Thesis
The companies built before AI will not survive the era it defines.
Not because AI will replace them. That framing is wrong, and it cost the market three years of misdirected effort. They will not survive because they were built for a world where intelligence was scarce, expensive and human. That world is ending. Almost none of the incumbents are being rebuilt for the one taking its place.
The consensus response has been to add AI on top: pilots, copilots, chatbots, automation layers, productivity gains. That is the wrong unit of work. Bolting AI onto a business built for scarce intelligence does not produce an AI-native company, any more than fitting an engine to a horse-drawn carriage produced a car. The ceiling most firms are pressed against is not the model. It is process, structure and how decisions get made. The firms doing it loudest will be the most exposed.
AI-native is what happens when the company is rebuilt on the other side of that ceiling. The model becomes a participant in the work rather than a tool applied to it. Knowledge is structured so the model can read it. Decisions get made with the model in the loop. Operations run on a substrate the model can act on. In regulated environments, the integrity layer is engineering work, not posture. The evidence has to hold up to an auditor.
Most firms cannot do this work. Consultancies write the strategy deck and leave before the rebuild. AI vendors sell the tools and assume the integration is someone else’s problem. System integrators deliver the implementation but not the architecture. The missing work is the rebuild itself. A senior team doing the architectural thinking, the engineering work and the cultural change together, in the same room, for the duration.
That is the work somai does. We embed senior teams inside companies and to help rebuild them block by block: cognition, perception, reaction, regulation, execution. Typically in eight-week blocks and with value-anchored pricing. If the substrate does not deliver the value agreed at brief, the outcome bonus is not earned. At the end of the block, the client team owns and operates the substrate.
The next decade will be defined by companies that were built for it. We exist to build them.
Five substrates that we believe make a company AI-native. We work with you to rebuild them one at a time from pilot to production.
Signal architecture, four flow types.
Memory architecture, at a glance.
A block rebuilds one substrate in four stages, typically over eight to ten weeks.
Brief
first two weeksWe meet the leadership team and the company. We read what is written down and listen for what is not. We pick one substrate, agree the hypothesis, and agree the value it earns if the hypothesis holds.
Output: a contracted brief and a value hypothesis, on one page.
Build
weeks 1 to 8A senior team working directly with the the businesses operators. We design in test, build in production, both against real workloads, with the people who will run it after we leave. No parallel shadow IT systems.
Output: the substrate live, observed under load, with the client's team operating it.
Handover
week 9The substrate is aligned across to the business. We document the seams, the failure modes and the next reasonable extensions. You own the code, the data and the playbook outright. No licence, no retainer hook, no ongoing seats. But we continue to support where we can.
Output: a working substrate the client owns, and a thirty-page operator manual.
Interlock
months 1 to 6 afterWe return for a quarterly half-day to read the operating metrics against the value hypothesis. If the substrate has delivered the value, the outcome bonus is earned. If it has not, it is not.
Output: a verdict on the block, written and signed by both sides.
If the substrate does not deliver the value hypothesised at brief, the outcome bonus is not earned. The fee for the work itself stands. Value is priced before the work, contracted against during it, and proved in production after it.
A studio of senior operators. No juniors, no offshore, no associates in the room.
somai is a team who have run the kinds of companies we work in: heads of operations, engineering leaders, CFOs, chiefs of staff, and founders. Everyone has a fundamental belief in the power of AI when properly implemented and built into the foundations of a modern business.
Most of the bench earned its scars in places where ambiguity is not forgiven: under regulators, auditors and risk committees. Those are the surfaces an AI rebuild has to survive. The principles travel; we take work outside them too.
We take the principle that the studio installs and the client operates seriously. We do not sell retainers, managed services, or ongoing seats. We are not the team that runs your operations after we leave. Your team is. Our job is to make sure your team can.
We hold limited capacity for a few deep engagements per year. That is not scarcity theatre. A senior studio at this depth cannot run wider without becoming the consultancy it was founded to displace.
Essays from inside the rebuild. Lessons we’ve found worth writing down.
RSS- 05.04
What stays human
As working intelligence becomes abundant, the scarce thing is no longer analysis. It is judgement, taste, and the will to mean it.
- 05.03
Decisions with the model in the loop
AI as a tool you pick up, or an agent that replaces you. Both pictures miss the real change. The shift is to the loop.
- 05.02
A company your model can read
A company knows far more than it can say. Until that knowledge is made legible, a model cannot do anything useful inside it.
- 05.01
The end of scarce intelligence
Every company alive today was built around the assumption that working intelligence is scarce. That assumption has quietly stopped being true.