for heat: the loud demo, the charismatic owner, the Slack thread that feels like demand because everyone can see it. Assume the market will be gamed. Build until gaming gets expensive.
The currency is not enthusiasm. It is usage pulled into real work, owner time, budget, reduced pain, and changed behavior. Praise is not demand. A project with no owner is not a bet. A tool that saves one team by pushing its cost onto three others has not cleared the market. It has exported pain.
Do the unglamorous work. Maintain the paths. Merge duplicates. Protect boring work. Notice when public evidence becomes public performance. Markets do not replace management. They force management to do its real job in public.
The company has to make truth-seeking cheaper than politics: show receipts, name the owner, price the risk, and make polished theater too heavy to carry.
At GrowthLoop, Gaia and Grimoire split the job. Gaia creates supply. Grimoire prices consequence. Local fixes become shared machinery, then named bets with owners and a way to die.
Every six weeks, Grimoire turns the portfolio into a docket. The owner has to stand in public with the facts while there is still time to change the bet. Grimoire does not just show reality. It forces the company to decide while the facts still matter.
In one recent review, the record made the evasions visible: cold bets wearing active names, owners carrying work because nobody had called it dead, and leaders treating fear and ambiguity as weather instead of work.
So we moved capital. We stopped funding most of those bets, moved people to the few still worth backing, and made the standard explicit: a lead is not done when they report resistance. They own turning resistance into a decision.
That is the directed chaos bargain: build locally, face standards publicly, and let the institution decide what survives.
Founder Power Must Leave Artifacts
That bargain still needs a center. Some bets still belong there: strategy, taste, culture, standards, incentives, safety, budgets, existential risk, and the commons no local team would fund early enough.
Founder power belongs here too. Some bets have no natural owner yet. Founders can absorb risk, spend trust, fund the commons, and say, “We are moving now,” before the market has enough evidence to price the move.
Those are war powers. Use them. The failure is when the emergency becomes the constitution.
The failure is not concentrated power. The failure is power that leaves no artifact.
Compound Marketing Engine was one of those founder bets. The category did not exist as a clean budget line, so no local team could prove demand yet. So the center made the call: we reorganized around it, moved people, killed a multi-month bet that no longer fit, and made hiring decisions for the initiative. Then the bet had to go back through standards, ownership, and consequence.
That power earns its place when it comes back with receipts: risk named, owner assigned, budget visible, success criteria stated, review date set. Founder intervention becomes central planning when it is constant, hidden, or detached from local evidence. Intervene loudly. Own the bet. Make the cost visible. The best use of concentrated power is to make more of the company powerful.
Power does not get better by pretending it is harmless. It gets better when reality has enough independent paths back into the room that power cannot quietly grade its own work.
Theory Tested Through Pressure
That was the model. The test was whether it could survive pressure.
A year ago, the bets we made demanded a faster company. We were slow and cautious; moving carefully was not going to teach us enough. So we pushed speed first, knowing more motion would expose what the system could not handle.
It did. More work surfaced real breakage: customer pain, platform instability, and drag hidden inside the team’s week. That is when central power had to act. We called a platform stability Code Red: stop normal motion, name the cost, and fix the spots hurting customers and slowing the team.
Within three months, the facts changed. High-urgency incidents were cut in half. Resolution time fell from four days to less than one. Pager acknowledgements went from a coin flip to nearly automatic.
Shipping did not just recover. It kept compounding. A year ago, about 30 engineers merged 700 PRs per quarter. Last quarter, a smaller team merged more than 1,900 PRs (more than 3x PRs/person)!
PR count is easy to fake. Slice work smaller, ship cosmetic churn, stay safe. I only care here because the obvious failure mode did not show up: incidents fell, recovery sped up, more than 70% of the work still landed inside named bets, and stale work got harder to carry.
The important change was not 1,900 PRs. It was the shape of the work. More of it arrived with the pain that caused it, the person carrying it, and the artifact others could reuse. A project could not stay alive just because it sounded important. A local fix could not stay invisible just because it started small.
Most companies get slower when they raise standards. They get procedural, political, and approval-bound. The machine learns to call friction maturity.
We got faster because useful work stopped waiting to become official. AI made the first move cheaper. Gaia gave those moves a place to live. Grimoire forced owners and review. The center still made the bets no local team could justify early enough.
The Company Learns
AI has already removed enough barriers to building that companies are being flooded with drafts, scripts, workflows, fixes, and tools from the edge. People are building whether the company wants them to or not. The choice is what the institution does when the work arrives. Turn it into summaries, dashboards, approvals, and cleaner permission queues, and the company will know more while changing less. Make the work bite instead: a draft changes a workflow, a script earns users, a fix exposes a standard, and a tool makes the next decision harder to dodge.
The company that wins is not the one that sees more. It is the one where truth moves money, people, and power.