Archetype / operating layer
The systems that run without you.
Archetype is the operating layer behind every venture Sebastien Poulet runs. One operator. A constellation of autonomous systems working while he's asleep. Nothing irreversible is ever left to judgment.
Read the operating doctrineEvidence
The first system is already running.
Daemon operates Seb's own development pipeline unattended. The tiles below are illustrative telemetry of what it tracks; the verified mechanism is documented in the playbook.
trailing 7 days
trailing 30 days
all-time
since last manual fix
Illustrative telemetry from a running instance. Every system below eventually reports here, live and daily.
The architecture
One operator. Four systems. No bottleneck.
Each system is scoped to one part of the operation and answers to the same rules. None of them wait on Seb unless the decision is genuinely his.
-
DaemonOperating now
The operator. Reads, decides, and acts inside the permission tiers Seb set, escalating only when authority is genuinely ambiguous.
-
LogosOperating now
The memory. Everything Archetype has learned about each venture, structured so nothing has to be re-explained to the next agent.
-
PrometheusOperating now
Growth. Runs marketing and sales across channels on its own schedule. A human approves the send; nobody drafts it from scratch.
-
OrpheusIn development
Artist and live operations, end to end. Not open yet: the operator doesn't ship what it hasn't proven.
The doctrine
Anything irreversible sits behind a hard gate. Everywhere.
Credential scope, spend caps, human approval on anything external-facing: every system above enforces its non-negotiables in infrastructure, not in a system prompt a model could talk itself out of. Judgment is for taste. Gates are for consequence.
A rule that lives in a prompt is not a rule. It has to live in the architecture.Archetype's operating doctrine
Questions
Asked before the descent.
What is Archetype, Sebastien Poulet's operating layer?
Archetype is the operating layer behind every venture Sebastien Poulet runs: one operator and a set of autonomous agent systems (Daemon, Logos, Prometheus, Orpheus) that work inside hard, architectural gates, so nothing irreversible is left to a model's judgment.
How can an AI agent team run unattended without doing something irreversible?
Put every irreversible action behind a gate that lives in infrastructure, not in a prompt: scoped credentials, spend caps enforced before the spend happens, and human approval on anything external-facing. Judgment is for taste; gates are for consequence.
How do you stop AI agents from falsely claiming work is done?
Audit the claim instead of trusting it: after an agent reports done, the runner re-executes the required checks itself. In Archetype's own suite, 22 deterministic checks run against a stubbed agent CLI and cost zero tokens.
What systems make up Archetype?
Four systems: Daemon (the operator), Logos (the memory), Prometheus (growth: marketing and sales), and Orpheus (artist and live operations, still in development).
Is there a playbook for running an AI agent dev team?
Yes; the operating doctrine is a free public playbook on this site. The companion Agent Team OS kit is $99 at launch ($149 after), and done-for-you setup is $1,500 with 2 slots per month.
Get in
The proof is public. The rest is built in the open.
Daemon's operating model, the hard-gate architecture, the audit mechanism behind it: written up as a free playbook. Everything else above is running against Seb's own ventures before it runs anyone else's.