AI amplifies what exists: what DORA 2026 and Gartner say about architecture maturity

Two 2025-2026 publications say the same thing in different ways. The 2026 DORA report on AI in software development concludes that AI is an amplifier: it amplifies the performance of an organization with solid foundations and the disorder of a fragile one (InfoQ). Gartner forecasts that over 40% of agentic AI projects will be canceled by the end of 2027, for three reasons: escalating costs, unclear value, inadequate risk controls. None of these causes is model capability.

What "foundations" means

DORA lists capabilities that condition AI's return: quality internal platforms, good version-control practices, internal data accessible to AI, among others. Translated into architecture language, these are properties of an organization that knows what it has, why it decided so, and how it evolves it. That is precisely what the TEAF maturity pyramid seeks to locate.

The seven levels, reread through AI

• Levels 0 and 1 (documentary chaos, PowerPoint architecture): AI will produce faster in disorder. Each agent adds one more version of information that already existed in three copies. This is where the amplifier amplifies noise.
• Levels 2 and 3 (structured, instrumented): documentation and decision traceability exist. AI can begin to draw on them; gains become possible, but execution remains partly manual, and bottlenecks (review, validation) appear.
• Levels 4 and 5 (connected, living): knowledge, decisions and execution are linked, and gaps are measured continuously. This is the ground where amplification works in the organization's favor, because you can see what agents do and whether it serves the intent.
• Level 6: a horizon, not a prerequisite, as the article on the pyramid recalls.

Gartner's three causes, read in the loop

Each cause of cancellation matches a component of the TEAF loop that, when missing, lets the problem appear:

• Unclear value → Intent. An agent shipped without a measurable intent can't be evaluated: nobody can say whether it succeeds.
• Escalating costs → Observation. Without measuring cost per task and retry loops, cost drifts until the budget review decides. See also our reading of the IVG and IDA indicators.
• Uncontrolled risk → Decision and AI Control Plane. Overly broad rights, unlogged actions, no human validation where errors are costly: that is the subject of the AI Control Plane.

What this reading does not prove

Avoid circular reasoning: these two studies do not validate TEAF. DORA speaks of software engineering; Gartner of a survey of agentic projects. They converge with TEAF's thesis — governance and foundations condition what AI brings — without demonstrating it. TEAF's validation remains the workstream described in the article on field evidence. Saying the correspondence is consistent is honest; saying it is proven would be false.

One concrete action

Before the next agent deployment, place your level honestly with the maturity diagnostic, then ask three questions: what measurable intent? what tracking of cost and results? what limits on the agent's rights and what human validations? If any of the three remains unanswered, that is the next workstream, before the agent.

• DORA 2026: AI amplifies what exists, good or bad; Gartner: cancellations come from costs, unclear value and poorly controlled risks, not models.
• The TEAF maturity pyramid locates an organization's ability to benefit from amplification.
• Each Gartner cause maps to a loop component: Intent, Observation, Decision / AI Control Plane.
• This convergence is consistent with TEAF but does not validate it: field validation is still in progress.

Sources

• InfoQ — New DORA report: strong engineering foundations drive AI ROI
• Gartner — Over 40% of agentic AI projects will be canceled by end of 2027
• Kodus — DORA 2026: the ROI of AI runs through code review