Certified AI Native Developer
Certified to build with agents: spec-driven development, agentic TDD, and production agent systems with evals.
SPECIMEN. A demonstration entry from the seed registry, published to show the shape of a record. It is not an issued credential and no person holds it.
Every node this credential stands on — its own track up to its level, plus the full prerequisite closure behind that, derived from the skill graph rather than restated by hand. Each is matched to the work on this record that closed it.
- n-l0-1Mental modelsFNDART-11Calibration log: three variants, one driftVERIFIED
- n-l0-2Tool stackFNDART-12Stack map with two identified gapsVERIFIED
- n-l0-3Prompt specsFNDART-13Constraint analysis across three tasksVERIFIED
- n-l1-1First workflowFNDART-14Delivery-loop definition with a caught checkpointVERIFIED
- n-l1-2Personal AI OSFNDART-15Personal AI OS with three days of retrosVERIFIED
- n-l2-1AI native envDEVART-16AGENTS.md and an agent diff that failed twiceVERIFIED
- n-l2-2Spec-driven devDEVART-17Executable spec, agent plan, green diffVERIFIED
- n-l2-3Agent anatomyAGIART-18Agent config with an annotated execution traceVERIFIED
- n-l2-4Tools & MCPAGIART-19Tool contract v1 to v2, with the failure that forced itVERIFIED
- n-l3-1Agentic TDDDEVART-20TDD transcript with mutation-test evidenceVERIFIED
- n-l3-2Review machine codeDEVART-21Risk-tiered review of an agent PRVERIFIED
- n-l3-3Ship an agentDEVART-22Production agent with two green eval runsVERIFIED
- n-l3-4OrchestrationAGIART-23Three-agent orchestration with both gates evidencedVERIFIED
- n-l3-5Agent evalsAGIART-24Three-layer eval suite and the adversarial fixVERIFIED
What AINO-D requires of anyone who holds it. Published terms for the credential class, not a transcript of this holder's assessment.
- 14 nodes: L0–L1 foundations, agent anatomy through evaluation, developer track through L3
- Reviewed capstone evidence: an agent release package with an eval suite
- Running gate — live-reviewed artifact coverage across the required graph
- Planned defense — architecture, evals, and rollback after vetted examiner coverage
This page publishes the credential, the gate it stands on, and the artifacts linked to it. It does not publish artifact bodies, assessment transcripts, reviewer identities, or the holder's name — the registry is public and the work is not. An id that resolves to nothing returns 404, never a page.
Every linked artifact carries its recorded reviewer, date, and verdict. Each review states whether its exact recorded standard matches today's rubric. AI and expert review remain distinct.
- AI REVIEW · MODEL NAMED
- A named model returned the recorded score and verdict. Each review states whether its recorded standard matches the current rubric. This is a machine verdict on submitted work, not a human judgment or a statement about the person.
- AI REVIEW · NO MODEL RAN
- No model reviewed this artifact. A keyword and length check ran because no review provider was reachable, and that check cannot verify anything: the pass flag is forced false on this path, so it can neither mark an artifact verified nor close a graph node. It is recorded because the submission happened, not because it was assessed.
- EXPERT REVIEW
- A human practitioner recorded this verdict under their own account. Each review states whether its recorded standard matches the current rubric. Expert review is recorded separately from AI review; no model produces it and no process converts one into the other.
No expert review is recorded against any artifact on this record. Model reviews and offline checks are identified separately below. Neither is a human judgment.
Mental models · AI Native Foundations
Published today. Each review below states whether its recorded standard matches.
- 01All 3 prompt variants target the same question; outputs are labeled as live observations or authored-example analysis
- 02Analysis checks specific claims against available evidence, records any drift, and does not invent a failure or cause
- 03Trust rules are stated as testable conditions, not vibes
- AI REVIEW2026-03-05PASSEDREVIEWED BYanthropic/claude-opus-4-8RECORDED SCORE83 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
Tool stack · AI Native Foundations
Published today. Each review below states whether its recorded standard matches.
- 01Every tool on the map carries a job label from the four-job taxonomy
- 02Map fits one screen and every entry names a tool the learner currently has access to, not a wishlist tool
- 03Both gaps are tied to a real task and justify a manual, fixed-workflow, or agent approach with a verification check
- AI REVIEW2026-03-19PASSEDREVIEWED BYanthropic/claude-opus-4-8RECORDED SCORE75 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
Prompt specs · AI Native Foundations
Published today. Each review below states whether its recorded standard matches.
- 01After-prompt includes an output schema and defined failure behavior
- 02Both versions run on the same three inputs, outputs attached
- 03Analysis names which constraints actually changed behavior
- 04At least one test input is an edge or break case
- AI REVIEW2026-04-02PASSEDREVIEWED BYanthropic/claude-opus-4-8RECORDED SCORE81 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
First workflow · AI Native Foundations
Published today. Each review below states whether its recorded standard matches.
- 01Definition covers trigger, gather, generate, verify, act
- 02Two run logs on different inputs are attached
- 03Checkpoint analysis states what the gate caught or would have
- 04Hours-saved estimate shows the math
- AI REVIEW2026-04-16PASSEDREVIEWED BYanthropic/claude-opus-4-8RECORDED SCORE88 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
Personal AI OS · AI Native Foundations
Published today. Each review below states whether its recorded standard matches.
