AI Impact Brief
The AI Amplification Effect
What two years of DORA data and five industry reports reveal
PRs merged vs. delivery stability. Same teams. Same year.
AI is accelerating individual output while organizational outcomes stay flat. The variable that determines which side you land on? The quality of your engineering practices.
2024: The warning shot
When DORA's 2024 Accelerate State of DevOps Report landed, the AI findings surprised everyone. For the first time, AI adoption showed up in the data—and it correlated with worse delivery outcomes, not better.
The high-performing cluster shrank from 31% to 22%. The low-performing cluster grew from 17% to 25%. A 25% increase in AI adoption predicted a 1.5% decrease in delivery throughput and a 7.2% decrease in delivery stability. The only positive AI signals were documentation quality (+7.5%) and perceived code quality (+3.4%).
The mechanism was batch size. As Laura Tacho observed: "AI introduces risk not because of garbage code, but because batch size seems to increase." More code per change means more risk per deployment. And 39.2% of developers reported low or no trust in AI-generated code.
2025: The amplification thesis
A year later, DORA didn't just update the data—they rebranded the entire report. The 2025 edition is titled State of AI-Assisted Software Development. That name change tells you where the industry's center of gravity has moved.
The throughput story reversed: task completion up 21%, PRs merged up 98%. But stability stayed negative—bug rates up 9%, PR size up 154%, review time up 91%. The net organizational impact? Flat.
"AI is an amplifier, not a fixer."
— DORA State of AI-Assisted Software Development, 2025
Teams with strong engineering practices saw AI multiply their effectiveness. Teams with weak practices saw AI multiply their dysfunction. The variable wasn't AI adoption—it was the quality of what AI was amplifying.
The 2025 report also introduced a fifth metric—Rework Rate—measuring the percentage of unplanned deployments to fix user-facing bugs. It captures exactly the hidden cost that throughput metrics miss: shipping faster without shipping better.
The industry converges
DORA isn't alone. Five major 2025 reports tell the same story from different angles:
Up from 31%. "Almost right" code is the #1 frustration.
AI saves ~10h/week. Org friction loses ~10h/week. Coding is only 16% of dev time.
22% of merged code is AI-authored. "Adoption doesn't equal impact."
80% of new devs use Copilot in first week. Volume ≠ quality.
of organizations qualify as "AI high performers." The rest see 10–20% cost reductions but can't translate them into sustained delivery improvement.
Every report tells the same story: AI is accelerating individual output while organizational outcomes stay flat or degrade. The variable that determines which side you land on is the quality of your engineering practices—the structural health of your codebase, your testing discipline, your review processes, your knowledge distribution.
Free 1-hour DX coaching session
A focused session for engineering leaders who want to understand what's actually happening in their codebase.
- Complete data tables from DORA 2024 and 2025
- Side-by-side comparison of 5 adjacent reports
- DX Coach capability mapping to industry findings
The data: report by report
DORA 2024 — AI Impact on Delivery Metrics
| Metric | Impact per 25% AI Adoption Increase | Direction |
|---|---|---|
| Delivery throughput | -1.5% | Worse |
| Delivery stability | -7.2% | Worse |
| Documentation quality | +7.5% | Better |
| Perceived code quality | +3.4% | Better |
| Code review speed | +3.1% | Better |
| Time on valuable work | -2.6% | Worse |
| Individual productivity | +2.1% | Better |
DORA 2025 — The Amplification Effect
| Metric | AI-Assisted Change | Direction |
|---|---|---|
| Task completion | +21% | Better |
| PRs merged | +98% | Better |
| Bug rates | +9% | Worse |
| PR size | +154% | Worse |
| Review time | +91% | Worse |
| Net organizational impact | Flat | Neutral |
Adjacent Reports — 2025 Convergence
| Report | Sample | Key Finding |
|---|---|---|
| Stack Overflow 2025 | 65K+ devs | 46% distrust AI output (up from 31%). "Almost right" is the #1 frustration. |
| Atlassian DevEx 2025 | 3,500 devs | AI saves ~10h/week. Org friction loses ~10h/week. Net: zero. Coding = 16% of dev time. |
| GetDX Q4 2025 | 135K devs | 91% adoption. 22% of merged code is AI-authored. 3.6h/week saved. "Adoption ≠ impact." |
| GitHub Octoverse 2025 | Platform | +25% commits YoY. +23% PRs merged. 80% of new devs use Copilot in first week. |
| McKinsey State of AI 2025 | Enterprise | Only 6% are "AI high performers." 10–20% cost reductions in software engineering. |
What this means for your codebase
Each industry finding maps to a structural signal in your code. DX Coach measures these signals deterministically—no surveys, no self-reporting, no dashboards that show you what you want to see.
| Industry Finding | Source | DX Coach Dimension | What We Detect |
|---|---|---|---|
| Batch size inflating risk | DORA 2024/25 | Code Health | Complexity hotspots, structural complexity scoring |
| Convention drift accelerating | DORA 2025, SO | Code Health | Convention variance detection, pattern drift scoring |
| Bug rates increasing (+9%) | DORA 2025 | Change Safety | Test quality assertions, change-readiness analysis |
| Review time ballooning (+91%) | DORA 2025 | Change Safety | Code churn hotspots, coupling analysis |
| "Almost right" AI output | SO 2025 | Code Health | Intent duplication, code duplication scanning |
| Knowledge concentration risk | DORA 2025 | Team Resilience | Bus factor analysis, knowledge concentration scoring |
| AI-generated insecure patterns | SO 2025, McKinsey | Ops Readiness | Security posture scanning, injection detection |
| Rework rate (new DORA metric) | DORA 2025 | Change Safety | Churn analysis, rapid re-edit detection |
Measure your baseline before AI amplifies it
DORA's amplification thesis has a clear implication: you need to know the quality of your engineering practices before AI scales them. If your codebase has complexity hotspots, convention drift, knowledge silos, or empty tests—AI will multiply all of it.
DX Coach measures exactly the layer DORA identifies as the determining variable. It reads your codebase's structural fingerprints against a catalog of engineering standards—complexity, duplication, test quality, coupling, security posture, documentation freshness, knowledge concentration—and tells you where to invest before AI amplifies what's already there.
No surveys. No dashboards that only show you what you want to see. Deterministic signals from the artifact itself, translated into practices from an open, DORA-aligned playbook.
Free 1-hour DX coaching session
A focused session for engineering leaders who want to understand what's actually happening in their codebase. Live walkthrough of what DX Coach reveals.
Book your session