Is GitHub Copilot worth it for a
100–1,000 person engineering team?

The money, named

What you pay. Copilot Business is $19 per granted seat per month and Copilot Enterprise is $39, both published by GitHub. Multiply by your headcount and twelve months and the whole band looks like this:

What it has to beat. The break-even is the same at every headcount in this band, which is the single most useful fact on this page:

$19 / seat / month x 12                    =  $228 per engineer per year
$228 / $90 per hour ( = $180,000 / 2,000)  =  2.5 hours per engineer per year

Copilot pays for itself if it saves each engineer about 38 minutes per quarter. That is a very low bar, and no credible study — including the ones most hostile to AI coding tools — claims the effect is zero for everyone all of the time. On licence cost alone this is not a close call at 100 seats or at 1,000, and any vendor-comparison spreadsheet that stops at the $19 is measuring the wrong thing.

What it can cost you.Here is the line nobody puts in the quote. Assume 20% of engineering time currently goes to rework — fixing defects, reverting, re-doing work that shipped wrong. (This is the one number you should replace with your own; everything below scales linearly off it.)

1 engineer x 2,000 hours x 20% baseline rework  =  400 rework hours / year
500 engineers                                  =  200,000 rework hours / year

Now apply the measured effect of AI-assisted authorship on rework, at three published severities, costed at $90/hour:

So at 500 engineers: the licence is $114,000 and the exposure is $900,000 to $7.4m. Central estimate ~$2.7m, which is 23.7x the licence. At 1,000 engineers double every figure; at 100, divide by five. The ratio never moves, because both sides scale with headcount — and that is the point. You are approving a rounding-error line item that moves a very large number. Whether it moves in your favour is decided by the controls below, not by the purchase.

The reason, at this size

1. The most-quoted argument against buying has been withdrawn by the people who made it.

If somebody forwards you the METR study — “a randomised trial found developers were 19% slowerwith AI” — that number is a snapshot of February–June 2025 tooling, and METR themselves published a follow-up in February 2026 superseding it. Their new estimates point the other way: −18% time (a speedup) for returning developers, −4% for newly recruited ones. Both confidence intervals cross zero (−38% to +9%, and −15% to +9%), so the honest reading is inconclusive on magnitude, no longer evidence of slowdown. METR also flagged their own 2026 data as biased, because developers increasingly refuse to participate in a study that makes them work without AI.

Anyone still quoting the 19% at you in 2026 is quoting a retracted snapshot. Don’t let it block the purchase — and don’t let anyone use it to justify a six-month evaluation across an org this size, which costs more in delay than the licence costs in cash.

2. The argument for it is also weaker than the vendor page implies.

GitHub’s headline — 55.8% faster — comes from a 2023 controlled experiment where 95 developers wrote an HTTP server in JavaScript from scratch (95% CI [21%, 89%], p=0.0017). Greenfield, single-file, no existing codebase, no review, no on-call. Nobody in a 500-person org does that. They work in a codebase with years of history, a required review, and a change-approval process.

The number that describes your Tuesday is GitClear’s January 2026 finding: heavy AI users, measured against their own past selves, gained about 25% velocity. Real, meaningful, and roughly a quarter the size of the marketing figure. Budget on 25%, not 55%, and the purchase still clears the bar by an order of magnitude.

3. The failure mode is specific, measurable, and it compounds with headcount.

GitClear analysed 623 million code changes from 2023–2026 (AI-assisted commits are now about a quarter of all commits). Every maintainability signal moved the wrong way:

• Code block duplication +81%
• Within-commit copy/paste +41%— copy/paste now exceeds refactoring by ~5x, reversing a 2x preference the other way in 2022
• Refactoring line moves −70%
• Legacy maintenance (touching code older than 12 months) −74%
• Error-masking constructs — catch blocks and safe-navigation operators that swallow a failure without diagnosing it — +47%
• Function connectivity −35%, and two-week churn +15%

Churn is the expensive one; error-masking is the dangerous one. If you run Datadog or anything like it, code that silently catches errors instead of surfacing them doesn’t page anyone. It degrades quietly and arrives as a support ticket rather than an alert. A 47% increase in error-masking against a monitoring stack tuned for thrown exceptions is how you acquire a class of bug your dashboards are structurally blind to — and at 500 engineers you acquire it in five hundred places at once.

