Notes: what we’re learning.

deepmoat.ai is an OpenAI Select Partner. The relationship supports our work helping people understand where AI can be useful and put an agreed approach into practice.

Our recommendations begin with the work and the people using it. We are model agnostic, and we are also a registered partner in Anthropic’s Claude Partner Network. We consider OpenAI, Anthropic’s Claude, Google Gemini, other providers, and open-source models according to the need.

The relationship broadens what we can bring to the work. Your needs guide the recommendation.

Where OpenAI is a good fit, we connect the possibilities to a defined task, appropriate access, and the support people need to use the result.

Where another approach fits better, we say so. Any relevant referral or resale interests are disclosed before a commitment.

The tools can widen what is possible. People still review the work and stand behind the recommendation.

— Keri Nakamoto, deepmoat.ai

It took us the last quarter of 2022 to find and build the team. From 2023 through 2026, that team built an AI practice inside ANVL Entertainment, our working film and television company. We funded it ourselves. The first focus was internal work: what was AI actually good for, here, now?

We built a lot, and we built early. Script coverage. A development pipeline. A story diagnostic. Storyboarding, including video storyboards. Research tools for the subjects and settings our scripts touched. Decades of producing and operating judgment went into instruments we actually ran. It cost us real money and real time; every lesson was paid for.

We kept testing against the work, not the promise.

At that stage, we chose not to turn those early pilots into public products. The tools and models were changing quickly, and we did not yet see a durable product worth maintaining. We kept the work internal while we learned which capabilities could hold up in practice.

We also killed pilots that worked. That's the part that surprises people. A pilot that demos well and a pilot that survives real use are different animals, and some of ours were the first kind wearing the second kind's clothes — tools that needed more supervision than they saved, builds whose edge would evaporate with the next model release. Working isn't the same as durable. We stopped them early, before they could quietly absorb budget and belief.

We worked across commercial LLMs and smaller open-source models, comparing what each could do against the job at hand.

That experience now informs Folio, our plugin for script coverage, story bibles, pitch decks, and storyboards. It is coming soon. It also informs deepmoat.ai’s work helping people explore where AI can be useful and put it into practice.

— Keri Nakamoto, deepmoat.ai