Prove what your engineering delivers.

Measure output, rework, quality and the real effect of AI from the code itself, benchmarked against a large industry dataset. Built on research from Stanford’s Software Engineering Productivity Research Group, on the P10Y platform. We run the evaluation with you.

Book a 20-minute call See a sample read-out

Evaluation run by

use AI at workGoogle DORA, 2025

Experienced developers with AIMETR randomised trial, 2025

After AI, in one 355-engineer organisationStanford research, 2025

typical net gainonce the extra rework is subtracted · Stanford research, n = 2,562

Why the usual metrics mislead.

Activity counts disagree with each other for the same work, and they reward the wrong behaviour.

Lines of code

More lines are not more output, and counting them rewards verbose code.

Commits and pull requests

They reward splitting work into small pieces, not delivering more.

Story points

Subjective, easy to inflate and not comparable across teams.

Self-assessment surveys

A weak predictor of real output (r = 0.17). People are poor judges of their own productivity.

DORA metrics

Useful for delivery speed and DevOps maturity, but they do not measure what was delivered.

Same work, four measures

Four equivalent pieces of work. Activity measures disagree by more than three times; output units rate them the same.

	Eng 1	Eng 2	Eng 3	Eng 4
Lines of code	2,029	1,511	802	583
Commits	21	12	6	26
Story points	5	13	3	6
DORA frequency	Daily	Weekly	Monthly	Bi-weekly
Output units	10	10	10	10

Illustrative example.

How it works.

From your Git history to a read-out, without surveys and without anyone filling in a timesheet.

Your Git repositories

Read-only access to the repositories in scope. Nothing is written back.

Every commit, reviewed

A model trained to replicate a panel of 10 to 15 expert reviewers scores each commit for output, complexity and maintainability.

Output and quality dashboard

Output, rework, quality and the effect of AI, by team, against the industry dataset.

What the model reads in each commit

Cohesion
Complexity
Coupling
Data structures
Interfaces
Methods
Persistence layers
APIs consumed
Architectural patterns
Dependencies
Dependency injections
And more

Research basis: arXiv:2409.15152, which reports correlations of 0.82 and 0.86 with expert panels on two evaluation tasks. Results are set against the research dataset of 600+ companies and 120,000+ engineers (Stanford Software Engineering Productivity Research Group, November 2025). The commercial platform is developed by P10Y Software, Corp.

A sample read-out, explored.

Seven views from an anonymised composite read-out, illustrative only. Hover or tap the charts, switch the series, and move the sliders.

Anonymised composite, illustrative. Built from real engagements and published research; values are redrawn from report charts and rounded. No client is identifiable.

AI in the code.

No surveys. AI use is read from the Git history itself, then set against output, rework and quality. The figures below are illustrative.

Evidence of AI in each commit

Every commit is checked for signs that AI was involved. Pick a signal to see which commits it catches.

claude/add-retry-testsAdd retry tests for payment webhooksAgent branch
copilot-swe-agent[bot]Bump dependency and fix the breaking importAgent as author
feature/invoice-exportExport invoices to CSVCo-authored-by: Cursor AgentMessage pattern
feature/search-filtersFaceted search on the product listConnected tool
fix/login-timeoutIncrease the session timeout for SSO usersNo evidence
codex/refactor-pricingSplit the pricing service into two modulesAgent branch
hotfix/null-currencyGuard against a missing currency codeNo evidence
feature/audit-logAudit log for admin changesGenerated with Claude CodeMessage pattern

6 of 8 commits show AI evidence

“AI-assisted” means evidence that AI was involved, not a claim that AI wrote the code. Commits shown are illustrative.

Adoption by team, quarter by quarter

Share of engineers with evidence of AI use. The rollout rarely spreads evenly.

	Q1	Q2	Q3	Q4	Q5	Q6
Team A	5	15	35	60	75	85
Team B	0	5	10	20	40	55
Team C	10	30	45	55	70	90

0%90%

AI maturity: how well, not how much

Each repository is placed on a four-level ladder. Weighting by output shows whether the most mature practice sits on the work that matters.

