AI coding agents can now write, test and ship code with very little help.
That is great news, until your dashboard shows more pull requests, more commits and more lines of code, and your customers still wait just as long for new features.
To lead an agentic SDLC well, you need the right agentic SDLC metrics. This guide shows the engineering metrics for CTOs that matter, starting with DORA metrics, and the ones that fill the gaps.
Because they count activity, and AI agents create activity for free. Commits, lines of code and pull request counts rise even when real delivery does not. In an agentic SDLC you must measure outcomes instead: what reaches production, how safely and at what cost.
Faros AI analysed data from 10,000+ developers across 1,255 teams. Teams with high AI adoption completed 21% more tasks and merged 98% more pull requests. However, review time increased by 91%, showing that higher developer output does not always mean better business productivity. (Faros AI Research)
A study by METR found another challenge: AI does not always make developers faster. In a trial involving 16 experienced developers and 246 tasks, participants expected AI to speed up their work by 24%. Instead, tasks took 19% longer with AI. (METR Study)
DORA’s 2024 research found that a 25% increase in AI adoption was associated with a 1.5% drop in delivery throughput and a 7.2% decline in delivery stability. Its 2025 report showed that AI can improve throughput but also amplifies existing engineering practices. DORA 2024 | DORA 2025
For CTOs, the takeaway is clear: measuring Agentic SDLC performance requires looking beyond developer output to overall delivery speed, quality, and stability.
Track agentic SDLC metrics in five layers: delivery (DORA metrics), flow, AI impact, quality and risk, and cost and people.
These are the engineering metrics for CTOs that matter most.
Together they answer one question: is AI making our delivery faster, safer and cheaper, or only making coding faster?
| Measurement Area | What It Evaluates | Key Metrics |
|---|---|---|
| Delivery | Measures how efficiently the entire software delivery system performs. | Change lead time Deployment frequency Failed deployment recovery time Change failure rate Deployment rework rate |
| Flow | Identifies where work gets delayed during the development process. | Review wait time Pull request size |
| AI Impact | Measures how much AI agent-generated work successfully reaches production. | Agent-assisted share of merged changes Agent pull request merge rate |
| Quality and Risk | Evaluates software quality and compares the risks associated with AI-generated and human-written code. | Change failure rate by code origin Rework on agent code vs. human code |
| Cost and People | Measures the cost of delivering software and the impact of AI adoption on developer experience. | Cost per shipped change Developer trust and workload |
Think of these five layers as an engineering dashboard. Delivery measures speed, flow identifies bottlenecks, AI impact tracks productivity gains, quality and risk highlight issues, and cost and people reveal resource efficiency and team health.
This approach aligns with the three measurement layers highlighted by most of the engineering analytics platforms: usage, impact, and quality. Adding flow and cost gives CTOs a broader view of Agentic SDLC performance and its impact on engineering outcomes.
DORA metrics are five research-backed measures of software delivery performance: change lead time, deployment frequency, failed deployment recovery time, change fail rate and deployment rework rate. Three measure throughput (speed) and two measure instability (how often things go wrong).
DORA (DevOps Research and Assessment) is a Google Cloud research program that helps engineering teams measure software delivery performance. It originally introduced four key metrics, which expanded to five in 2024 with the addition of Deployment Rework Rate and the updated Failed Deployment Recovery Time metrics.
DORA metrics help CTOs understand whether AI is actually improving software delivery. They matter for three key reasons:
→ Measure speed and stability together: High-performing teams focus on both faster delivery and reliable releases.
→ Understand overall performance: DORA metrics help evaluate delivery practices, team performance, and engineering outcomes.
→ Apply across technologies: These metrics work across different applications, services, and technology stacks, including AI-driven development.
One important rule: Measure each application or service separately. Combining data from different teams or systems can give misleading results.
Agents change how every DORA metric behaves. Coding time shrinks, but review and test queues grow. So lead time can stay flat, and change fail rate and rework rate can quietly rise. Read the five metrics together and compare agent-assisted work with human-only work.
| DORA Metric | What Agents Change | What to Watch | A Healthy Sign |
|---|---|---|---|
| Change Lead Time | AI agents speed up coding and test writing. Review and CI waiting times can become the biggest parts of the delivery process. | Lead time remains flat or increases even as coding time decreases. | Lead time decreases as review waiting time also improves. |
| Deployment Frequency | AI agents make smaller code changes faster and cheaper, potentially increasing release frequency. | Deployment frequency increases, but changes become larger or releases are bundled to manage review workloads. | Deployments become more frequent while change sizes remain stable or decrease. |
| Failed Deployment Recovery Time | AI agents can help investigate incidents and suggest fixes, provided that monitoring and rollback mechanisms are in place. | Recovery takes longer as code volume grows and fewer team members understand the changes. | Rollback is fast, and time to detect and recover from failures decreases. |
| Change Failure Rate | More AI-generated code can reach production when testing is weak. DORA's 2024 research linked AI adoption with lower delivery stability [3]. | The failure rate increases for agent-assisted changes compared with human-written changes. | The failure rate stays at or below the human-only baseline. |
| Deployment Rework Rate | AI agents may generate code that passes initial tests but requires unplanned fixes after deployment. | The number of unplanned deployments increases following production incidents. | The rework rate remains stable or decreases as deployment volume grows. |
Not on its own. DORA metrics provide a strong foundation for measuring software delivery performance, but Agentic SDLC requires a broader view. CTOs also need to understand how AI agents contribute, where workflows slow down, whether code quality improves, and how AI impacts engineering costs.
