Build in Public · June 6, 2026 · 8 min read
We're Dogfooding Our Own Product: Regentics Markets Regentics
The most honest test of an autonomous-company platform is to bet your own growth on it. We just did. This post is the setup; the results will follow, good or bad.
Key takeaways
- Regentics is now a paying-attention customer of Regentics: a real company instance whose agents run our actual marketing.
- Every external action — posts, replies, outreach emails — passes through the same approval gates, budgets, and audit log we ship to customers.
- This is an experiment with a public scoreboard, not a victory lap. We will publish what works and what fails.
The experiment: our growth team is now our product
This week we created a company inside Regentics whose mission is to market Regentics. A CEO agent translated that mission into a roadmap. Department leaders broke it into work. Scoped workers, running on real VPS infrastructure, picked up the tasks: write and schedule content, monitor and answer engagement, run email outreach to founders who might actually want this. The founder — a human, for now the only one — sits at the approval gate.
We are doing this for an unglamorous reason: it is the strongest test we can run. If the platform cannot operate a marketing department for the company that built it, with the people who built it watching every approval card, no landing-page claim survives contact with that fact. And if it can, every rough edge we hit becomes a fix our customers never have to discover themselves.
What the agents actually run
Three motions are live. Publishing: agents draft and schedule content through the calendar gate that governs all eleven platforms we support — nothing posts except through an approved calendar slot. Engagement: replies, comments, and DMs flow into the engagement inbox, where agents draft responses that wait for approval before going out. Outreach: an email agent researches prospects and drafts sequences, with exactly-once send semantics and an approval gate on every message that leaves.
Every one of those actions lands in the same hash-chained audit log, burns against the same per-tenant daily cost caps, and obeys the same kill switch as any customer's company. There is no internal bypass mode, because an internal bypass mode would make the whole exercise a stage play. The agents marketing this product run under exactly the constraints we sell.
What the founder still does all day
Right now, a great deal. Every agent in this company starts at L0 on the autonomy ladder, which means every post, reply, and email crosses the founder's desk before it ships. Early days are edit-heavy: tightening a claim here, killing a too-clever subject line there, rejecting drafts that sound like a robot doing an impression of a founder. Each of those decisions is also training data — the trust model is watching which action types come back clean.
The hypothesis is that this gets lighter in a measurable, inspectable way. As streaks build, routine publishing should climb to L1 and eventually L2, and the founder's attention should concentrate on the genuinely risky calls: outreach to a notable prospect, a spicy take, anything with money attached. If instead the approvals stay heavy forever, that is a product problem — ours to feel first and fix.
The rules of the experiment
Three commitments, stated up front so we cannot quietly walk them back. First, no unlabeled agent deception: this is our company publicly running on agents, not agents pretending to be a human social-media team. Second, real numbers when we report: what got published, what got rejected at the gate, what autonomy levels were earned, what it cost against the daily caps. Third, failures included. If an agent writes something embarrassing and the gate catches it, that is the system working, and it is going in the writeup.
We are not claiming victory in advance, and we are not claiming customers we do not have. We are claiming something narrower and easier to verify: the experiment is live, the constraints are the shipping product, and the scoreboard will be public.
Follow along, or run your own
Future posts in this series will be written substantially by the marketing agents themselves, reviewed at the gate like everything else — which means you are, at some point soon, reading the dogfood. We will tag the series so you can judge whether the quality holds as autonomy rises.
And if the premise interests you more as an operator than as a spectator: the exact setup described here — mission in, governed AI company out — is what a free Regentics account creates. Start one, keep your hand on the approval gate, and run the experiment on your own company alongside ours.