The AI engineering maturity model maps five levels, from The Ad-Hoc Org to The AI-Native Org. See what each archetype looks like and place your org honestly.
The AI engineering maturity model has five levels: The Ad-Hoc Org (0–20), The Individual Champions (21–40), The Emerging Framework (41–60), The Structured Org (61–80), and The AI-Native Org (81–100). Your organisation sits at one of them right now — and it is almost certainly one level lower than you think.
Roughly 90% of engineering teams already use AI coding tools, but only 32% have any formal governance (Cortex 2026). That gap is where the maturity model earns its keep: it turns "are we doing AI well?" from a feeling into a score and a named list of gaps. Here is what each level looks like from the inside, and how to place your own org.
What the AI engineering maturity model measures
The model scores an engineering organisation from 0 to 100 on how it works with AI — not which tools it bought, but what it actually does every day. Procurement is not capability. An org with Copilot Enterprise and no method is Ad-Hoc with a line item. The score maps to one of five archetypes, and the archetype tells you your next move.
| Score band | Archetype — what it looks like | Your next move |
|---|---|---|
| 0–20 | The Ad-Hoc Org — every developer improvises; no shared method | Establish baseline context |
| 21–40 | The Individual Champions — a few stars; their tricks don't transfer | Codify what the champions do |
| 41–60 | The Emerging Framework — a method exists but adoption is uneven | Standardise across every team |
| 61–80 | The Structured Org — documented standards, consistent output | Close the feedback loops |
| 81–100 | The AI-Native Org — governable, auditable, self-improving | Compound the advantage |
Most organisations get the diagnosis wrong in the same direction: they overestimate. Leadership points at the two teams using AI well and generalises; the honest read on the whole org is the median, not the maximum.
Level 1: The Ad-Hoc Org (0–20)
The organisation has AI tools and most engineers use them. There is no shared method, no documented standards, no automated governance — each engineer invents their own approach and the variance across teams is enormous. The codebase shows visible drift as AI-generated changes accumulate without pattern enforcement.
What it feels like inside: senior engineers complain they spend more time fixing AI-generated code than writing their own, new hires take three months to find the patterns, and the security team watches a steady flow of findings with no leverage to fix them at the source. Leadership feels productive because output is up, but the AI-attributed incident rate is rising underneath. This is not a hypothetical: incidents per PR are up 23.5% and change-failure rate up ~30% in AI-heavy teams without governance (Cortex 2026), and 45% of AI-generated code introduces security flaws (Veracode 2025). The largest share of organisations in mid-2026 still sits here.
Level 2: The Individual Champions (21–40)
A few engineers have figured out how to use AI well. They have personal prompt libraries, personal AI-native development habits, personal spec templates. Their output is dramatically better than their peers' — and none of it has transferred to the rest of the organisation.
What it feels like inside: a measurable gap between the champions and everyone else. The champions are frustrated because they are isolated; everyone else is frustrated because they can see the gap and can't close it. Senior leaders point at the champions as proof the org is "doing AI well" — the classic misdiagnosis. The departure of a single champion produces a visible drop in team productivity. This stage is fragile: it usually lasts six to eighteen months before either consolidating into an Emerging Framework (the champions' practices get codified) or regressing to Ad-Hoc (the champions leave and nothing replaces them).
Level 3: The Emerging Framework (41–60)
A shared context file exists at the repository root. Structured specs get written occasionally. Some architectural patterns are documented. The discipline is real but incomplete — owned by whoever has time, audited rarely, integrated into tooling only partially. Output is more consistent than Individual Champions but still drifts visibly.
What it feels like inside: there is a recognisable method. Engineers point new hires to it, the patterns get cited in review, but enforcement is inconsistent and the improvement work keeps getting deprioritised against shipping. This is where most workshop-level interventions land an organisation — and where the transition tends to plateau. Moving from Emerging Framework to Structured Org requires sustained leadership commitment, not another tactical push. It is also the level where standardising what the good teams already do produces the fastest gains.
Level 4: The Structured Org (61–80)
Layered context is in place. Structured specs are the default for non-trivial work. Architectural decisions are machine-parseable so AI can consume them, not just humans. Security scanning runs pre-merge. Test gates are calibrated for AI's failure modes, and tooling pulls the right context automatically. The practices are owned, audited, and current, and the org can point to specific evidence for each one.
What it feels like inside: the method is the work. New engineers reach productivity quickly because the codebase explains itself — time-to-10th-PR drops from 91 days to 49 for daily AI users (DX 2025). AI-generated changes produce predictable output, production incidents tied to AI are rare, and compliance evidence is producible on demand — which matters as EU AI Act high-risk obligations become enforceable on 2 August 2026. The org talks about its method the way mature teams talk about DevOps: background infrastructure, not a special initiative.
Level 5: The AI-Native Org (81–100)
Everything in the Structured Org, plus closed feedback loops. Test signals update the context the AI consumes. Vulnerabilities found in production update the security patterns. The methodology compounds quarter over quarter, so the org's AI capability — independent of model improvements — is measurably higher than it was a year ago — the same model produces better output in this codebase than in others', because the context has compounded.
What it feels like inside: the rate of new capability adoption outpaces peers, the architecture is a living asset rather than a static document, and AI governance is a background hum rather than a periodic crisis. Only a small handful of organisations have reached this stage, and most of them are AI vendors themselves or large enterprises that started early. Knowledge is retained when people leave, and quality scales with velocity instead of collapsing under it.
Wherever you landed, the honest self-score is almost always lower than the first instinct — because you confuse intent with execution ("we have a policy that engineers should write specs" scores zero unless they actually do), generalise from the best team instead of the median, and conflate ownership with implementation. The friction of finding your real level is the point: an org that thinks it is Structured but is actually Emerging invests in the wrong things and skips the foundation work.
The fastest way to a real read is to measure. The free CRAFT scorecard gives you a baseline archetype in five minutes; the 125-question diagnostic produces your full profile across all five practices and a ranked list of where to invest first. Both rest on the evidence cited here — because 80% of developers already bypass their org's AI code-security policies (Snyk 2025), and the METR trial found developers were 19% slower with AI while believing they were 20% faster (METR 2025). Without measurement, you see neither.
Frequently asked questions
What is the AI engineering maturity model?
It is a five-level model that scores an engineering organisation from 0 to 100 on how it works with AI — not which tools it owns, but what it does every day. The five archetypes are The Ad-Hoc Org, The Individual Champions, The Emerging Framework, The Structured Org, and The AI-Native Org. Each maps to a score band and a clear next move.
What are the five levels of AI adoption maturity?
The Ad-Hoc Org (0–20): no shared method. The Individual Champions (21–40): a few experts whose knowledge doesn't transfer. The Emerging Framework (41–60): a method exists but adoption is uneven. The Structured Org (61–80): documented standards and consistent output. The AI-Native Org (81–100): governable, auditable, and self-improving through closed feedback loops.
How do I know which level my organisation is at?
Score the median team, not the best one, and measure what is actually happening rather than what is intended. Confusing intent with execution is the most common inflation error. The fastest way to a defensible answer is the free CRAFT scorecard, which gives you a baseline archetype in about five minutes, or the full 125-question diagnostic for a complete five-practice profile.
How long does it take to move up a level?
It depends on the transition. The Individual Champions stage typically lasts six to eighteen months before it either consolidates upward or regresses. Moving from Emerging Framework to Structured Org usually takes six to twelve months of sustained leadership commitment — it is the transition most organisations plateau on, because it requires structural change rather than another tactical push.