GitHub Insights vs Deveval: Why Activity Metrics Miss AI-Era Output

GitHub Insights is where most engineering leaders start. It is free, it is already there, and it counts the things that used to matter. PRs opened. Commits pushed. Review latency. Contributor activity.

That worked in 2018. In 2026, it actively misleads.

Not because GitHub is wrong about the data. GitHub is right about the data. The problem is that the data no longer maps to the thing you actually care about — whether your engineering team is shipping value.

This post is about what GitHub Insights shows you, what it cannot show you, and why every engineering leader running an AI-augmented team needs a second layer sitting on top.

What GitHub Insights Actually Measures

Open the Insights tab on any repo and you get a specific set of things. Pulse gives you a snapshot of recent activity — PRs opened, closed, merged. Contributors shows commit volume per person over time. Code frequency shows additions and deletions week by week. Community Standards checks whether you have a README and a LICENSE.

On the enterprise side, GitHub Insights (the product, formerly Pull Panda + expanded reporting) adds pull request metrics: time to first response, review turnaround, coding time before a PR opens, and code review load distribution. Useful signals about the review pipeline.

All of these measure one thing: flow. How fast work moves through your review process. How much activity is happening. How evenly it is distributed.

Every single one is a signal about motion, not outcome.

Why Motion Metrics Used to Be a Decent Proxy

For most of engineering history, motion correlated with output. If a team was pushing more PRs, they were probably shipping more features. If commits were up and cycle time was down, the team was probably in a good rhythm. Activity was expensive, so more activity meant more work being done.

That correlation is broken.

The moment Cursor, Claude Code, and Copilot entered the average engineering team, the cost of generating code went to near-zero. An engineer using AI can open more PRs, push more commits, and touch more lines of code in a week than the same engineer could in a month two years ago.

GitHub Insights sees all of that motion. It cannot see whether any of it turned into shipped business value.

What GitHub Insights Cannot See

Specifically, four things:

1. Whether the code that got merged actually solved a problem. A PR that adds a login flow and a PR that renames a variable both count as +1 merged PR. Pulse does not know the difference.

2. Whether AI-generated code got reviewed the same way human-generated code does. Review coverage percentages look identical. Review depth does not.

3. Whether an engineer is generating volume that never merges. Some of the highest-motion engineers in the AI era are also the lowest-output. They open PRs that get rewritten, abandoned, or reverted. GitHub counts the activity. The activity did not ship.

4. Whether the team is actually leveraging AI, or just performing it. Two engineers can each merge 20 PRs a week. One is using AI to compound their impact. The other is generating suggestions their reviewers spend hours cleaning up. Insights sees them as equal.

How the Developer Impact Index Measures the Gap

The Developer Impact Index (DII) is built to answer the question GitHub Insights cannot: Is any of this motion turning into shipped value?

It is a 0–100 score composed of four weighted dimensions, all derived from the same GitHub data GitHub Insights uses. The difference is in the interpretation.

• Code Quality (CQ) — 30 points. Architecture, maintainability, defect risk, review signal. Not just "was it reviewed" — was it changed materially in review, and did it come back defective.

• Delivery Capability (DC) — 30 points. Business logic shipped. Features merged that made it to production. This is the biggest weight because it is the thing that matters most and gets measured least.

• Engineering Efficiency (EE) — 25 points. PR cycle time, first-pass review rate, productive AI adoption. Not adoption of AI in general — adoption of AI in a way that produces mergeable work.

• Contribution Breadth (CB) — 15 points. Cross-repo ownership, review help, team leverage. Are you a solo engineer with high output, or are you making the people around you better?

Each engineer gets a DII score and a red/yellow/green status. Each team gets an aggregate. Reports export to PDF in one click.

Side-by-Side: Same Team, Two Lenses

Consider a 12-engineer team that just rolled out Cursor company-wide. Here is what each tool tells you after 90 days.

Question

GitHub Insights answer

DII answer

Are we shipping more?

PR volume up 40%. Commits up 65%.

Delivery Capability up 8 points. Volume up, but merged-to-production ratio held flat.

Is AI helping every engineer?

Contributor graphs all show activity increases.

4 engineers gained 10+ points on Engineering Efficiency. 3 engineers dropped. First-pass review rate diverged.

Is code quality holding?

Review turnaround time unchanged.

Code Quality dropped 4 points on average. Defect signal up. Review depth (comments per PR touching real substance) down.

What do I tell the board?

"Activity is up." (True and unhelpful.)

"AI is compounding for 4 of our 12 engineers. Here is the plan for the other 8."

Same team. Same GitHub data. Two completely different conversations.

Why Most Teams Need Both

This is not an argument to replace GitHub Insights. GitHub Insights is the right tool for the questions it answers: is the review pipeline healthy, are reviewers overloaded, is there a hotspot repo. Those are real questions.

DII is the right tool for a different set of questions: is AI adoption converting to shipped value, which engineers are compounding, what does a board-ready answer look like.

You want both lenses. The pipeline one for operational decisions. The output one for coaching, hiring, and the AI ROI conversation you are going to have with your board this quarter whether you are ready or not.

Get Your First DII in 5 Minutes

Connect your GitHub org. No dashboards to configure, no agent to install, no meetings on the calendar. You get a DII score for every engineer, a team aggregate, and a one-click PDF export in about five minutes.

There is a free plan you can start on today — no trial clock, no credit card. If your gut says AI is helping and your data cannot say why, that is a good place to start.

Start your free DII evaluation →