This is where Azilen takes a broader approach. Along with DORA metrics, we focus on additional measures to evaluate AI impact, workflow efficiency, code quality, and engineering costs. This helps businesses understand not just how fast they deliver software, but whether AI is making the entire development process more efficient and reliable.
DORA provides the performance baseline, while these additional metrics help identify bottlenecks, measure AI contributions, and guide continuous improvement across the SDLC.
Track pull request wait time, review time, and pull request size. In an Agentic SDLC, bottlenecks often shift from coding to code reviews, testing, and approvals.
When AI generates more code, teams need more time to verify it, slowing down overall delivery.
Count the full cost: licenses, model and compute spend, review time and rework. Divide it by what actually reached production. And plan for a J-curve, because early results often dip before they improve.
The DORA 2026 research highlights three reasons for the early dip:
→ The Learning Curve: Teams need time to learn new AI-driven workflows, temporarily reducing feature development.
→ The Verification Tax: More AI-generated code means more time spent reviewing and validating changes.
→ Pipeline Adaptation: Testing and approval processes need to scale to handle increased development output.
They measure results tied to the task. Airbnb tracked how many files passed each check. Amazon counted developer-years saved and hosts removed. Stripe tracks merged agent pull requests per week and limits CI loops. None stopped at “suggestions accepted”.
| Company | Reported Results | How It Works |
|---|---|---|
| Airbnb | 75% in 4 hours Approximately 3,500 test files were migrated, with 97% completed after four additional days of tuning. |
Each test file was tracked through migration stages, including testing, linting, and type checks. This helped identify exactly where files were getting stuck. |
| Amazon | 4,500 developer-years Amazon estimated the time saved through Java dependency migrations and reported $260 million in annualized efficiency gains. |
Amazon measured time savings based on migrated Java dependencies and cost savings from removing hosts after upgrades. |
| Stripe | 1,300+ pull requests per week Stripe reports that its coding agents produce more than 1,300 merged pull requests every week, with every pull request reviewed by a human. |
Each agent is limited to one or two full CI rounds, helping control execution costs and prevent unnecessary retry loops. |
Pick one service. Set a DORA baseline in week 1, tag agent work in week 2, build a one-page view in week 3, and hold your first monthly review in week 4. Keep it simple and fix the biggest bottleneck first.
Avoid seven mistakes: counting activity, trusting one metric, comparing very different teams, turning metrics into targets, ignoring review capacity, tracking individuals, and judging results too early.
CTOs should track DORA metrics, workflow efficiency, AI agent contributions, code quality, engineering costs, and team productivity. Together, these metrics show how AI affects overall software delivery.
The cost depends on the number of AI agents, tools, integrations, infrastructure, and governance requirements. A small pilot may require a limited investment, while enterprise-wide adoption can involve higher costs for platform integration, security, monitoring, and ongoing AI usage.
AI-assisted development mainly helps developers with tasks such as writing code, debugging, and generating tests. Agentic SDLC extends AI involvement across multiple stages, including planning, coding, testing, code reviews, and deployment, with human oversight.
Yes, Agentic SDLC can reduce repetitive tasks, automate workflows, and help teams complete development activities faster. However, the actual productivity improvement depends on code quality, review time, testing capacity, and how well AI agents fit into existing workflows.
Common risks include insecure code, incorrect AI-generated outputs, excessive agent permissions, compliance issues, and increased review effort. Businesses should establish testing, access controls, monitoring, and human approval processes before scaling adoption.
Implementation time depends on the existing development environment, AI readiness, integration requirements, and project scope. Businesses can begin with a focused pilot and gradually expand AI agents across development workflows based on measured results.
"*" indicates required fields
5432 Geary Blvd, Unit #527 San Francisco, CA 94121 United States
320 Decker Drive Irving, TX 75062 United States
6d-7398 Yonge St,1318 Thornhill, Ontario, Canada, L4J8J2
71-75 Shelton Street, Covent Garden, London, United Kingdom, WC2H 9JQ
Hohrainstrasse 16, 79787 Lauchringen, Germany
12, Zugerstrasse 32, 6341 Baar, Switzerland
5th floor, Bloukrans Building, Lynnwood Road, Pretoria, Gauteng, 0081, South Africa
12th & 13th Floor, B Square-1, Bopal – Ambli Road, Ahmedabad – 380054
B/305A, 3rd Floor, Kanakia Wallstreet, Andheri (East), Mumbai, India
22 mins
Talk to Our
Consultants
Chat with
Our Experts
Write us
an Email