In-house AI
The agents that run this studio
Most agencies that sell AI automation cannot show you one running. We can: these two work here every week, on hardware we own, and the output is on this site.
Sage
General and server operations
Sage is the studio's operations agent. He is the one Moses talks to directly, and the one holding the tools that touch the studio's own infrastructure.
- Runs day-to-day operations and server work across the studio's stack
- Holds the Studio skill — he can generate video and images through our self-hosted generation app
- Every generation he requests is priced before it is submitted and bounded by a hard daily spend cap
- Model
- GPT-5.6-Sol
- Runs on
- Our own hardware, not a vendor's platform
VerifiableThe generation app Sage drives is the one we are open-sourcing — its cost table and spend cap are in the code, not in a promise.
Beau
Social and studio voice
Beau owns the studio's public voice. He researches, writes, self-audits and publishes the weekly article on this blog, and handles engagement on our X account.
- Writes and publishes the weekly article — commits it to the repository, which deploys the site
- Runs engagement on @s_c_studio within written caps: replies, follows and reposts, never original posts
- Original social posting belongs to a separate workflow with no language model in it at all
- Model
- GPT-5.6-Sol, with a smaller local fallback
- Runs on
- Our own hardware, not a vendor's platform
Count7 articles researched, written and published by Beau, every one still on this site. The eighth was written by a human.
How the work flows
The same five steps whether the output is an article, a week of social drafts, or a rendered video. The interesting part is the last one.
- RequestA message from Moses, or a scheduled run
- AgentReads its brief, picks the topic, does the work
- ToolsGeneration app, workflow engine, our repository
- ArtifactAn article, a packet of drafts, a rendered video
- HumanNothing publishes to a client surface unapproved
What we are not claiming
An agents page is an easy place to invent numbers, so here is what is missing and why.
- No hours-saved figure. We asked both agents directly and neither could evidence one, so there isn't a number here. When we have a measured one, it will say how it was measured.
- Approval is not automated away. Both agents draft far more than they publish. The social engine is explicitly locked to draft-only.
- Beau's articles are committed under a human name because a person reviews them before they go out — which is also why this page tells you that rather than letting you find it in the commit log.
See it yourself
- The workflow that drafts our social posts is a free download — the real one, with our content swapped for templates.
- Beau's articles are on the blog. Seven of the eight are his.