The DX AI-assisted engineering impact report
DX's Q1 2026 benchmark of 400+ companies found 92.7% AI adoption and 3.9 hours saved per developer each week, yet quality swung both ways and CI waits and meetings still cost more time than AI saves. Enablement and measurement, not tool choice, decided the outcome.
Written by
-
Benedikt Stemmildt Founder & Co-CEO
Updated:
Evidence 7
Our position. We hold that adoption is easy to measure and tells you almost nothing. If installing tools produced the gains, the usage curve and the throughput curve would move together; they do not, and the space between them is the organization (our reading of the DX research (opens in a new tab)).
DX’s numbers bear that out. Its quarterly benchmark uses one methodology across 400+ companies and reports quarter-over-quarter deltas, which makes it the closest the industry has to a neutral reference (State of AI Impact in Engineering: Q1 Report (opens in a new tab)). Adoption sits above 90%, while PR throughput for non-users and daily users differs by less than one pull request a week.
Period covered: 2025-10-17 to 2026-02-01 Sample: 400+ companies, <50 to 10,000+ employees, North America / Latin America / Europe / APAC Report cadence: Quarterly (follow-up to Q4 2025 report; DX plans to continue quarterly)
What did DX measure across 400+ companies?
Adoption and time savings
| Metric | Value | Change vs. Q4 2025 |
|---|---|---|
| Industry AI adoption (monthly+) | 92.7% | Up from ~90% |
| Non-users / no-data | 7.3% | — |
| Average weekly time savings | 3.9 hrs | Up from 3.6 hrs |
| Daily users’ time savings | 4.72 hrs/week | — |
| Weekly users’ time savings | 4.04 hrs/week | — |
| Monthly users’ time savings | 3.63 hrs/week | — |
| Shadow-AI users’ time savings | 3.34 hrs/week | — |
AI-authored code
| Cohort | % AI-authored | Change |
|---|---|---|
| Average | 27.4% | Up ~5pp |
| Daily users | 30.8% | Up from 24.1% |
| Weekly users | 26.9% | Up from 21.8% |
| Monthly users | 24.5% | Up from 20.2% |
| Unknown / no telemetry | 22.3% | Up from 20.2% |
PR throughput (median weekly)
| Cohort | Q1 2026 | Q4 2025 |
|---|---|---|
| Daily users | 2.4 | 2.2 |
| Weekly users | 1.9 | 1.7 |
| Monthly users | 1.6 | 1.5 |
| Non-users | 1.5 | 1.2 |
DX’s separate longitudinal study, from November 2024 to February 2026, showed PR throughput gains of only 10-15%, far below what self-reported time savings imply. The AI productivity studies find the same gap between perceived and measured speed.
Tool performance (daily user PR throughput, weekly)
- Cursor: 4.1 (up from 2.8 in Q4 — +46% QoQ)
- Claude: >4.0 for weekly/monthly users (up from 2.6/2.2)
- GitHub Copilot: 3.61 (up from 2.5)
- Tabnine: 1.83 (lowest; concentrated in large enterprises)
Newer AI-native and agentic tools lead by a widening margin.
Quality is volatile
DX tracked Change Failure Rate, Change Confidence, and Code Maintainability quarter-over-quarter per company. Findings:
- Some companies saw CFR rise by up to ~2 percentage points against a 4% industry benchmark — roughly 50% more defects shipped.
- Companies swung significantly in both directions on Change Confidence and Maintainability.
- No uniform “AI improves quality” or “AI degrades quality” pattern. Outcome depends on measurement, training, and integration strategy — not on tool choice.
Developer onboarding
- Time-to-10th-PR: 33 days for daily AI users (down from 39 days in Q3 2025 — cut by more than half since Q1 2024).
- Onboarding curve continues to steepen quarter over quarter.
Role evolution
- Engineering managers using AI daily ship 4x the code of non-users (doubled from 2x in Q4). DX calls this the “player-coach shift.”
- Designers and product managers using AI: 75% (up from ~60% in Q4). Handoff friction between design/PM and engineering is decreasing.
Company size: reversal
- Q4 2025: larger companies (1,000+ devs) saved more time.
- Q1 2026: smaller companies (<200 devs) now lead in time savings per developer, despite larger companies having higher overall adoption.
- Interpretation: smaller firms completed rollouts first and moved into sophisticated, higher-value use cases.