L1Ad hoc45%20%
L2Grounded prompting30%25%
L3Agent-augmented20%40%
L4Orchestration5%15%

Before and after, against a 12-month baseline

Index, pre-AI = 100.

Effective outputhigher is better118
Output per engineerhigher is better112
Rework ratehigher is worse135
Churn ratehigher is worse120

Pre-AI = 100

AI spend per output unit

From each tool’s own billing. Lower is better.

Tool AUS$4.2
Tool BUS$7.8
Tool CUS$11.5

US$ per output unit

All figures in this block are illustrative, from an anonymised composite. Works from Git alone; connecting AI tools adds usage and spend.

What the research found.

AI raises activity almost everywhere. Whether it raises output depends on the team and the codebase, which is why it has to be measured.

About 4×
The gap between leading and lagging AI teams widened about four times.
Interquartile difference in net gain: 4.8% in April 2023, 19% in July 2025. 46 AI teams matched with 46 non-AI teams. ^A
R² ≈ 0.2
Token spend is a weak predictor of gain.
Spending more on AI tokens explained little of the difference between 27 teams. ^A
R² 0.3–0.4
Cleaner engineering environments go with bigger gains.
Codebase hygiene explained more of the gain from AI than spend did. ^A
+14% / +1%
In one case study, pull requests rose 14% while effective output rose 1%.
Rework grew 2.6 times; code quality fell 9%, with 3.6 times the variance. ^A
28–38%
Coding agents brought 28 to 38% more commits at every maturity level.
Across 441 repositories. ^B
53% vs 27%
Without committed AI configuration, cognitive complexity rose 53%. With it, 27%.
A few pages of committed rules and context for the agents went with a lower quality cost. ^B

^A Denisov-Blanch, “How to Quantify AI ROI in Software Engineering”, Stanford Software Engineering Productivity Research Group, November 2025.
^B Denisov-Blanch et al., “A Few Pages of Markdown: Committed AI Configuration and Lower Quality Cost after Coding-Agent Adoption” (2026), arXiv:2608.25241.

Why delivery partners use it.

For software engineering vendors, and for the consultancies and integrators that buy from them.

Win more bids with independent evidence

Show a buyer your output, rework and quality against the industry, scored from the code rather than claimed in a proposal.

Defend your rates with output per dollar

Turn “we are good value” into a number: output per engineer and per dollar, next to onshore alternatives.

Move to output-based pricing

A shared, neutral unit of output makes it possible to price on what is delivered instead of hours.

Prove your AI delivery claims

Show where AI is used in the code, how mature the practice is, and whether it raised output or only activity.

Run a tighter delivery engine

Spot rework hotspots and lagging teams early, and fix the cause upstream before it reaches the client.

Your data stays yours.

You decide what is connected, who sees which view, and what leaves the room.

Read-only, cloud or on-premise

The platform reads the repositories in scope and never writes to them, in the cloud or on your own infrastructure.

No source code kept

P10Y states that no source code is stored and that data is deleted after processing.

Certified

P10Y states ISO 27001 and SOC 2 Type 2 certification and GDPR compliance.

You choose the views

Company, team or individual views. Individual views can be switched off.

Nothing shared without approval

No result goes to anyone, a client included, without your written approval.

Start with mock data

See the platform working on mock data before you connect a single repository.

From interest to a decision.

Four steps, each one a chance to stop. Nothing is connected to your code until the pilot, and the pilot is scoped with you first.

Product demo

A walk through the platform and a sample read-out, with your questions answered.

Mock-data sandbox

Provisioned by P10Y. Explore the dashboards yourself, with no repository connected.

Scoped paid pilot

The cohort, repositories, history, read-out, timing and terms are agreed with us before it starts.

Rollout decision

You decide on the evidence from the pilot: roll out, extend, or stop.

Find out what your engineering really delivers.

Tell us about your teams and what you need to prove. We will show you the platform and a sample read-out.

Book a 20-minute call or ask us about a benchmark