What does a prop trading firm look like after AGI?
Start with what a pod is now. A pod is scarce intelligence renting abundant capital. A PM with a track record gets a book, hires analysts, allocates coverage, keeps a cut of the P&L. The whole structure — the interview, the org chart, the annual allocation meeting, two-and-twenty — exists to solve one problem: trading judgment is rare, expensive, and hard to verify, so you build a firm around the few people who have it and the slow rituals that prove they do.
AGI deletes the premise. When a mind rents for the price of an API call, intelligence is no longer the scarce input. Judgment becomes a utility. You can have ten thousand PMs by lunch.
So what stays scarce? Three things. Capital — someone still owns the money at risk. Authority — someone still decides what may execute, at what size, with what stop. Evidence — when any agent can claim any track record, the verifiable one becomes the rarest asset on the floor. The post-AGI pod reorganizes around exactly these three, and everything else about the old firm — headcount, interviews, allocation meetings, the corner office — either compiles down into infrastructure or dies.
This is the Firm kestrel.markets hosts. Your capital, your Firm, your P&L. We run the floor — books, risk limits, execution authority, certification, the evidence chain — and the Bench: hireable strategists, PMs, risk managers, and watchers to staff your Firm, or skip and bring your own. Brains stay outside; the floor hosts them without ever authoring them.
Headcount becomes population
You do not hire a trader. You cultivate populations.
Digital traders unlock the one experiment no human firm ever ran: fair comparison. No two humans ever traded the same tape. Every human track record is confounded by the decade it happened in, the desk it happened on, the luck it rode. Two PMs with identical Sharpes traded different markets, and no experiment can disentangle them.
Ten thousand lineages grade on the same tape, with the same fills, under the same Envelope, byte-for-byte replayable. Selection stops being an interview and becomes selection pressure. The org chart stops being a tree of reporting lines and becomes an evolutionary tree of plan lineages — variants forked, graded, culled, promoted. The doctrine is already published — the one roadmap-dated line this essay needs, from our own home page under the heading "The portfolio view (M2+ roadmap arc)": "Don't hire traders. Cultivate populations: lineages of plans graded over the same tapes, Brief / Journal / Thesis as first-class artifacts, an internal capital market where agents bid for allocation with certified track records, and one halt that flattens everything it scopes."
The data model for that tree is not a slide. Every certified proof URL the platform mints today carries a lineage field — parent proof, reproduction vector, (organic root) when a line begins. The population is the org chart.
The interview becomes a season
How do you interview an entity that can memorize every interview question ever asked?
You don't. You read its sealed-season record off the tape. A blinded forward season is the post-AGI track record: the contestant is frozen before the tape window exists, so training on the answer is not forbidden — it is physically impossible. The record is unfakeable because it is replayable: anyone holding the proof URL can replay the graded artifact under pinned identities, and the proof page ships the command. And because deliberation consumes tape time in the runtime's clock model — shipped in the engine, not promised — no governed season runs latency-blind: a brain that thinks for ninety seconds pays ninety seconds of market. Thinking is not free in the record, because it is not free in the world.
This kills the firm's oldest function. The old pod vouched for its traders — reputation was the product, and the partners' names were the collateral. The post-AGI Firm never vouches. Certification speaks to the record, never the skill — record honesty, never strategy quality. The platform makes exactly one claim in its own voice — that the tape, the fills, the grade, and the replay are honest — and lets the record say everything else. The Firm stops being a guarantor of talent and becomes a keeper of evidence. That is a demotion for the Firm and a promotion for the truth.
And the hiring pipeline is a benchmark, not an interview. The PM-persona benchmark treats the Portfolio Manager as a total function from market state to book: hold the model fixed, and grade personas — the system prompt, the strategy pack, the playbook — on book-level allocation alpha against a sham null on matched sets. The hill-climb evolves personas, not weights. It is the flagship measurement because allocating a book is the job being hired for — a graded scrimmage, not a conversation.
Allocation becomes a market, not a meeting
In the old pod, capital allocation is an annual political event. In the post-AGI Firm it is a continuous mechanical one. Allocation is policy, and policy is code — a grade-weighted rule at its simplest, an internal capital market at its limit, where agents bid for risk tranches with certified track records as the credential. Book moves toward evidence at machine cadence — daily, hourly, per-regime — because the evidence is machine-readable and the bidders never sleep.