Enterprise segment
- Traditional / regulated enterprises lead adoption over big tech. Rollouts are more deliberate, structured, and governance-heavy — and that correlates with higher adoption in the long run.
Shadow AI
- Users with no telemetry from sanctioned tools still report meaningful weekly usage, time savings, and AI-authored code.
- DX’s prescription: acceptable use policies over access blocking. Provide better internal alternatives.
Language-level gains
- Modern languages (Python, Java, Go) see the strongest time savings because models are trained on them.
- Rust leapfrogged competitors in Q1 2026. The Rust compiler acts as a feedback partner for agent self-validation loops; Claude 3.7 and Gemini 3.0 crossed a reasoning threshold on Rust-specific issues (lifetimes, async).
Non-AI factors still dominate
Annualized developer time lost to CI wait times, meeting bloat, approval bottlenecks, and context switching eclipses the time AI saves on coding. DX’s conclusion:
AI is a local optimizer. Global optimization requires fixing the human and systemic processes surrounding the code.
What do practitioners in the report say?
On context engineering and systematic prompts
Teams at Google are starting to set some team norms about how they interact with their code base in the form of standard prompts so that every agent that’s deployed to their code base follows a certain set of prescripted prompts that helps standardize the way that the code is written.
Team-level standardized prompts, not individual prompt skill.
On blurring roles
We’re blurring the line between an engineer and other disciplines. The path from an idea to a prototype to something that can be passed on to an engineer has widely changed in the past few months, and I think it’s going to speed up the way we can develop and ship features.
On AI as local optimizer (key recommendation)
AI is saving nearly 5 hours a week for top users, but CI wait times and meeting bloat still eclipse these gains. AI is a local optimizer. Global optimization requires fixing the human and systemic processes surrounding the code.
On enablement as success factor
For engineering leaders, these findings underscore that successful AI integration requires more than just tool deployment; it demands thoughtful measurement, implementation strategies, and proper training.
How does DX measure AI impact?
DX groups metrics into three pillars:
| Pillar | Metrics |
|---|---|
| Utilization | AI tool usage (DAUs/WAUs), % PRs AI-assisted, % committed code AI-generated, tasks assigned to agents |
| Impact | Core 4: PR throughput, perceived rate of delivery, DXI, code maintainability, change confidence, change fail %; human-equivalent hours (HEH) |
| Cost | AI spend per developer, net time gain per developer, agent hourly rate (HEH / AI spend) |
The load-bearing idea is the separation of AI metrics (what is happening) from Core metrics (whether it is working). Adoption without outcome is a vanity number; outcome without adoption cannot be attributed.
What changed against the previous quarter?
- Junior engineers overtook staff+ engineers in daily time savings for the first time. Consistent high-frequency usage moved juniors up the learning curve to peak efficiency (4.9 vs. 4.8 hrs/week).
- Smaller companies (<200 devs) now lead in efficiency, reversing Q4.
- Rust leapfrogged other languages on time savings due to reasoning model improvements plus compiler-as-feedback-loop dynamics.
- Engineering managers doubled output again (2x → 4x) — player-coach role consolidation.
DX’s three 2026 recommendations, summarised:
- Shadow AI risk is high — move from blocking to providing internal alternatives.
- CFR swings of ~2% = 50% more defects. Focus shifts from speed to validation. AI-driven automated testing is no longer optional.
- AI is a local optimizer — fix the human and systemic processes around the code.
A fourth point runs through the report body without being listed as a recommendation: structured enablement and measurement determine AI success, not tool choice.
Related
- The DORA report on AI-assisted development: DX’s earlier reading of DORA, on developer experience and the KPIs that follow.
- AI productivity studies: METR, Stanford, DORA and Multitudes: the controlled and longitudinal studies to set beside DX’s benchmark.
- Amazon Cost to Serve Software (CTS-SW): an outcome metric in money, where DX counts hours and pull requests.
- The State of AI-Native Engineering report: what leaders changed in their organizations while adoption rose.
- Measuring impact: how to separate usage metrics from outcome metrics in your own team.
Resources
Articles
- AI Impact Report: Q1 2026 (opens in a new tab) - DX blog post summarizing its Q1 2026 benchmark on adoption, time savings, quality swings and the non-AI factors that still cost more time.
Papers and reports
- State of AI Impact in Engineering: Q1 Report (opens in a new tab) - DX’s quarterly benchmark across 400+ companies on AI adoption, time savings, PR throughput and quality, with its three-pillar measurement framework.