In April 2026, GitHub announced a change to how it charges for Copilot, its coding assistant. Until then, every plan came with a fixed monthly usage allotment. Starting in June, billing began following actual consumption, measured by task. The company's own justification, published on its official blog, was blunt: "a quick chat question and a multi-hour autonomous coding session" used to "cost the user the same amount" under the old flat plan — and that no longer made sense.
That line sums up a problem bigger than one coding tool. According to a Bain & Company survey of more than 30 software vendors, 65% already blend traditional per-user licensing with some form of AI usage or outcome fee — and none of the vendors surveyed had moved fully away from the seat-based model. Bain's conclusion is that per-seat licensing is not dying, but it is no longer enough on its own.
The reason is simple: an AI agent is not a person logging in. It can process an hour's worth of several people's work, or sit nearly idle on a slow day. Charging — and budgeting — by user when the work happens by task measures the wrong thing. And that distortion does not stop at the relationship between a company and its software vendor. It repeats itself, point by point, inside every company that hands AI to its own team through a flat license.
Two ends of the same waste
Per-seat licensing wastes money on both ends at once, and it is rare to see both mentioned together.
On one end, whoever rarely uses the AI still costs the full license price every month. Flexera's 2025 IT asset management report, drawing on more than 500 IT professionals worldwide, found that 35% of respondents said software subscription waste increased over the past year — and that reclaiming unused licenses became one of the fastest-growing priorities among the most mature teams in that kind of management. An idle license is idle budget.
On the other end, whoever genuinely depends on AI day to day runs into an artificial ceiling: the individual license's monthly limit, sized for an average use that this specific person has already outgrown. The team that would benefit most from AI — because it handles more volume, more urgency, more repetition — is often the first to hit the plan's limit.
The result is a company that, at the same time, overpays for people who barely use it and underserves people who use it heavily. Neither problem shows up on the same invoice line, so finance sees only the total bill, never the shape of the waste inside it. No flat license solves both sides at once, because both sides call for opposite fixes.
Why just switching plans does not solve it
The most common reaction to this waste is buying a bigger plan for whoever hits the ceiling and trying to reassign or cancel the license of whoever does not use it. That helps, but it does not reach the root of the problem.
Bain itself points to a real obstacle in this transition: sales and procurement teams are used to negotiating by headcount, not by volume of work — and changing that logic requires rethinking internal process, not just the vendor contract. Until that shifts, the company keeps buying "seats" for work that is no longer done sitting in one specific seat.
On top of that, according to PYMNTS's coverage of this shift across the broader software market, many vendors are stacking AI charges on top of the seat license instead of replacing it — which helps the vendor capture revenue, but does not necessarily solve the visibility problem inside the company that is buying. Without knowing how much each team actually consumes, the AI budget becomes just another generic line on the spreadsheet, the same way the old license used to be.
The real blind spot is not "which plan to buy." It is not having, today, a simple answer to the question: how much is each team actually using AI — and for what. Without that answer, every renewal conversation repeats the same guesswork instead of starting from actual numbers.
What has to be in place
Fixing this means changing the unit of billing and internal tracking: from person to task, with real-time visibility.
Visible cost per conversation, not just per subscription. Every interaction with AI carries a real cost, and that cost can — and should — be shown, not hidden behind a flat monthly fee that averages everything away.
Credits shared by the team, not a license locked to one person. A shared budget lets whoever needs more in a given week use more, and whoever needs less does not leave the difference wasted — without requiring anyone to negotiate an individual plan upgrade every time the volume of work changes.
Automatic routing to the cheapest AI model that solves the task. Not every question needs the most expensive model available. Automatically directing routine tasks to more economical options — and reserving the more capable model for what actually needs it — cuts cost without requiring each person to understand the technical difference between models.
Consumption visible by team, not just the company total. A single number at month's end does not say who is using AI for what. Consumption broken down by department is what lets leadership decide where to invest more and where usage is still low — the same visibility gap Flexera's report identifies in traditional license management.
This is how Skyller was designed: transparent cost per conversation and AI credits shared by the team, instead of a flat per-person license that no one can calibrate correctly.
What shows up once consumption becomes visible
The most interesting effect of trading a per-person license for a shared budget is not just financial — it is discovery. Once AI consumption per team becomes visible, patterns emerge that the flat license used to hide.
One team may be using AI far more than any individual plan anticipated, signaling that a specific process is ready for more automation. Another may be underusing an expensive tool, signaling a lack of training — not a lack of need. A third may be concentrating usage on one expensive model for simple tasks, when automatic routing would already handle most of them at a much lower cost. None of these three signals show up when the cost is a flat monthly fee split by headcount.
That also changes the budget conversation: instead of negotiating "how many more licenses we need this year," the discussion becomes "where AI usage is growing, and the budget needs to follow." That is a question much closer to what the company actually wants to know.
Questions to bring to the next budget meeting
Before renewing another per-user license contract for AI tools, it is worth bringing these questions to the conversation with finance and IT:
- Do we know, today, how much each team in the company actually uses the AI available to it? If the answer comes only from the number of active licenses, the company is measuring seats, not usage.
- Has any team already hit its individual plan's limit while another barely uses its own? If so, a flat per-person license is wasting money on both ends at once.
- Does the AI budget rise and fall with actual task volume, or is it a fixed number negotiated once a year? A shared budget with visible consumption tracks real demand instead of locking it to an old number.
- Is there a routine for reclaiming idle licenses, or does it only get noticed at contract renewal? Waste that only surfaces at renewal has already gone unaddressed for months or years.