Google’s DORA data agrees on direction: in 2024, every 25% increase in AI adoption came with a 7.2% reduction in delivery stability(and −1.5% throughput). DORA’s 2025 report found throughput had turned positive as teams learned — but the stability penalty did not go away, at 90% adoption with 30% of developers reporting little or no trust in AI output. Two years running, the same stubborn finding: AI makes you faster at producing changes and worse at landing them safely. Large orgs feel that as incidents, not as vibes.

4. Buy Business, not Pro — and not Enterprise for everyone.

Copilot Pro is $10 per user per month and would look like a saving of $114,000 a year across 1,000 people. Don’t. Business ($19) adds organisation-wide policy management, content exclusion, audit logs, centrally-managed seats and — the one that matters — IP indemnity, which GitHub extends to unmodified suggestions only when the public-code filter is enabled. Pro, Pro+ and Max carry no indemnity at any price, including the $100/month Max tier. At this size you will be sent a security questionnaire by every enterprise buyer for the next three years; pay the $9.

Enterprise at $39 is the other trap, in the other direction. It requires GitHub Enterprise Cloud, so on GitHub Team it is a platform migration with its own per-user licence cost, a different admin model and its own project plan — not a Copilot upgrade. GitHub’s documentation says enterprise owners can assign Copilot Enterprise or Business per organisation, or Business directly to users and teams, so the middle path is supported and it is where most orgs this size should land:

500 engineers, all Enterprise      500 x $468  =  $234,000 / year
50 platform on Enterprise           50 x $468  =   $23,400
450 everyone else on Business      450 x $228  =  $102,600
                                   mixed total =  $126,000 / year  (saves $108,000)

Before you spend the difference, write down which enterprise-only capability you are buying and who asked for it. “Priority access to new models” is a real benefit and a poor six-figure justification; a documented compliance requirement is a good one.

5. Granted seats and credits are where the money actually leaks.

Copilot Business bills per granted seat, not per active user. At 100 seats that is a rounding error; at 1,000 it is a budget line. Every 100 granted seats that nobody uses — leavers, finished contractors, the people who never opened it — is $22,800 a year. A 10% idle rate on a 1,000-seat org is exactly that. A fifteen-minute monthly seat audit is usually the largest single saving available on this purchase, and it needs no negotiation and no tool change.

The second leak is usage. GitHub moved Copilot to usage-based billing on top of the base plans: each plan includes an allowance of GitHub AI Credits and consumption past the allowance is billed as usage. Base prices did not change — Business stays $19, Enterprise stays $39 — and code completions and next-edit suggestions are explicitly excluded from credit consumption; the pool covers chat, agents and CLI. We are deliberately not quoting an allowance number: GitHub’s own announcement post cites different figures in different sections. Open your billing settings, read the current allowance and consumption, and set a spend limit with an alert before you roll out to hundreds of people.

6. Kill the double-pay before you roll out, not after.

Orgs this size almost always have a pocket of engineers already paying for another assistant. If 80 of your 500 are on Cursor, rolling Copilot to everyone means those 80 seats carry two paid assistants doing substantially the same job: 80 × $228 = $18,240 a yearof duplicate Copilot spend, or $30,720 a year counted on the Cursor side at its published $32 per seat per month annual price. Either exclude them from the Copilot rollout or sunset the other tool there after a 30-day preference check. It is small money next to the rework figure — and it is exactly the kind of line that quietly becomes six of them by next year.

7. Two procurement facts that change your rollout date.

As of 22 April 2026, GitHub paused new self-serve sign-ups for Copilot Business for organisations on GitHub Free and GitHub Team plans. If that is you, you cannot put this on a card this afternoon — the route is GitHub sales or GitHub Enterprise Cloud, and your “decide by Friday” timeline is now a multi-week timeline. Check which plan you are on before you promise a date.

Copilot is not currently available for GitHub Enterprise Server. If your source of truth is self-hosted, this purchase is not available to you at any tier, and no amount of budget fixes it.

What to do instead of just buying it

Three controls. Together they cost roughly one engineer-week to set up at any headcount in this band. They are what separates the $900,000 outcome from the $7.4m one.

