The second half of 2025 marks a turning point for engineering analytics. As generative AI assistants have grown ubiquitous, the traditional signals that reliably helped leaders understand productivity -- Commit Count, Diff Delta, PRs Opened -- have begun to break down. Assistants are inflating code metrics across the board, while applying gravity toward deep-seeded tech debt via quiet code duplication.

Given these challenges, GitClear's latest round of updates takes aim at the key question: Which AI agents are delivering the most value right now, and which teams are measurably getting ahead thanks to their use of AI?

open_in_new

Q3 introduces the most comprehensive upgrade in our history: granular, AI usage visibility, cohort-based analytics, context-rich metrics, and a fully consolidated Developer Analytics Encyclopedia. Together, these tools form a framework we call Contextual Intelligence—the ability to not only see that code changed, but to understand why, how, and by whom.

If Q2 was about uncovering AI’s growing influence, Q3 is about making that influence measurable, comparable, and actionable.

linkHeadline Upgrade: Quantify Long-term Impacts of Heterogeneous AI Use (Copilot, Cursor, Claude Code)

A defining challenge of 2025 has been code line inflation. Earlier GitClear research revealed a dramatic surge in duplicated code, churn, and “easy wins” generated by LLM coding assistants. I spoke about this at LeadDev New York, and now we’re delivering the tooling to quantify the phenomenon at scale.

open_in_new

A slide from LeadDev New York, showing how developers have crossed to the dark side of "moving code" to "duplicating it"

Now, we have not only the stats to go beyond these generalities. We also have a brief 3 minute video to introduce GitClear's AI usage stats:

linkAI Usage Data: Piercing the Black Box of AI Attribution

The foundation of the Q3/Q4 strategy is the ingestion and processing of AI Usage Data, like "All Suggestion Count":

open_in_new

Every snippet of AI involvement is now captured:

linkAI Acceptance Rates: What Has Your Model Done for You Lately?

In a full-circle moment in the AI wars, it was recently reported by WSJ and others that OpenAI declared a "Code Red" in response to Google...about one year after Google had declared a "Code Red" in response to OpenAI. 🎭 These billion-dollar companies are fighting tooth-and-nail to keep their edge. How can developers and execs ensure that they are leveraging the best option of this month?

GitClear's API integrations now ingest telemetry regarding the intensity & quality of every team's AI interactions. This goes beyond binary "enabled/disabled" flags to track granular details about how developers from various teams are liking the LLMs they choose to interact with:

open_in_new

The AI Usage charts will relate per-team (and for Cursor/Claude, per-developer) truths like

This gives engineering leaders a grounded way to evaluate model performance—without relying on anecdotes or marketing claims.

linkAI Cohort Stats: Tease out the True Impact of AI Use

The most significant breakthrough of Q3 is GitClear's AI Cohort Stats framework. This feature acknowledges a critical reality: AI adoption is non-uniform. In any given engineering organization, distinct behaviors emerge as ad hoc A/B tests.

You can ignore the A/B tests happening around you, or you can leverage them into wisdom that can be applied across teams. Recognizing the truth of who has played the role of "Power Users," "Regular Users," and "Skeptic/Non-Users" lets you tie behaviors to KPIs.

Treating the team as a monolith dilutes the signal of "which metrics are impacted by AI?" Segmenting them reveals the truth.

open_in_new

"Commit Count" is a common, if easily gamed, measurement used among "productivity gauging" metrics

Cohort levels are broken into five strata, explained on the Cohorts help page. GitClear dynamically segments developers per-week, based on their interactions with AI tools (Copilot, Cursor, Claude).

Note that, for each cohort, the value shown on the cohort graph is the average value per developer among developers in a particular cohort. This distinction ensures that the cohort with the highest value isn't de fact the cohort with the highest amount of "Diff Delta," "PRs Opened," etc. For a Cohort curve to soar, the developers who inhabit that cohort must each prove their mettle.

linkCohort Metrics Available

Some representative customer graphs, showing real world data. It has been comforting to see our initial customers who deploy AI Cohorts have observed results in line with what one might naively expect (more AI use = more productivity, but also more AI sludge to process).

open_in_new Metric to minimize: How much time is being soaked up reviewing code generated by each cohort?

open_in_new Metric to maximize: How many pull requests are being opened by developers in each AI cohort?

There are a wide and growing variety of AI Cohort stats available, see the entire set of them (Diff Delta by Cohort, Story Points Completed by Cohort, Copy/Paste per Cohort, etc) on our dedicated AI Cohorts help page.

linkComprehensive Sprint Stats for Jira users

For teams that plan with Story Points, GitClear now offers an entire tab dedicated to comparing teams' performance (teams plural because this feature most often applies to Enterprise customers) at planning and executing the goals of the company.

All of the current sprint stats are explained on our new help page dedicated to configuring and using Sprint Stats.

open_in_new

What percent of the team's issues worked were planned at the beginning of the sprint?

