For engineering leaders
Defend your tooling budget with proof, not opinions
We count what your team pays for and does not use, across Copilot, Cursor, Claude, OpenAI and the rest at once. The budget talk starts with a number instead of an argument.
Free Snapshot. No card. Nothing installs. Nothing writes to your systems.
Finance is going to ask what the AI tools are doing. In most companies the honest answer is: some of it works, some of it is waste, and nobody can tell you which. That answer loses budget.
We count the part that can be counted. Seats billed and never opened — a seat is a paid license for one person, and an idle seat is one nobody has used lately. Two coding assistants doing one job. Token spend with no team attached. A token is the unit AI vendors bill for, roughly a few characters of text.
It comes straight out of the vendor consoles and your billing, and it runs 10 to 15% of seat and subscription spendacross the public benchmarks and audits we've run.
The tooling budget you defend with proof is the tooling budget you get to keep.
Four questions you get asked. Four answers you can show.
Not a dashboard. Four answers.
- “What are we getting for the AI spend?” — An AI cost report: one page that shows what your AI tools cost and what you get back. Every figure says how we know it.
- “Do we still need both Cursor and Copilot?” — Overlap counted per team, not guessed.
- “Is the rollout you pushed for actually being used?” — Seats used in the last 30 days, by team, against a benchmark.
- “Is AI working?” — A range, with the cost of going faster beside it. This is the question where most tools start lying.
No one gets ranked. That is built in, not a setting.
We only ever report on teams of eight people or more. There are no individual developer rankings. Not by default, not as an upgrade, not ever.
We read metadata — counts, dates and names, never the contents. Seat activity, PR timing, token volume by team. Never your code, diffs, prompts or messages. And the audit never writes to your systems, because an auditor that can change the books is not an auditor.
That is the point, not the caveat. It cannot become a productivity tracker.
Renew, standardize, or keep both
The duplicate-assistant question usually goes to whoever argues hardest in the room. It does not have to. We show, per team, who uses Cursor, who uses Copilot, who uses both in the same week, and who has opened neither since their first day.
Sometimes the honest answer is to keep both. Inline completion and agent work are different jobs. The point is not to cut a tool, but to stop paying for the copy nobody opens.
Why you need your own number, not a vendor's
DX tracked more than 400 companies from November 2024 to February 2026. The median gain in PRs shipped from AI coding tools was 7.76%. The spread is the interesting part.
That spread is the argument. A vendor citing “55% faster” is quoting someone else's middle number. Nobody knows where you sit until you count it.
So we will not hand you one number. We show a band, 20% either side, labelled estimated — worked out from assumptions you can see and change. Always a range, never a single number.
It never appears without the cost of going faster beside it: rework within 21 days, bugs per PR, unreviewed merges, incidents per PR. Rework is work that has to be redone. A speed number without its cost is propaganda.
Bring the band. It survives the second question.
What it costs, in your units
Observe is $16 per engineer per month, per engineer covered, never per employee. At 200 engineers, that is $3,200 a month.
DX puts total AI cost per engineer at $200 to $600 a month for teams mixing inline and agent tools. Call it $300. Your 200-engineer company spends $60,000 a month on AI tooling. We cost 5.3% of that bill to tell you what the bill bought.
Break-even is 107 unused ChatGPT Business seats at $30, or 80 idle Cursor Teams seats at $40. If you cannot find that much, you run a tight shop. Stay on the free Snapshot: a real product, not a trial, and it costs nothing.
Compare Snapshot, Observe and Control
What you walk away with
One page you can forward without editing. The spend you can get back, by vendor and team. Seat use against a benchmark. Where two tools overlap. A productivity band with the cost of going faster attached. Every figure shows its formula and its proof level: counted, compared or estimated.
Free account, no card. Connect GitHub and one vendor console, and you have a number the same day.
Questions engineering leaders ask
Can this be used to rank or judge individual engineers?
No. There is no individual view to unlock, and a team smaller than eight rolls up. This is not a policy you have to trust us on. There is nothing to turn on.
What do you actually read?
Seat activity from vendor consoles, PR and commit timing from GitHub, token volume by team, transactions from billing. No source code, no diffs, no prompts, no message contents.
What has to be installed?
Nothing. No agent, no IDE extension, no CI step. Connectors are OAuth or API key, and every one is read-only. Native today: GitHub, Anthropic, OpenAI, Cursor, Ramp, Okta and Google Workspace. CSV import covers the rest.
How long before I have something I can show finance?
Recovery findings land in days, straight out of systems of record. The productivity band needs a quarter of history to be worth anything, and we would rather say so than guess.
Walk into the budget conversation with the numbers already counted
The idle seats are already on the invoice. This is the audit that turns them into a number you can defend, team by team, with the formula attached.
Snapshot is free and needs no card. Observe is $16 per engineer per month. Control is $40 per engineer per month.