1. Turn on the public-code filter before you assign a single seat.

Org admin setting. It suppresses suggestions matching public code at 65 lexemes or more (~150 characters), and it is the precondition for the IP indemnity you are paying Business rates to get. Costs nothing. Buying Business and leaving this off is the single most common way to pay for indemnity and not have it.

2. Put a meter on duplication and two-week churn before rollout, not after.

Take a 60-day baseline of duplicated-block rate and the share of code rewritten within two weeks. Re-measure at 90 days. If churn is up more than 10 points you have bought the expensive outcome and you need to intervene — but you will only know that if you took the baseline first. Do not skip this because it delays the rollout by two weeks; at 500 engineers the baseline is the entire difference between managing this tool and hoping about it. Roll out in waves by team so the baseline has a control group.

3. Cap PR size and rehearse rollback.

DORA’s consistent finding across 2024 and 2025 is that the instability comes from AI generating large, hard-to-review batches — and that small batches plus frequent, practised use of revert/rollback is the documented moderator that turns AI adoption net-positive. Enforce a diff-size threshold in CI that flags oversized PRs for splitting, and make sure reverting a bad deploy is a rehearsed one-command operation rather than an incident-time improvisation.

Add one review norm to your engineering handbook: reviewers must reject catch blocks and safe-navigation operators that suppress an error without handling it. That is the +47% signal, it is invisible to linters, and it is cheap to stop at review and expensive to find in production.

And instrument three numbers on the day you turn it on: median cycle time from first commit to merge, the share of merged pull requests reverted or hotfixed within seven days, and seat utilisation over the last thirty days. The first two tell you whether it works, the third pays for the audit that funds it, and all three together make next year’s renewal a chart rather than an opinion.

What would change our mind

• Your engineers are mostly senior and work in one large, tightly-coupled monolith. That is closest to METR’s population, where the evidence for net speedup is weakest and the review burden is highest. The verdict downgrades to a 90-day paid pilot on two teams before an org-wide rollout.
• You cannot instrument churn and duplication. Without control #2 you are buying blind and the range above is not a range, it is a coin flip. Pilot on 50 seats instead of rolling to 500.
• Seat utilisation shows a third of granted seats unused after 90 days. Then the answer is not a different tier, it is fewer seats — and a conversation about why, before the renewal.
• GitHub moves code completions from unlimited to metered credits at the Business tier. Today completions and next-edit suggestions are explicitly excluded from AI-credit consumption; the pool covers chat, agents and CLI. If that changes, your fixed line becomes a variable bill and every break-even above has to be re-run.
• Business rises above roughly $30 per seat.The break-even moves from 2.5 to about 4 hours per engineer per year. Still likely worth it — but it stops being the non-decision it is at $19.

What we checked

Every figure above came from a page we opened, so you can check our work before you forward this to finance.

• GitHub Docs — Plans for GitHub Copilot — Business $19, Enterprise $39, Pro $10, Pro+ $39, Max $100 per granted seat per month; tier requirements; the Enterprise Server exclusion; the 22 April 2026 self-serve pause notice. Read 16 August 2026.
• GitHub — Copilot plans and pricing — IP indemnity conditional on the public-code filter; the 65-lexeme (~150 character) duplication threshold; completions and next-edit suggestions excluded from AI-credit metering. Read 16 August 2026.
• GitHub Docs — Usage-based billing for organizations and enterprises — Billing is per granted seat, not per active user; AI Credits are consumed against a monthly allowance with usage-based overage past it. Read 16 August 2026.
• GitHub Blog — Copilot is moving to usage-based billing — Base plan prices unchanged at $19 and $39. The credit allowance figures in this post are internally inconsistent, which is why no allowance number is quoted anywhere on this page. Read 16 August 2026.
• GitHub / Microsoft controlled experiment, 2023 — 55.8% faster, n=95, writing an HTTP server in JavaScript from scratch; 95% CI [21%, 89%], p=0.0017.
• METR — Early-2025 AI and experienced open-source developer productivity — Randomised trial, 16 developers, 246 tasks: +19% completion time (slower), CI [+2%, +39%]. Superseded by METR's February 2026 update: −18% for returning developers (CI [−38%, +9%]) and −4% for new ones (CI [−15%, +9%]), with the authors flagging their own selection bias.
• Google DORA, 2024 and 2025 — 2024: −1.5% delivery throughput and −7.2% delivery stability per 25% increase in AI adoption. 2025: throughput turns positive as teams learn, the stability penalty persists; 90% adoption, 30% report little or no trust in AI output.
• GitClear — The Maintainability Gap (June 2026) and the January 2026 velocity note — 623 million code changes, 2023–2026: duplication +81%, within-commit copy/paste +41%, refactoring line moves −70%, legacy maintenance −74%, error-masking constructs +47%, two-week churn +15%, function connectivity −35%. Separately: heavy AI users measured against their own past selves gained ~25% velocity.
• Cursor — Pricing (comparison price only) — Teams at $32 per seat per month on annual billing. Read 16 August 2026.

