AI made your team faster.
Delivery got less stable.
Copilots and agents write code faster than teams review, test and release it. The research is in: more output, more warnings in the code, less stable delivery. Agile Analytics measures where that shows up in your loop — from the tools you already run.
AI doesn't fix a team.
It amplifies what's already there.
DORA’s 2025 conclusion, from close to 5,000 respondents. Its 2024 report, from 39,000, shows the split: the work gets better, delivery gets shakier. Same tools, same engineers — the difference is whether anyone can see where the extra output goes.
Better code. Less stable delivery.
Estimated change for every 25% more AI adoption. Documentation, code quality and review speed improved; delivery stability fell. In 2025 DORA found that throughput now rises with AI too.
Measure where it lands.
Output is measured where it is typed. Value is measured where it ships. Agile Analytics reads Git, Jira, CI/CD and monitoring and puts the two side by side, per team:
- FlowLead time, deployment frequency, Sprint Insights
- Quality gatesChange failure rate, MTTR, Error Budgets
- MaintenanceFeature vs. non-feature work, per sprint
- PeopleDevEx surveys, interruptions, Kudos
Also in the picture: Carnegie Mellon followed 806 open-source projects that adopted Cursor (MSR 2026). Velocity rose for two months and then faded; static analysis warnings (+30%) and code complexity stayed up. Bain puts the typical gain from AI assistants at 10% to 15%, and finds the time saved often isn’t put to higher-value work, so it doesn’t show up as a return. Carnegie Mellon 2026 · Bain 2025
Speed is easy to see.
Quality has to be measured.
Four questions every team building with AI should be able to answer from data. Agile Analytics answers all four today.
Lead time for changes and deployment frequency per team, from Git and CI/CD. Sprint Insights shows how much of each sprint actually ships as features. Review wait and PR size as first-class metrics are next.
Change failure rate and MTTR from your deploys and incidents. Error Budgets and SLOs from your monitoring, with alerts when a budget burns. This is where DORA’s “higher instability” shows up first.
Sprint Insights classifies Jira tickets as feature or non-feature work, so the share of each sprint that goes to maintenance rather than features is visible per team — and whether it moves as AI output grows.
US companies spent $4 billion on AI coding tools in 2025, up from $550 million a year earlier (Menlo Ventures), and almost nobody sets that against what shipped. The AI Impact page does: adoption per tool, and cost and return in euros per AI-assisted merge request, next to delivery speed and change failure rate.
Your AI bill,
per merge request.
Connect your AI vendors once. Every night Agile Analytics reads what they billed and sets it against the AI-assisted work that shipped and the hours your people booked. In euros, with the source next to every number.
GitHub Copilot and Cursor report their seats and who used them. Twelve seats and eight people active this month is a number you can take to the renewal.
Licences on one side, token and agent spend on the other, per tool and per model: Claude Sonnet in Copilot, GPT-5 in Cursor, the Claude Code workspace. Converted at the ECB rate of the day.
Spend divided by the merge requests an assistant helped write, recognised from the Co-authored-by lines it leaves in commits. So Claude Code and Copilot can be compared on what they cost per change.
Hours per merge request in each person’s AI-heavy months against their quieter months, from the hours they actually booked, valued at your hourly rate. Shown only when enough people take part to make it reliable.
- Anthropic: Claude API and Claude Code
- OpenAI
- Google Gemini, through your billing export
- GitHub Copilot
- Cursor
- Any other tool, as a licence you enter
Vendors report use per person. We keep that only to link accounts to teams and never show it: an assistant’s use appears when five or more people used it, a team when five of its members did.
Connect. Baseline. Compare.
No new process, no self-reporting. The data already exists in the tools your teams use every day.
Access tokens you create and can revoke, for GitHub, GitLab or Bitbucket, Jira, your CI and your monitoring (CloudWatch, Google Cloud Monitoring). Slack for Kudos, if you want it.
The first sprint gives every team its starting point: lead time, deployment frequency, change failure rate, MTTR, the feature-to-maintenance ratio.
Teams that lean on AI against teams that don’t. This sprint against last. The gates against their targets. The question “did it ship?” gets a per-team answer.
Engineering productivity
has an SLA now.
A team sets targets for lead time, change failure rate and review turnaround, and Agile Analytics tracks every breach. The values below are an example of what that looks like — not customer data.
Stop guessing.
Start measuring.
Connect Jira, your Git host and your CI. Your first sprint report shows where AI output waits, whether the gates hold, and how much of the sprint shipped — from your own data.
- DORA, State of AI-assisted Software Development 2025
- DORA AI Capabilities Model, 2025
- DORA, Accelerate State of DevOps 2024
- Carnegie Mellon University, He et al., Speed at the Cost of Quality, MSR 2026
- Bain & Company, From Pilots to Payoff: Generative AI in Software Development, Technology Report 2025
- Menlo Ventures, 2025: The State of Generative AI in the Enterprise