What is token adoption, and how do companies measure it?
Short answer
Token adoption is how far a team has made AI models and coding agents part of its normal work. Companies measure it by how many people use the tools each week, how often, and how much of the paid capacity gets used, then check outcomes such as merged work. Raw token totals are a weak measure and invite gaming.
On this page
What token adoption means
Token adoption is the degree to which the people on a team use AI models, and AI coding agents in particular, as a regular part of their work, measured by active use and by how much of the AI capacity the company pays for is put to work. The name comes from tokens, the units AI models read and write, which is how providers meter and limit use.
The term spread in 2025 and 2026 as companies moved from offering AI tools to expecting staff to use them. Some counted tokens, and several of those efforts were rolled back. The lesson is to keep two questions apart: are people using the tools, and is the use producing work.
Adoption is not volume
A team can burn a lot of tokens with low adoption (two people running agents all day while ten seats sit idle) or reach high adoption with modest volume. The two need different numbers.
| Adoption | Volume | |
|---|---|---|
| Question it answers | Who uses the tools, how often, on real work | How much the tools were used in total |
| Typical numbers | Weekly active people per seat, sessions per person, share of weekly limit used per seat | Tokens, requests, dollars at API rates |
| Good for | Finding idle seats and people who need help getting started | Budgets, cost caps, capacity planning |
| Fails when | Counted without outcomes (a session is not a shipped change) | Ranked per person: rewards spending, invites gaming |
What companies have done, 2025 to 2026
The public examples, with the date each source shows. Most were reported from internal memos, not published by the companies.
| Date | Company | What happened |
|---|---|---|
| Apr 2025 | Shopify | CEO memo: “Reflexive AI usage is now a baseline expectation at Shopify.” Per the report, AI competency joins performance reviews and hiring, and teams must justify why AI cannot do a task before asking for more resources.1 |
| Aug 2025 | Coinbase | CEO Brian Armstrong said engineers who had not onboarded to the company's AI coding tools within a week, and had no good reason, were fired. He called the approach heavy-handed.2 A later report put AI-written code at about a third of new code, with a 50% goal.3 |
| Nov 2025 | Meta | A memo from the head of people made AI-driven impact a “core expectation” from 2026. Individual AI usage metrics were left out of 2025 reviews.4 |
| Apr 2026 | Meta | An employee-built leaderboard called “Claudeonomics” showed the top 250 token users out of more than 85,000 employees. Total use passed 60 trillion tokens in 30 days, and the top user averaged 281 billion. It went down two days after the news broke.5 |
| By Jun 2026 | Amazon | Scrapped an internal leaderboard that ranked staff by AI usage after people gamed it, which sent compute costs up.6 |
| By Jun 2026 | Uber | Used its whole 2026 AI coding budget by April. Now caps employees at $1,500 a month per tool.6 |
| By Jun 2026 | Microsoft | Found some staff spending $500 to $2,000 a month on Claude Code tokens alone.6 |
| By Jun 2026 | Databricks | An engineering leader said the AI budget for engineers is still unlimited.6 |
The pattern: mandates in 2025, token-volume contests in early 2026, then caps and a shift to cost control by mid 2026. The Next Web called the turn “tokenminimizing.”6
Industry baseline numbers
Surveys give context, not targets. None measures coding agents specifically.
- Stack Overflow Developer Survey 2025: 84% of respondents use or plan to use AI tools, up from 76% the year before. About half of professional developers (50.6%) use them daily. More distrust the accuracy of AI tools (46%) than trust it (33%), and 66% name AI answers that are almost right, but not quite, as their top frustration.7
- DORA 2025 (nearly 5,000 respondents): AI adoption among software professionals reached 90%, a 14% increase, with a median of two hours a day working with AI. 30% trust it only a little or not at all.8
- GitHub Octoverse 2025: more than 180 million developers on GitHub, and 80% of new developers use Copilot in their first week.9
- Anthropic Economic Index, March 2026 (data from February 5 to 12, 2026): Claude Code “has grown to represent a large share of sampled traffic” on Anthropic's own API. No exact share is given.10
Metrics that work
For a team on Claude Code, these five numbers, read weekly, cover adoption without turning it into a contest. Each has its own guide.
