How do a few people run two companies with agents?

Alterra owns two companies. Alterra Studio, opened in 2025, is a design and development studio for technology companies. Alterra Editions, started in 2026, is a media company that makes short films about the past. Its first brand, Then & Theory, is on Instagram as @thenandtheory and at thenandtheory.com.

A few people run both. That works because most of the repeat work is done by a crew of software agents we wrote ourselves, each with one job and a written brief.

What does the agent crew actually do?

Each role is a program that runs on a schedule and reads its brief before it starts. On Then & Theory the roles are:

• Research and fact-check finds sources, checks dates and claims, and pulls reference images from museum open-access collections.
• Production builds to written briefs. Each agent owns its own files, so two agents never edit the same thing.
• The critic reviews every piece before it moves on, and never builds anything itself.
• The quality gate runs automated checks on the finished cut.
• The producer keeps production two days ahead, reruns a failed stage once, and collects everything that needs a person into one batch.
• Publishing posts on a schedule and records the link to each post.
• Analytics takes regular snapshots of every film’s numbers.

A manager role reads those numbers each morning and rewrites the growth brief that every other role reads first.

A film goes on the schedule only when the quality gate passes and the critic says post. The reasoning behind that split is in why the checker never builds.

How do the agents remember what worked?

Each agent starts every job cold, so the memory lives in files: a brief for each role, a production playbook, a style guide and a schedule. When an episode teaches us something, the lesson goes back into the relevant brief and the next run starts from there. Our shorthand for this is “briefs are files, not memory.”

One example of a lesson going back into a brief: analytics showed that the share of people who skipped in the opening seconds tracked views closely, and caption wording made no difference. The opening picture and the pacing were the problem. That finding is now a check the critic runs on the first three seconds of every film.

What do people still decide?

Anything that commits money or the company. Agents never spend money, accept licenses or open accounts. When a role needs one of those, the producer puts it in a single batched request to a person.

People also set direction and standards: which companies we run, what each brand is for, what the critic judges against, and how much compute a film may use. When an overnight crew on one large episode used far more than planned, the fix was a human rule. Full crews now need sign-off, and each part of a scene gets at most two passes before the critic sees it.

Who watches the crew when nobody is at a desk?

A check runs every 30 minutes and costs nothing. It wakes the manager only when two things are both true: no episode log has moved in 45 minutes, and fewer than two posts are booked ahead. The manager then finds what is stuck and restarts or nudges that role in a team chat people can also read and write in.

What goes wrong, and how is it caught?

Most failures are plumbing. Early on, a director agent started helpers in the background and then finished, which shut the helpers down, and the next script rendered half-built shots. The fix was a policy line in the shared brief (no background work inside a run) and a check that every shot is marked as built before anything renders or goes to review.

The other common failure is drift in quality, because a builder that likes its own work will pass it. That is why the critic is a separate role and the gate is automated.

Is the crew fully autonomous?

No, and we don’t describe it that way. The crew runs scheduled, repeatable work so that people can spend their time on direction and judgment. We pick tools agents can run without anyone clicking through a screen, mainly Blender, Python and ffmpeg. Apps that only work by hand are optional extras and never a dependency.

What does this mean for a studio client?

Alterra Studio runs on the same base: one small team, one design and 3D pipeline, and the same software. When you hire the studio you work with a small team led by Ryan Hofmann, and the slow parts of the job are already automated, including research, versioning, checks and renders. There is more on the setup in how we work.

In short:

• Agents do scheduled, repeatable work, each to a written brief.
• Lessons are written back into the briefs, so the next run starts from them.
• Only the critic and the automated gate can put a film on the schedule.
• People own money, contracts, accounts, direction and standards.

If you want that setup working on your brand, site or launch film, book a call with Alterra Studio.