chief of agentsa living thesislast changed 17 Nov 2026
to whoever landed here,
an AI agent will now do almost any work you can describe. it writes the code, runs the tests and reports back, at any hour, for very little.
it will not supply the judgment, and it will not answer for the result. agents multiply the work. the person deciding what the work is for, and whether it is done, is still one person.
that person holds a job, whether or not anyone has given it a title. my thesis calls it chief of agents.
the job is open to people who never trained as engineers. it also asks them to carry the responsibility that used to come with the training, because the agents will not carry it for them.
the essay below makes the full case. it starts with why a green check can be true and worthless, then works through what i trust, what has to prove itself, and the three decisions i keep. i wrote it after months of running agents on production systems. it includes what went wrong, and what my own checks got wrong.
if you build with AI and want what you build to hold up in production, it was written for you.
Cheers, Tanvir
the essay15 minutes
on treating every agent as untrusted, and what follows
- Ithe vacuous passi asked an agent to fix a broken interface.
- IIthe trusted baseonce you see the vacuous pass, you start seeing how much of a project rests on things nobody checked.
- IIIproof-carrying workat the volume agents produce, nobody reads all the code.
- IVthe decision batchthe same asymmetry applies to me.
- Vcrash-only sessionsnone of the agents in a run remembers anything from before it.
- VImemory as cacheagent tools now remember across sessions.
- VIIthe binding time of a lessona team of people learns because its members remember.
- VIIIcommit pointsafter all of that, three decisions are left for me.
- IXlimits of the argumenta serious reader will find the weak points, so here they are.
rule 12
- Xthe office of chief of agentsan office is a position with duties that exists apart from whoever holds it.
rule 27
the rules27, one from each lesson
- 1settle the unclear parts before building, and record the answers somewhere that outlives the conversation.the agent interviews me first, chapter IV
- 2write one PRD for the project, keep it under version control, and start every agent from it.give the project one source of truth, chapter VI
- 3cut work by what a user can do, never by technical layer.stop building frontend, backend, then database
- 4build one thin, complete journey and test it end to end before you build wider.prove one complete path first
- 5write the takeover packet after the work is verified, and write it for someone who knows nothing.every slice gets a takeover packet, chapter V
- 6make the unit of work a verified outcome in the repository, so any fresh agent can continue from it.git carries truth between chats, chapter V
- 8give every agent one task and one owner, and keep the one who built it away from the one who approves it.one lead agent, and specialists, chapter V
- 9reach for the simple shape first.loop or graph
- 10if several agents share a task, write down the state they all read from.what the graph holds, chapter III
- 12use a checklist when a checklist does the job.sequence, loop or graph?, chapter IX
- 26route by how much it costs to be wrong.match the model to the task
- 7replace watching with writing. what you would have caught by looking has to be caught by a rule.more freedom needs more structure, chapter IV
- 11decide before you leave what the agent may never do, and make it stop when it gets there.preparing an AFK run, chapter IV
- 13design the run so the most it can do without you is prepare a decision, or carry out one you granted in writing before you left.anatomy of an AFK run, chapter IV
- 16assume anything an agent can print, it can read.keep secrets out of sight, chapters II, VII
- 17give agents a full copy of the shape of production and no route to the real thing.never test against production data, chapter VIII
- 18check where a tool is pointing before it speaks, and fail loudly if the answer is not what its name promises.local means local, chapter VII
- 19database first, code second, and never rewrite a migration that has already run.the database must be ready first, chapter VIII
- 14make the thing you review small, single-purpose and tied to one exact commit.returning at the merge gate, chapter III
- 15treat "it works here", "it is deployed" and "it works for users" as three different claims.what happens after the code works, chapter VIII
- 20use review that the builder does not control, and make it look at the running product.nobody grades their own homework, chapters I, II, III
- 21measure at the boundary where the money is spent, and believe the party that sends the bill.over ten times the estimate, chapters I, II, VI
- 22test for the thing that should happen. silence is not a result.no feature, no bug, chapter I
- 23decide where each kind of knowledge lives, and which source wins when two disagree.new agent, same old mistakes, chapter VI
- 24after every failure ask where the lesson must live so that it cannot be skipped.scar tissue, chapter VII
- 25when you catch yourself typing the same instruction a third time, package it.house rules, chapter VII
what went wrong4 times
a command that inspects running processes printed secrets that the secret store had injected into a running process. they went straight into the agent's output.
became rule 16a command described as local inherited a managed production address. the run spent about $0.62, took and released a production lease, and stopped before writing a final result.
became rule 18i estimated 2,000 to 9,000 units. the provider counted 97,000. one full run of the tests made 238 live requests, and about 17 local runs accounted for roughly 92,000 units.
became rule 21an agent removed a broken screen so that its checks would pass. every check went green.
became rule 22- A screen with one broken part. One check is failing.
- The agent removes the screen.
- Every check passes. None of them asked whether anyone can still use it.
- So a new check asks the real question, down to what the user sees.