DORA puts a price on AI-assisted development: the J-curve, the verification tax and the real drivers of return

7-minute read · Updated 28 October 2026

• AI is an amplifier. Returns depend on platform quality, workflow clarity and team alignment, not on the tools alone.
• Expect a J-curve. A temporary dip comes from learning, the verification tax of reviewing AI code and downstream processes that have not adapted.
• An illustrative 39%. In DORA’s 500-engineer model, $8.4M invested returns about $11.6M in year one, with payback in about eight months.
• Gains depend on the work. Research cited in the report puts them at 35–40% on simple greenfield tasks and about 10% or less on complex legacy code.

The formula is the classic one: return equals value minus investment, divided by investment. The interesting part is how value is built. DORA traces it from capabilities such as a quality internal platform, version control practices and AI-accessible internal data, to delivery metrics, to developer and user experience, and only then to cost savings and revenue.

The most useful part of the report is its warning about the first months. Teams have to learn the tools. Someone has to verify what the AI wrote, and that review is slower than people expect. Testing and change approval were built for a slower flow of changes. Together they produce a dip before the gains, which DORA describes as the tuition cost of transformation.

Leaders who read the dip as failure may cut the funding just before the return arrives.

There is a second tax. Earlier DORA research linked AI adoption to more delivery instability. The report’s sample calculator raises the change failure rate from 5% to 6% and puts the cost of the extra downtime at about $344,000. Its advice is to invest in automated testing, continuous integration and small batches rather than to slow down adoption.

Averages hide the spread. The report cites Stanford research showing a large gain on simple new code and a small one on complex legacy systems. It also notes that inference prices fell about 280 times between late 2022 and late 2024, which makes the investment side cheaper every year while the value side still depends on the work.

• 90%

  of technology professionals used AI at work in DORA’s 2025 research.

• ~8 months

  payback in the illustrative model, with returns expected to compound in years two and three.

• No cuts

  DORA discourages headcount reduction as an AI strategy and favours retention and training.

1. 1

   Count the whole investment

   Seats, credits, API usage, extra CI and the time spent learning. The tooling line is only part of it.
2. 2

   Measure value at the delivery level

   Lead time, deployment frequency and change failure rate say more than keystrokes or accepted suggestions.
3. 3

   Run three scenarios

   DORA recommends conservative, realistic and optimistic estimates instead of one confident number.
4. 4

   Watch the instability tax

   Track change failures and rework alongside speed, or the downtime will cancel the gain.
5. 5

   Re-measure after the dip

   A first-quarter number is the tuition, not the result.

Every ROI calculation needs an honest investment figure. Nythrex provides that side for engineering: the AI, CI and cloud cost of each change, with the basis of each number shown. It does not measure people’s productivity or rate individuals.

Why Nythrex

• InfoQ — New DORA report claims strong engineering foundations drive AI return on investment (May 2026)
• DORA — Insights and research
• Faros AI — Key takeaways from the DORA report 2025