A model knows things. A coworker knows what is going on.

An AI coworker needs more than access to Slack, GitHub and Linear. It needs a continuously updated understanding of what the company is doing, what changed, what belongs together, what needs action, what it is allowed to do, and what happened the last time it acted. Today every agent — beckett included — reconstructs that state on wake-up, from scratch, with tokens. That reconstruction is the product Kowo should own.

Airbyte already calls itself "a data and context layer for AI agents"; Glean is building context graphs from enterprise activity; Dust wires company knowledge to agents. Competing horizontally on "we connect your data to agents" is a fight against connector catalogs and enterprise distribution. Kowo can own something narrower and more actionable: turning the exhaust of engineering work into executable work state — events → objects → episodes → state — for a small set of sources (GitHub, Discord/Slack, Linear, CI, Sentry) and make the cross-source understanding unusually good before adding a sixth.

The rest of this page tests that claim against the disk: is there a real episode that already runs the loop? What raw material exists and in what shape? What would it take to produce the "second object" — the curated WorkItem — from the five raw events? And, because the CTO-seat memory records two vetoes in one conversation on 16 Aug (supply is not a moat; name who pays), what does Kowo sell?

The bar you set on 16 Aug (from the memory graph)

Two vetoes in one channel conversation, both saved as calibration nodes: "our data pipeline is not quality at all, so we don't have anything to offer that Reddit or Twitter can't" (supply is not a moat) and "we already can't make money off of you… maybe something that's marketable as a service we provide. Like idk consulting" (name who pays, how much, how soon — before the moat argument). Every lane below is written to that bar. Where a number is a hypothesis it says so; nothing on this page has a customer yet.

Three lanes, ordered by time-to-cash

The moat that isn't supply

The commodity-data veto was right: Kowo will never out-collect anyone. The asset that accrues in a Kowo node is not volume, it is a dataset nobody else can have for that customer:

human activity      (messages, reviews, decisions)
+ company state     (work items, changes, incidents, deploys)
+ agent activity    (runs, casts, tool calls, cost)
+ agent outcomes    (proof, review verdicts, what shipped, what broke after)
────────────────────────────────────────────────────────────
= "when this service fails after this kind of migration, these three checks find it"
  "PRs touching auth need Samantha"
  "this workflow succeeded 18/21 times; the 3 failures were all X"
  "you fixed this class of bug in 0e196c3 three weeks ago"

That last line is already true on this box for exactly one incident, and it took eleven minutes and three dollars to create. It is organizational memory the customer owns, in their environment, that gets more valuable every week they keep the node — and it walks out the door if they churn. That is a retention moat, not a supply moat, and it passes both vetoes: it is not "we collect data", it is "your coworker remembers what happened when it acted last time, and nobody else's does". Your own market-research doc already names memory/company knowledge as the line between a tool and a coworker; Kowo is the place that memory lives.

1. Is "ro" in the channel logs you? The two vetoes are attributed to ro; this page treats them as your bar. If that's someone else, the revenue section needs a second reading.
2. Which team is the first lane-A pitch? A name, or "nobody yet" — both are answers; the second one changes the 30-day plan to "find one" before "build E3".
3. Is Kowo willing to be a services company for two quarters? Lane A bills first and productizes later. If the answer is no, lane B moves up and needs the installer before anything else.
4. Which of the portfolio (beckett, babble, chilltext-arena, beckett-metrics, fal-pipelines, freebuff, pixe-eval, betterwright-bench, 0xbeckett.me) is on the Kowo story and which is not? The Aug 14 review said the CTO seat has no company object; company.md is still a template. Answering "what Kowo is" in one paragraph there unblocks beckett's own free-time thinking about it.
5. Do we keep the events under ~/.beckett or is Kowo a separate process/package from day one? This page assumes "same daemon, new module, second writer" for E1–E3 and a real boundary only when a second consumer shows up. If you want the boundary now, say so and E1 becomes a package first.
6. Which of the six missing surfaces matters for the first customer — Linear, CI, Sentry? As far as this box shows, kowo-co doesn't use Linear or Sentry, so the model can't be proven on them here. The first customer decides connector #2.

The Aug 14 review found five roots and shipped the fixes for four of them in a weekend; the fifth — no company object, a CTO seat that proposes its own parts — is what this page is actually about, seen from the other side. Beckett doesn't have a company to think about because Kowo hasn't said what it is. The cleanest thing to say is: Kowo is the data plane for digital coworkers, and Beckett is the first one.

The evidence that this is not a retrofit is on the disk. Eleven stores already emit the exhaust. The event and job tables already exist, empty. One routine that runs twice a day already produced a complete incident → work item → run → change → deploy → report loop with no human, in eleven minutes, for three dollars — and the only reason it isn't a Kowo record is that the links between the pieces are English. Making those links a column, those pieces objects, and that noticing a query is a backfill script, a resolver module, and one better input to a primer that already exists.

The revenue answer is not "Kowo Data as a horizontal platform". It is: lead with Beckett as the reason to install Kowo, sell the install as a service because services bill on day one and you said so, let every engagement leave the same node behind, and let the node be the thing that gets more valuable per customer per week — because it holds what happened the last time their coworker acted, and nothing else does. A model knows things. A coworker knows what is going on. Kowo is the second part.

"i want to begin thinking of kowo as a company i can posit as valuable and then make and show real value."jason, 16 Aug 2026 — Part 2 is the first piece of real value; Part 4 is how to make it repeatable; Part 5 is who might pay for it. The next thing to make is a buyer.