AI Landed On Top of the Work: What Linear's Whole-Workflow Telemetry Shows

Between January and June 2026, AI adoption inside Linear workspaces more than doubled in every function the company tracks. Product went from 12% to 34%. Engineering from 12% to 30%. Founders from 14% to 30%, Design from 6% to 22%, GTM from 5% to 18%, across a base of 127,000 paid users. The single largest jump belongs to CEOs at companies with more than 200 employees: 9% to 36%.

Here is the number that did not move: planning time. Across every function, time spent planning shifted by zero to one minute per user per month. AI adoption doubled everywhere, and the existing work stayed exactly where it was. Nothing got handed off. The new activity settled on top.

That combination, explosive adoption with zero displacement, is the finding worth an operating review. And it comes with a twist: the vendor that produced the telemetry now calls the industry’s favorite value metric, token spend, “a relic.”

The First Whole-Workflow Baseline

This is Linear’s first whole-workflow telemetry report, and its scope is what sets it apart from single-function measurement. Because the product hosts the workflow from issue creation to PR attachment, across product, engineering, design, and go-to-market, its telemetry captures how AI use flows through an entire delivery pipeline rather than one seat in it.

The caveats are stated in the report body and they matter. The data covers paid workspaces only. PR counts measure PRs opened, not merged. The telemetry is blind to AI use that happens outside Linear. And Linear itself describes what it measures as motion rather than value. Carry all four caveats into any decision you make with these numbers. What survives them is still a whole-workflow, cross-function baseline.

Two findings anchor it. First, AI is now creating roughly 2.4 million issues per week in Linear, near parity with the roughly 2.5 million created by humans, up from approximately zero in June 2024. Machine-originated work items went from a rounding error to half the intake in two years. Second, authorship is spreading past engineering: the share of product people attaching PRs went from 3% to 10%, designers from 1% to 8%, GTM from 1% to 3%. Linear states these as floors, since attribution undercounts.

The Agent Delta

The sharpest number in the report is a fixed-cohort comparison of 6,887 teams between June 2024 and June 2026. Teams that adopted coding agents went from 21 to 65 PRs per week. Teams that did not went from 8 to 10.

Tripled output against near-flat output, in the same tool, over the same window. Whatever discount you apply for opened-not-merged and for motion-not-value, a delta that size inside one instrument is the closest thing to a procurement argument the market has produced. We wrote earlier about why measuring adoption through surveillance backfires; this cohort design shows the alternative. Compare teams to themselves over time, and the instrument stops being a leaderboard and starts being evidence.

But the delta also sharpens the question the rest of the report forces. If agent-equipped teams open three times the PRs, someone reviews them, triages the issues they spawn, and comments on the work. Who?

Where the Overflow Went

Linear’s time-accounting answers it. Planning time moved zero to one minute per user per month in every function. Founders are the exception, in the wrong direction: founder triage time rose 17 minutes per month and founder commenting rose 26. About 43 extra minutes a month of coordination work, absorbed by the people at the top of the org chart.

This is the Jevons pattern applied to work itself. Cheaper production did not reduce the total amount of work in the system. It increased it, and the increase pooled as triage and review. The report’s telemetry shows AI adding a new stratum of activity above the existing ones, with the coordination overflow flowing uphill to whoever owns the backlog. At a small startup that is often the founder personally. At larger companies the same overflow plausibly lands on engineering managers and leads, though Linear’s role taxonomy resolves the absorption clearly only for founders.

Forty-three minutes a month is small in isolation. The direction is what matters. Every adoption curve in the report is still climbing, machine-created issues are at parity with human-created ones and rising, and the absorption cost scales with the volume of generated work, not with the number of people available to absorb it.

The Meter Disowns Itself

Which brings us to the metric most organizations are using to discuss all of this. Linear’s own report calls token spend as a proxy for value “a relic.” That claim deserves attention precisely because of who wrote it: a vendor whose telemetry could easily have been packaged as a token dashboard.

John Cutler’s essay on measurement proxies, published the same month, explains the seduction. Tokens, he writes, are “a beautiful meter and invoice, which can create a much stronger illusion that you have a denominator suitable for ROI.” The meter is precise, the invoice is real, and neither tells you whether anything of value happened. His argument is conceptual rather than empirical, and the dollar scenarios in his piece are illustrative hypotheticals, so treat them as frames, not findings. The frame is enough. A precise number attached to spend will always outcompete a vague number attached to outcomes, and it will be wrong.

We made the case against usage proxies when the token leaderboard died, and again when tokenmaxxing became a performance metric. What is new in August 2026 is the source. When the company whose product sits on top of the workflow says the token proxy is a relic, and simultaneously publishes what a workflow-level alternative looks like, the excuse for token-based ROI reporting expires.

Build an Absorption Ledger This Month

The governance question Linear’s data puts on the table: where did the absorbed work go in your organization? Answer it with your own telemetry, this month.

Pull the cohort comparison. Take your agent-equipped teams and your non-equipped teams and chart PRs opened per week for both, over at least a year if you have the history. If your delta looks nothing like 21-to-65 versus 8-to-10, find out what is different before renewing any agent contract.

Tag machine-originated work. Linear can report AI-created issues separately from human-created ones. If your tracker cannot distinguish agent-created items from human-created ones, you cannot see the intake shift that is already underway. Fix the tagging first.

Measure absorption, not just output. Instrument triage and review time per role, the way Linear instrumented founder minutes. The people absorbing coordination overflow rarely report it, because each increment is small. The ledger makes the pooling visible before it becomes attrition.

Retire the token slide. In the next AI review, replace token spend with two workflow numbers: the cohort PR delta and the absorption minutes by role. If the vendor selling the meter calls it a relic, your board deck should not be the last place it survives.

AI landed on top of the work. The organizations that thrive with it will be the ones that can see, in their own numbers, who is carrying the extra weight.

This analysis synthesizes How teams build: AI usage patterns in software teams (Linear, Tim Qi, Head of Data, August 2026) and TBM 437: Tokens, Hours, Points, and Other Curious Proxies (The Beautiful Mess, John Cutler, August 2026).

Victorino Group helps engineering organizations replace token-spend reporting with workflow-level AI measurement and absorption accounting. Let’s talk.