Provenance.Per-seat prices, tier requirements, the IP-indemnity condition, the Enterprise Server exclusion and the 22 April 2026 self-serve pause were read from GitHub’s own plans documentation and pricing page on 16 August 2026. The research effects (GitClear, Uplevel, METR, DORA, the 2023 GitHub/Microsoft experiment) are quoted as published. The annual costs, the break-even, the mixed-tier comparison, the idle-seat figure and the rework dollars are our arithmeticon those published numbers at a $180,000 fully-loaded cost per engineer, 2,000 hours a year and a 20% baseline rework rate — stated so you can substitute your own. No AI-credit allowance is quoted, because GitHub’s own announcement post is internally inconsistent on it. No sponsorship, no affiliate link, no vendor review of this page. Last reviewed 28 August 2026. Corrections: [email protected].

Questions we get asked about this one

Other verdicts at this size

• Is Atlassian Jira worth it for a 100–1,000 person engineering team? — Yes, on Standard — the tool is not the decision at this size, the edition is. Jira Cloud Standard is $36,500 a year at 500 users and $68,500 at 1,000, and the Premium step is $26,000 and $42,000 a year for a checkbox that applies to everyone on the site.
• Is LaunchDarkly worth it for a 100–1,000 person engineering team? — Yes, once you have counted service connections and client-side MAU — neither of which is your headcount. $10 per connection per month and $8.33 per 1,000 MAU means one extra environment across 80 services costs $9,600 a year, and the governance tier is quote-only.
• Is New Relic worth it for a 100–1,000 person engineering team? — Yes, if the user-type policy is written before the contract. Full platform users are $349 a month at Pro list against $49 for a core user, so fifty misassigned people cost $180,000 a year — several times the $0.40-per-GB data line everyone actually negotiates.
• Is Okta worth it for a 100–1,000 person engineering team? — Yes — identity is a control, not a convenience, at this size. The money is the suite: Okta publishes $6, $14 and $17 per user per month, so 500 identities is $36,000, $84,000 or $102,000 a year, and the upper suites only pay if somebody staffs the automation.
• Is PagerDuty worth it for a 100–1,000 person engineering team? — Yes, on Professional and billed annually. At 500 responders that is $126,000 a year against $246,000 on Business — and the money leaks through licences held by people who left the rota, at $252 each per year.
• Is Sentry worth it for a 100–1,000 person engineering team? — Yes, and paid plans include unlimited users so hiring does not raise the bill. The money is the rate card: Business re-prices every error at two to three times the Team rate, which is $5,268 a year at two million errors a month and $19,668 at ten million.
• Is Cursor worth it for a 100–1,000 person engineering team? — Yes, on Standard seats billed annually. At 500 engineers that is $192,000 a year against $576,000 for the all-Premium version of the same purchase — and the rework exposure is 4.7x to 38x the licence.
• Is Datadog worth it for a 100–1,000 person engineering team? — Yes, but only on a host count you measured yourself. Infrastructure Pro and APM on the same 500 hosts is $276,000 a year at list — and Datadog's own published retention rate says bills like yours grow about 20% a year.
• Is Snowflake worth it for a 100–1,000 person engineering team? — Yes, on Standard at $2.00 a credit unless somebody can name the Enterprise feature. One X-Large warehouse left running is $280,320 a year — the settings decide this bill, not the contract.
• Every published verdict, in one index — the same arithmetic on the rest of your renewals list.

Weighing one tool rather than a whole estate? Our published One-Tool Verdict on Copilot Business at 40 engineers is the same discipline at a smaller scope, and the do-not-adopt library carries the calls we make against other vendors.