| Metric | Formula | Read it as |
|---|---|---|
| Weekly active share | People with at least one agent session this week ÷ seats paid | Who has started. Low means onboarding, not laziness. |
| Sessions per active person | Agent sessions this week ÷ weekly active people | Whether use is habitual or occasional |
| Capacity used per seat | Share of the weekly limit used at the reset, averaged over 4 weeks | Whether you pay for capacity nobody uses |
| Agent wait time | Time agents sit waiting for a human answer or approval | Where people, not models, are the bottleneck |
| Outcomes | Merged pull requests, lead time, change failure rate | Whether the use produces work |
How to collect each number is in How do you measure Claude Code adoption across a team? The capacity number, with a worked dollar example, is in Are you using the Claude capacity you pay for? For setting it up from scratch, see How do you roll out Claude Code to an engineering team?, and for spend, How much should you budget for AI tokens per developer?
Backlash and gaming risks
- Gaming. Amazon dropped its ranking after people gamed it and compute costs went up.6 See Should you rank engineers by AI token use?
- Cost. Uber spent a year's AI coding budget in four months.6 Adoption goals need a budget and a cap next to them.
- Quality. With 46% of developers distrusting AI accuracy,7 more tokens can mean more review work. Count outcomes, not just use.
- Exposure. Meta's leaderboard came down after its data was shared outside the company.5 Per-person usage data is sensitive. Decide who can see it before you collect it.
Common questions
Is token adoption the same as token usage?
No. Token usage is a volume: how many tokens were read and written. Token adoption is about people: how many use AI tools regularly on real work, and how much of the capacity you pay for gets used. A team can have high usage and low adoption.
What is a good weekly active rate for AI coding agents?
There is no published benchmark for coding agents. DORA 2025 found 90% of software professionals use AI in some form, which is a broader measure. Set your own baseline during a pilot and track the trend.
Should AI use count in performance reviews?
Some companies do count it. Shopify made AI competency part of reviews in 2025, and Meta made AI-driven impact a core expectation from 2026 while leaving individual usage metrics out of 2025 reviews. Reviewing impact and outcomes is safer than reviewing token counts, which are easy to inflate.
Does Anthropic provide adoption metrics for Claude Code?
On Team and Enterprise plans, Anthropic's Claude Code analytics dashboard shows daily active users, sessions, lines of code accepted and pull requests with Claude Code. Check your admin console for what your plan includes.
Sources
- Internal memo: Shopify CEO declares AI 'non-optional'Digital Commerce 360, Apr 8, 2025Checked
- Coinbase CEO explains why he fired engineers who didn't try AI immediatelyTechCrunch, Aug 22, 2025Checked
- Coinbase CEO fired engineers who refused to use AITechSpot, Aug 2025Secondary report
- Meta to formally review employees' AI performance from 2026HR Grapevine, Nov 17, 2025Secondary report
- A Meta employee created a dashboard so coworkers can compete to be the company's No. 1 AI token userFortune, Apr 9, 2026Checked
- The tokenmaxxing era is over. Now companies are 'tokenminimizing'The Next Web, Jun 17, 2026Checked
- 2025 Developer Survey: AIStack Overflow, 2025Checked
- How are developers using AI? Inside our 2025 DORA reportGoogle, Sep 23, 2025Checked
- Octoverse: A new developer joins GitHub every secondGitHub, Oct 28, 2025, updated Feb 28, 2026Checked
- Anthropic Economic Index, March 2026 reportAnthropic, Mar 24, 2026Checked
- Customize your status lineClaude Code docs, undatedChecked
- Track team usage with analyticsClaude Code docs, undatedChecked