Full-stack AI developer
Builds and operates user-facing AI features across product, services, agent tooling, evaluation, and deployment boundaries.
Also observed as · AI application engineer · AI-native full-stack engineer
This is a training route, not an employer screen, placement promise, or claim about your current readiness. A pilot request enters manual consideration only; it creates no invitation, consent, enrollment, reviewer assignment, or payment.
- Target standing
- L3
- Dependency gate
- 18 modules
- Authored module time
- Role proof
- Ship an agent to production
Module time excludes project build, review, and revision time.
A role target with a visible entry and a fixed bar.
The published route begins at L0 with Thinking in models and carries every prerequisite through the L3 target. The assessment can change where you begin; it does not change the proof required.
No module credit is assumed in this public route. Bring prior evidence to the assessment; qualifying reviewed work is the basis for any individual capability claim.
AI literacy & judgment
Chooses model and system boundaries with calibrated tradeoffs rather than treating a model as a feature by itself.
Prompting & instruction design
Turns product behavior into testable instructions, structured outputs, and tool contracts.
Workflow decomposition
Breaks an ambiguous feature into an observable, shippable vertical slice across client, service, and model work.
Agent direction & oversight
Designs tool-using workflows with explicit handoffs, limits, and operator control.
Verification & quality control
Ships a regression-bearing evaluation and rollback story with the AI feature, not after it fails.
Security & governance awareness
Handles data, permissions, and user impact as engineering constraints in the implementation.
Every prerequisite. In learning order.
18 authored modules across 4 tracks. The 12 hr 37 min total is the curriculum estimate, not a promise of elapsed time to capability.
- FND2 hr 59 min
- AI Native Foundations
- 5 modules
- DEV3 hr 42 min
- AI Native Developers
- 5 modules
- AGI2 hr 46 min
- Agentic Workflows & AI Agents
- 4 modules
- SDL3 hr 10 min
- AI Native SDLC & Control Plane
- 4 modules
- 01 · L0Thinking in modelsFND · Calibration log: 3 prompt variants, drift analysis, and your trust rules.27 MIN
- 02 · L0The AI native tool stackFND · One-screen map of your AI stack with job labels and two identified gaps.33 MIN
- 03 · L0Prompting as specificationFND · Before/after prompt pair with constraint analysis on three test inputs.38 MIN
- 04 · L1Designing your first AI workflowFND · Workflow definition + two run logs + checkpoint analysis.43 MIN
- 05 · L1Your personal AI operating systemFND · One-page personal AI OS + three days of retro notes.38 MIN
- 06 · L2The AI native dev environmentDEV · AGENTS.md + agent-produced diff + failure analysis.37 MIN
- 07 · L2Spec-driven development with agentsDEV · Executable spec + agent plan + green diff.47 MIN
- 08 · L2Anatomy of an agentAGI · Minimal agent config + annotated execution trace.31 MIN
- 09 · L2Tools, MCP, and protocolsAGI · Tool contract v1→v2 with documented agent failure.36 MIN
- 10 · L2The AI native SDLCSDL · Eight-phase delivery map for a real agent workstream + gate definitions + accountability labels.46 MIN
- 11 · L2Risk tiers and the control matrixSDL · Risk-tier control matrix for five real systems + the design change classification forced.41 MIN
- 12 · L3Test-driven agentic codingDEV · TDD transcript with mutation-test evidence.41 MIN
- 13 · L3Reviewing machine codeDEV · Risk-tiered review of an agent PR with demanded changes.36 MIN
- 14 · L3Orchestrating multi-step agentsAGI · Three-agent orchestration + run evidence from both gates.51 MIN
- 15 · L3Evaluating agent behaviorAGI · Three-layer eval suite + adversarial fix.48 MIN
- 16 · L3Ship an agent to productionDEV · Working agent + consequence/coverage/variance release evidence + redacted event contract + measured cost sheet + exercised containment and rollback runbook.ROLE PROOF61 MIN
- 17 · L3The verifier layerSDL · Six-class verifier suite wired into a pipeline + signed results for one real change + an induced failure.55 MIN
- 18 · L3Agent observabilitySDL · Instrumented agent + one full trace + three runtime detection rules + a forensic reconstruction.48 MIN
Ship an agent to production
Working agent + consequence/coverage/variance release evidence + redacted event contract + measured cost sheet + exercised containment and rollback runbook.
- NODE
- n-l3-3
- STANDING
- L3
- ALIGNMENT
- AINO-D · Certified AI Native Developer
Completion alone is not verification. Only qualifying reviewed work can support the evidence record; credential requirements remain mechanically derived from the Capability Graph.
A deployed AI product slice
A reviewer can trace the user need, service boundary, model interaction, and operational fallback.
Produced through
An evaluation suite with a release decision
The suite includes representative failures, thresholds, and evidence of what blocks release.
Produced through
A code-reviewable implementation history
Changes show tests, review decisions, and corrections rather than a generated-code dump.
Produced through
Published availability, without a sales fiction.
These are the current Repo-backed delivery formats aligned to the role proof track. A schedule or staffing commitment exists only when its published record says so.
AI Native Developers — C03
- Schedule
- Schedule follows the founding cohort
- Cadence
- Weekly live · 12 weeks
The route names what it does not create.
Operating a high-traffic production service: incident rotation, latency budgets, and distributed-systems depth remain workplace experience.
These are authored role targets already carried by the LockedIn Labs curriculum Repo—not job listings, a labor forecast, an employer screen, or a placement promise. They are not a measurement of any learner. Dimension floors describe the target; only qualifying reviewed work can support an individual capability claim.
No source was used to infer a universal years-of-experience threshold or a current opening count.
Test the operating standard before choosing the full route.
The public Operator Trial compresses one synthetic release decision into about ten minutes. It creates no learner score, record, or hiring signal.