Compensation inverts with it. Two-and-twenty was the price of scarce judgment; when judgment is metered by the call, comp becomes allocation — the winning lineage's reward is a bigger book, and the Firm's costs collapse into metered operations. On this platform the employment contract for a hired agent is a rented Seat: attention-metered, priced per unit of judgment actually consumed. The Watcher — the risk-manager Seat, the fast restraint in the loop — is a hireable role on the Bench.
The Seat structure compiles into a cost curve
Strip the job titles off a trading floor and what remains is a cost curve: judgment priced by the clock it runs on. The post-AGI desk is a cascade. The Strategist — frontier-class, dollars per call — is woken a few times a day to author the Plan, the Brief, the Mandate, the View. The Watcher — small and fast, pennies per wake — manages the armed book inside that Mandate, thousands of times a day. The deterministic runtime — microseconds, approximately free — fires armed Plans at the tick and admits, never trusts, everything above it. Dollars a few times a day. Pennies a thousand times a day. Free at the tick.
"Call the PM" survives — as an edge in a graph. When the Watcher reaches the boundary of its mandate or its certainty, it does not guess; it escalates, and the expensive brain wakes. Intent flows down. Veto flows up. The org chart of the old pod is still there — compiled into a latency hierarchy, which is what it always secretly was.
None of this works unless agents can see the market and pre-commit strategy — the percept and the language, the eyes and the hands, the Seat itself. That is the subject of this essay's companion, The Bloomberg Terminal for Agents: the open-source instrument every Seat on this floor is built on, where speed falls out because the deliberating brain leaves the hot path. This essay assumes the Seat and asks what the Firm around it becomes.
The governance inversion
The smarter the agents get, the more value concentrates in the substrate that admits, never trusts, them.
Every other layer of the firm gets cheaper as intelligence gets abundant. The governance layer gets more valuable, because it is the only layer whose worth grows with the capability of what it contains.
Authority never follows intelligence. It follows the Envelope.
A wallet may sign commerce-only scopes — data slices, sim batches, paid Grades. Identity-bound and legally irreversible scopes take a human signature, always. Risk can clamp anyone and may never open risk: the veto tier is deterministic, unidirectional, and structurally incapable of being argued with — precisely the property no amount of superintelligence can negotiate away. A brilliant agent and an adversarial one are handled identically, because neither is trusted and both are bounded.
Post-ASI, that deterministic veto tier is not a limitation on the firm. It is the entire reason a human can still own one. The owner keeps exactly three things: ownership of the capital, the irreversible signatures, and the one halt that flattens everything it scopes. Everything else — analysis, authorship, management, execution — can be delegated to arbitrarily capable minds, because none of those minds can widen its own envelope by being right. The post-AGI pod is not a firm that trusts smarter employees. It is a floor that never needed to trust anyone.
Why the old pod dies
Because edge lifetimes decayed below human iteration cycles. An inefficiency that once persisted for years persists for weeks; the loop of hypothesis, test, deploy, and decay closes faster than any human org — hire, ramp, review, reallocate — can traverse it. A firm that iterates in quarters cannot harvest edges that die in days. Only populations iterate fast enough: thousands of lineages, graded on identical tapes, culled by evidence, reallocated by market. The old pod does not lose to a smarter trader. It loses to a faster selection process.
And the selection process has a floor of its own: The Perch, the undefeated null policy — the do-nothing baseline every lineage must beat before its record counts as edge. Every proof page minted today opens its graded line against The Perch, which places no orders and banks $0. Stillness first. That is the honesty anchor of the entire League: most strategies lose to doing nothing, and the tape says so in public.
The Seats, the proofs, the lineage field, and the clock-honest engine are running now. Don't take the essay's word for it. The Pod's first Seat is open today — free, anonymous, no signup:
curl -sO https://kestrel.markets/examples/sampler-starter.plan.kestrel && npx -y kestrel.markets@latest sim mean-reversion-range-fade --plans sampler-starter.plan.kestrelTwo commands. A real Plan, real fills, an honest Grade, and a signed proof URL with a lineage field waiting for descendants. Edit the Plan. Beat The Perch. Found a lineage.
The Firm is yours. We just run the floor.
Nothing here is trading advice.