The frontier moves. The standard holds.
AI Fieldcraft is the practice discipline inside LockedIn Labs: a repeatable path from unfamiliar capability to evidence, controls, release, and measured operation — without creating another course ladder or product.
A discipline inside the training system
AI changes faster than a conventional curriculum can be rewritten. New models, tools, research, and operating patterns arrive before most teams have finished learning the last generation. The wrong response is to make the catalog larger. The useful response is to teach a durable way to absorb the unfamiliar without relaxing the burden of proof.
We call that discipline AI Fieldcraft: reading a new capability precisely, reproducing it, finding its failure boundaries, adapting it to real constraints, and carrying it through evaluation, governance, release, and measurement. It lives inside the LockedIn Labs training system. It is not a separate product, subdomain, role ladder, or consulting label.
The frontier can change every week. The obligation to prove the work does not.
Levels answer distance. Fieldcraft answers method.
The AI Native Operator Framework and AI Fieldcraft solve different problems. The framework's L0–L4 levels describe how far a practitioner can operate. Role tracks aim that capability at the work. Fieldcraft describes how the practitioner trains when the capability itself is new.
The AI-Native Software Development Lifecycle begins where production responsibility becomes real: it governs how a system is designed, evaluated, authorized, released, observed, and improved. Fieldcraft prepares the operator to do that work. It does not replace the lifecycle, and it does not turn a practice artifact into production proof.
The loop is built around evidence
A verb is useful only when it leaves something another person can inspect. Each Fieldcraft stage therefore has an evidence output. The loop can be compressed for a small technique or expanded across a production system, but the obligations stay legible.
Measurement closes the loop rather than ending it. A material model, tool, dependency, policy, or workload change sends the capability back to the source record and through the relevant gates again.
- Read — separate the source, the claim, and the unknowns.
- Reproduce — establish a controlled, repeatable baseline.
- Stress — record failure modes, side effects, drift, and hidden cost.
- Adapt — show what changed for the actual workload and why.
- Evaluate — score the exact candidate against published gates.
- Govern — name controls, owners, approvals, and preserved evidence.
- Deploy — release the smallest safe scope with a tested way back.
- Measure — compare operational results to the intended outcome.
A mission begins with a real change
A Fieldcraft Mission is a source-backed, time-bounded field exercise in the existing practice workbench. The reference record points to a real release, paper, repository, or standard and states when LockedIn last reviewed it. The work packet remains synthetic, self-contained, and safe to practice against. No learner needs to paste employer data into the platform, and no fictional case implies a provider or customer relationship.
The first mission starts from OpenAI's 2026 hosted shell and versioned Skills capability. The synthetic case asks an operator to contain an unexpected network action, preserve incomplete evidence, pin the execution boundary, build an adversarial evaluation, and define an authorized recovery. The case will be reviewed when the source capability materially changes or by its stated review date.
Practice is not deployment
AI Fieldcraft is a training curriculum and mechanism, not a software package that LockedIn deploys into an environment. Practitioners develop the capability here. They then apply it when they evaluate, govern, deploy, and operate AI systems in the environments where they are accountable.
That distinction is load-bearing. A Fieldcraft Mission can produce a scored private portfolio artifact, but it is practice-only: it never closes a curriculum node, advances the capability graph, or creates a credential by itself. Real production authorization still belongs to the organization, workload, evidence, and named decision owners involved.
What stays fixed
Fieldcraft adds no new assessment dimensions, scores, levels, roles, credentials, portfolio, feed, Lab engine, or record system. The fixed assessment remains comparable. The graph still derives progress from authored curriculum evidence. Reviews still carry provenance. The existing portfolio remains the evidence record.
That is the design test for the concept: it should make LockedIn's training method more durable and more legible without making the platform more fragmented. New capability enters through the Signal, becomes constrained practice in Labs, and earns standing only through the evidence system already in place.