The 38 Way.
We get close to the process, find the constraint and build the missing capability. We do not sell a technology selected in advance, and we do not assume every opportunity deserves to be built.
External AI Department: a relationship, not the mission
The External AI Department is one possible way to work together: an external team close to the process, with a clear scope and enough continuity to accumulate knowledge.
It does not define what 38 Frontier Labs is. Proprietary software, products, custom systems, research and reusable modules are complementary forms of the same mission: turning research, technology and knowledge into concrete capabilities. It is not staff augmentation, a partner division or a promise of unlimited presence.
How the work starts
The initial commitment makes the problem decidable before anyone commits to Build. It produces four verifiable elements; if they do not justify the next step, stopping is a valid outcome.
- Process and constraint map
- What happens today, where it breaks and which conditions are non-negotiable.
- Capability brief
- What should become possible, for whom and with which accountable owner.
- Proof criterion
- The evidence that would justify proceeding, changing direction or stopping.
- Decision
- Proceed, do not proceed, test or wait—there is no automatic Build.
We build capabilities, not AI projects.
A capability is something the organisation could not previously do reliably and can then use, verify and govern.
Models, tools and architectures can change. Value sits in the combination of people, process, data, system and controls that makes a new operating behaviour possible.
- 01People
- 02Process
- 03Data
- 04System
- 05Controls
- 06Capability
From constraint to evolution
The method does not start with AI and does not end at deployment. Every phase leaves a decision and an artefact; it can also show that continuing is not worthwhile.
- Observe
- Decision: do we have enough context? Artefact: a map of process, people, data and exceptions. Stop if access or context is insufficient. Partner input: the people who do the work and access to relevant sources.
- Constraint
- Decision: which limit blocks the outcome? Artefact: a constraint statement. Stop if the problem is not important enough or does not require a new system. Partner input: priorities, consequences and non-negotiable constraints.
- Capability
- Decision: what must become possible, for whom and by which criterion? Artefact: a capability brief and accountable owner. Stop if value or responsibility is unclear. Partner input: a decision owner and operating criterion.
- Prove
- Decision: is the path feasible and worth further investment? Artefact: evidence, risks and a recommendation. Stop if proof does not converge or cost is unjustified. Partner input: representative data and experts for review.
- Build
- Decision: does the system meet the agreed criteria? Artefact: a working core with evaluation, controls and failure modes. Stop if the criteria fail in the real context. Partner input: technical access, reviews and integration decisions.
- Deploy
- Decision: are people and process ready to use and govern it? Artefact: an operating and adoption plan. Stop if ownership, fallback or change is unsustainable. Partner input: an owner, access, training and willingness to change.
- Measure
- Decision: do use and benefit justify cost and risk? Artefact: an evaluation record covering behaviour, use, cost and errors. Stop or revise if the criterion is missed. Partner input: usage data, feedback and willingness to compare outcomes.
- Evolve
- Decision: improve, maintain, replace or retire? Artefact: an evolution decision with opportunities and new risks. Stop if no further value is justified. Partner input: changes in process, security, priorities and dependencies.
- 01Observe
- 02Constraint
- 03Capability
- 04Prove
- 05Build
- 06Deploy
- 07Measure
- 08Evolve
The partner’s Capability Roadmap
It orders possible organisational capabilities—not software features and not the 38 Frontier Labs company roadmap.
It makes different decisions explicit: proceed, do not proceed, test, wait, scale, retire or reassess. It is built only when enough evidence exists to decide.
- 01Proceed
- 02Do not proceed
- 03Test
- 04Wait
- 05Scale
- 06Retire
- 07Reassess
From prototype to adoption
A prototype shows that a route may work. A capability also needs criteria, integration into the human workflow, responsibility, security, training, observability, exception handling and fallbacks.
If adoption requires change the organisation cannot sustain, stopping or reducing scope is a valid decision.
Continuous Capability Evolution
After deployment, performance, real use, technology, cost, security, risk, dependencies and the process itself can change. Where the relationship continues, the capability can be reassessed against those changes and new improvement opportunities.
This is not perpetual maintenance included by default or a universal recurring service. Cadence, responsibilities and criteria are agreed case by case.
- If agreed and value remains
- Observe use
- Evaluate
- Maintain
- Improve
- Replace
- Retire
Trust & ownership
Sources, evaluation, acceptance criteria, human oversight, limits, access and ownership are part of the system. The partner keeps domain decisions; code, infrastructure and transfer are defined per engagement.
Sovereign by choice means deliberately choosing models, hosting, data location and dependencies according to risk, without unnecessary lock-in.
The 38 Trust Layer is a methodological direction, not a certification.
You do not need to arrive with a solution.
A real process, a constraint and the willingness to define what should become possible are enough.