- 01Design fits one page: inputs, three AI stages, review gates
- 02Every stage names its review gate explicitly
- 03Retro notes cover three consecutive days
- 04Retro answers the stated weekly question with evidence
- AI REVIEW2026-04-30PASSEDREVIEWED BYanthropic/claude-opus-4-8RECORDED SCORE81 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
AI native env · AI Native Developers
Published today. Each review below states whether its recorded standard matches.
- 01AGENTS.md has exact commands, conventions, and forbidden zones
- 02Agent's diff was produced using only the file as guidance
- 03Failure analysis cites specific guidance gaps, not “model error”
- 04File is patched from the failure analysis — v1 and v2 shown
- AI REVIEW2026-05-07PASSEDREVIEWED BYanthropic/claude-opus-4-8RECORDED SCORE75 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
Spec-driven dev · AI Native Developers
Published today. Each review below states whether its recorded standard matches.
- 01Spec separates unresolved decisions and dependency edges, then contains intent, constraints, acceptance criteria, and a test plan
- 02Acceptance criteria are falsifiable and numbered
- 03Review notes on the agent's plan are attached
- 04Final diff is green on build, lint, and tests
- AI REVIEW2026-05-21PASSEDREVIEWED BYanthropic/claude-opus-4-8RECORDED SCORE88 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
Agent anatomy · Agentic Workflows & AI Agents
Published today. Each review below states whether its recorded standard matches.
- 01Config uses exactly one tool and an explicit stop condition
- 02Trace is complete from task to termination
- 03Each decision point is annotated with the dial it belongs to
- 04At least one poor loop decision is identified and diagnosed
- AI REVIEW2026-06-04PASSEDREVIEWED BYanthropic/claude-opus-4-8RECORDED SCORE81 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
Tools & MCP · Agentic Workflows & AI Agents
Published today. Each review below states whether its recorded standard matches.
- 01v1 contract has a strict input schema and enumerated errors
- 02Documented agent failure is reproducible from a trace
- 03v2 change directly addresses the observed failure
- 04Side effects and idempotency are noted
- AI REVIEW2026-06-11PASSEDREVIEWED BYopenai/gpt-5.5RECORDED SCORE81 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
Agentic TDD · AI Native Developers
Published today. Each review below states whether its recorded standard matches.
- 01Failing tests were written before the agent implemented
- 02Agent did not touch tests and implementation in the same step
- 03Transcript shows failing run → implementation → passing run
- 04One deliberate mutation is caught; the evidence names each applicable feedback layer and explains any omission
- AI REVIEW2026-06-18PASSEDREVIEWED BYanthropic/claude-opus-4-8RECORDED SCORE94 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
Review machine code · AI Native Developers
Published today. Each review below states whether its recorded standard matches.
- 01Every hunk in the PR is assigned a risk tier
- 02Agent's answers on boundaries and assumptions are recorded
- 03Two demanded changes are specific and justified by tier
- 04Review hunts invariants, not style
- AI REVIEW2026-06-25PASSEDREVIEWED BYanthropic/claude-opus-4-8RECORDED SCORE81 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
Ship an agent · AI Native Developers
Published today. Each review below states whether its recorded standard matches.
- 01Agent has scoped tools and structured outputs
- 02Captured events are minimum-necessary, redacted, access-controlled, and retention-bounded
- 03Release gates justify critical misses, coverage, and repetitions from consequence and observed variance; measured cost assumptions name the model/configuration, dated source, sample window, and expiry
- 04The owner has exercised containment, rollback, and recovery verification in the declared order
- AI REVIEW2026-07-09PASSEDREVIEWED BYanthropic/claude-opus-4-8RECORDED SCORE94 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
Orchestration · Agentic Workflows & AI Agents
Published today. Each review below states whether its recorded standard matches.
- 01Gatherer, drafter, critic roles with explicit handoff contracts
- 02Human gate sits before output ships
- 03Two run logs on real material attached
- 04Analysis separates what the critic caught vs the human gate
- AI REVIEW2026-07-02PASSEDREVIEWED BYanthropic/claude-opus-4-8RECORDED SCORE88 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
Agent evals · Agentic Workflows & AI Agents
Published today. Each review below states whether its recorded standard matches.
- 01Suite covers unit, trajectory, and outcome layers
- 02Adversarial case is designed to induce a real failure
- 03Results attached, including the reds
- 04Fix is verified by a rerun of the adversarial case
- AI REVIEW2026-07-12PASSEDREVIEWED BYanthropic/claude-opus-4-8RECORDED SCORE88 / 100Illustrative review
Authored specimen; these scores illustrate the record format.
- Which graph node each artifact was submitted against, with its module's current rubric as a reference. Each review states whether its exact recorded contract matches that current standard.
- Which kind of reviewer produced each verdict. Every review stores one of two values — AI or expert — at the time it is written.
- The date of each review and the verdict it returned.
- That nothing here was verified by an offline check. Verification requires a live model review or an expert one; the heuristic path is forced to fail before a status is written.
- That a human read the work — unless a panel on this page is marked EXPERT REVIEW. Where there is none, the record states it outright rather than staying quiet.
- That the holder can do this work today. Read the review dates; a credential carries a term, not a guarantee of current practice.
- Anything about the artifact's contents. Bodies, per-criterion notes, and reviewer feedback stay in the holder's workspace and are not published here.
- Employment, identity, or conduct. This record covers assessed work and nothing else.