First read
Keri usually carries the first decision read with the client: pressure, commercial stakes, timing, people, commitment surface, and what should stay shapeable.
The room forms around the decision.
DeepMoat is a small private bench. The client sees accountable human judgment; behind it, a deeper human bench and supervised AI agents and subagents help research, compare, pressure-test, prototype, and turn uncertainty into usable next moves.
DeepMoat assembles around the work.
The team forms around each problem. The shape of the work determines who comes closest to the client surface, the prototype, the model layer, or the operating system behind it.
We work from the problem outward: what needs to be understood, what needs to be seen, what needs to be built, what should wait, and what would make the next move concrete.
Who signs the read.
We're not machine learning researchers, and we won't pretend to be. We're operators who started working AI into a real company in 2022 — self-funded, no vendor to please — and earned our judgment the expensive way: demos that collapsed in production, pilots we killed, builds that survived. Before that, I ran a forty-division company, advised families on where serious money should and shouldn't move, and produced films that sold to Netflix and HBO. The bench here is AI. The signature is human. Every recommendation leaves with my name on it.
— Keri Nakamoto, DeepMoat
How the bench is assigned.
The room changes with the paid decision. DeepMoat assigns people, specialists, and supervised AI leverage around what must be understood, tested, protected, built, paused, or taught.
AI fluency and upskilling
Joseph helps clients and teams understand, use, supervise, and adapt AI systems without making the work feel theoretical.
Systems and implementation
James helps turn a clear AI route into working technical direction: model behavior, integrations, automation, reliability, and build choices.
Prototype and workflow surface
Kurtis helps shape usable systems, interfaces, automations, and workflow paths that can hold up close to the client surface.
Human deep bench
Additional specialists are brought in only when the paid scope requires added depth. DeepMoat stays small by design.
AI bench
Supervised agents and subagents help with research passes, comparison, drafting, testing, synthesis, and pressure checks. Human judgment decides what becomes client-facing.
Clients buy the right bench against the right paid scope.
Small public surface. Serious bench.
DeepMoat stays close to the work: client pressure, AI systems, practical builds, and the judgment needed to know what should be shaped, sequenced, or committed.
KE
Founder / first decision read
Keri
Keri started DeepMoat and usually carries the first read with the client. She keeps the work grounded in the actual pressure: what is being decided, what matters commercially, what should stay shapeable, and what kind of intelligence or prototype would make the next move visible.
KU
Prototype and workflow surface
Kurtis
Kurtis works close to both the client surface and the build surface. He helps turn rough ideas into usable systems, prototypes, interfaces, automations, and implementation paths that can hold up inside real workflows.
JO
AI fluency and upskilling
Joseph
Joseph helps people and teams understand, use, and adapt AI systems without making the work feel theoretical. He moves across AI fluency, upskilling, model behavior, open-source model tuning, retrieval, assistants, and practical system use.
JA
Systems and implementation
James
James helps turn AI direction into working implementation. He moves across model behavior, integrations, automation, reliability, and the practical system choices that let a prototype become something clients can actually use and trust.
Behind these first names is a deeper human bench and a supervised AI bench. Both are used carefully: to widen research, sharpen synthesis, test assumptions, and support the accountable DeepMoat read while keeping the work scoped and accountable.
The next useful move comes first.
Sometimes the answer is a sharper read. Sometimes it is a prototype, a system, a model path, a training layer, or a disciplined pause. DeepMoat is built to find the next useful move while the decision is still shapeable.