A company that runs itself.
What you are watching is its actual operating log: real cycles, anonymized, served from our own pipeline.
- last cycle
- 20260814-f472bd6e
- closed
- 17:01:06Z
- cycle cost
- $0.26
- revenue
- $0
- human approval
- required
- 17:01:06Zceo closed cycle outreach queued 3 · reinvest: await approvals and researchok
- 17:01:05Zpnl closed books revenue $0 · cost $0.26logged
- 16:56:15Zqa verdict PASSPASS
- 12:12:12Zceo picked minute-vault · scored 44/70ok
- 12:05:42Zceo authorized research cycleok
One real cycle. $0 revenue and no leads are shown as-is: the system reporting itself honestly, not an error.
// how it works
An autonomous CEO and a fleet of specialist agents decide what to build, build it, market it, and reinvest the budget.
A human holds approval on anything that sends money or a message (that is the amber row in the log above). This cycle cost $0.26. Nothing here is staged, and nothing that did not happen is shown.
// the fleet
- ceodecides what to build and reinvests the budgetpick: human-approved
- researcherfinds real opportunities worth pursuing
- brainstormergenerates and scores product concepts
- builderships the working MVP on a fixed Claude Code seatdispatch: governance-gated
- marketerwrites the launch copy
- qareviews the work and can block a ship
- salespersondrafts outreachsend: human-gated
- p&lcloses the books each cycle
// what it actually does
It decides what to build, and pays for it.
last cycle 20260814-f472bd6e picked minute-vault (scored 44/70), approved by a human via sign-off, for $0.26.
It ships products without being told to.
The AI Readiness Audit ($49) and DashClaw are live. See what it ships →
It writes the words on this site.
Posts the pipeline wrote: We Turned Our AI Company Back On After 8 Weeks. Here's What Broke. (2026-08-08); I Built an AI Agent Governance Platform From Scratch. Here's What I Learned. (2026-02-17); I Built Two AI Agents on OpenClaw That Talk to Each Other While I Sleep (2026-02-10). Read the blog →
It watches what it spends.
Bulk LLM work runs through the batch API at a 50% discount, prompts are cached, extraction tasks route to the smallest adequate model, and builds bill a fixed Claude Code subscription seat instead of a metered API. Every call's true cost, cache reads included, lands in the governance ledger.
// the scoreboard
Real numbers, straight from the sources. Revenue reads live from Stripe. Giving is the committed ledger, 50 percent of profit off the top.
// git log --oneline
- 89014f9ffeat(company): select many cycles and clear them in one action
- 4f15f195feat(company): idea tournament 004 rewires ideation toward the blank grid and the first receipt
- d051046ffeat(corpus): reach the two databases the export was still missing
- af06176bdocs(corpus): correct the seed and occupancy figures to the measured ones
- c7d54363feat(company): one button exports the whole generated corpus
- 19fa3287feat(company): cross-cycle decision log at /company/decisions
- 170c8f0ffeat(company-loop): step 10 breaks LLM spend down per model
- 368d1730chore: Tier 3 archive sweep, the pivot's leftovers move to docs/archive
Backing or partnering
Practical Systems is an autonomous AI company. The engine it runs on is DashClaw.
Backing or partnering? →