The Sprint Stats tab is ready to surface graphs that answer questions like:

Enterprise teams using Starred Reports Outlier View, will get an instant sense for which teams are the "role models" that can be used as inspiration and examples to inspire those teams that have struggled to execute on the plans they set.

linkArea Chart Improvements: Tooltips, Resizing, and Much More

A major focus this quarter was improving the readability and usefulness of GitClear’s charts.

open_in_new

Specific chart upgrades include:

linkMedian, 10th percentile and 90th percentile comparison lines: Now live

Most charts now support the means to show the "median," "10th percentile" and "90th percentile" performance across all teams in your entity. This is crucial information for understanding how far a team's results are from the median team among a large set, as viewed in Starred Reports Outlier View.

open_in_new

This is a favorite feature among our enterprise customers.

linkMore Q3 Upgrades: Documentation, Release Rules, Developer Velocity & Diff Viewer Enhancements

A grab bag of upgrades originating from customer requests.

linkContextualized Per-Committer Velocity

Now, when you hover on a committer's avatar on the left side of the CAB, you'll get the full rundown of how their recent velocity compares to historical velocity from the past year.

open_in_new

As always, use "Record time off" when a developer is taking PTO in order to avoid having them designated as "behind their average velocity."

linkUnified Metric Documentation

While the AI tools provide the mechanism for measurement, the new Encyclopedia of Developer Analytics, provides the meaning. This initiative represents an effort to standardize the taxonomy of developer analytics, addressing the "Consensus Gap" that has historically plagued our industry, and it's reports (even DORA).

open_in_new

A full, browsable rundown of available metrics from GitClear now resides in one place

Perhaps the most welcome upgrade for consumers of the GitClear API: each section of metrics now includes a table to summarizes each available metrics. This table includes how we define the date for each metric.

We also added a detailed reference explaining why GitClear often measures analytics in working hours, not raw timeline hours.

Side bonus: Did you happen to notice, as your read this blog, how you now have a scrolling sidebar to contextualize where you are within the page, and how you have a ubiquitous path straight to another section?

linkNew Release Rules: Github Actions and Merged Pull Requests

Two new ways to define when a release is occurring in your repos. These rules are particularly useful for capturing historical releases, which are difficult to reliably detect with historical branch-based rules.

One of the new options is a rule to allow a release to be designated any time that a pull request is merged to a specified branch:

open_in_new

Release defined by branch that release is merged into

There's also a new option to capture a release whenever a Github Action successfully completes:

open_in_new

Define a release to occur when a Github Action successfully completes

We can also integrate the new "Github Action" rule to apply to the "Pipeline Actions" offered by Bitbucket, Gitlab and Azure Devops, if paying customers write to us to request it.

Note that, if a single commit qualifies as a release by multiple rules, we will only designate the commit as a single release, as long as all of the rules converge to that single commit (e.g., if the commit pushes a tag, merges a PR, and has a particular deploy message in it).

linkDiff Viewer Improvements

For those using GitClear for pull requests (like us! 🐶), substantive progress was completed during Q3.

linkCopy code from "Unified" or "Split" view

linkAllow a diff file to be marked "Reviewed" from top of file

For those that are accustomed to the Github convention of marking a file as "Reviewed" using a checkbox atop the file, guess what:

open_in_new

Marking a file as "Viewed" will close, it while marking its contents as "Reviewed" unless/until the developer makes additional changes to the file during a subsequent PR review pass. As always, you can also automatically mark files as "Reviewed" by just scrolling through them and not clicking anything. 😇

linkPull Request Diff Viewer Includes Author Description

No need to visit the "Discussion" tab to see what changes a pull request seeks to implement. That is now present atop the main "Code" tab used to review the PR's contents:

open_in_new

This blends especially nicely with GitClear's updated Slack pull request notifications, letting you can jump straight from a Slack message about "This PR has been sitting around longer than your team prefers" to clicking the adjacent link to "Review the PR." Most every pull request goal now includes a link to review the pull request when the notification is sent via Slack.

linkAPI Upgrades Across the Board

Now that GitKraken Insights is utilizing GitClear's full breadth of stats, we've been busy with more upgrades than could be fully enumerated in one blog post. A taste of recent improvements include:

linkGithub Team sync will reuse an existing team when available

open_in_new

When you initiate an import of teams from Github, we will now only create a team if the team does not already exist on GitClear.

linkManually refresh list of importable repos

When choosing which repos to import, you can now manually refresh the list of importable repos:

open_in_new

This is especially useful if you decide to add new repos to your account while you are in the process of choosing which repos to import to GitClear.

linkContributors Page: Canonical Source of Truth for Seated Developers

The "Settings" => "Contributors" page now presents a wealth of information to answer common customer questions like "How many seats do we have?" and "How many seats will we have after our next renew?"

open_in_new

As always, you can manually adjust which developers are processed by clicking on the "Gear" icon at the right of any contributor line. We have also improved our bot detection handling to recognize Cursor and Copilot as types of bots.

linkOdds & Ends

Smaller improvements worthy of a mention.

link🔮 What's next? Help us pick

This quarter was about clarity -- illuminating how developers interact with AI, exposing patterns that were previously invisible, and restoring trust in engineering analytics.

But clarity is only the beginning.

In the coming quarters, we plan to leverage our AI late-mover advantage to move from analysis to action. As Amplenote is doing with “task-strategizing AI,” GitClear will begin helping teams execute more effectively — from identifying unstable code to proactively recognizing which charts looks amiss, which tickets that seem off-schedule, or model usage strategies.

The marriage of empirical developer analytics and intelligent AI assistance is the future toward which we’re accelerating.

As always, your feedback guides our roadmap! 🗺️

Drop us a line! We want to build this super-dashboard together.

linkAppendix: Feature & Upgrade Matrix (Q3/Q4 2025)

The following table summarizes the key features, architectural updates, and research initiatives launched or significantly updated between July 7, 2025, and the end of the year.