On June 1, 2026, GitHub Copilot moved every plan to usage-based billing, charging for AI work by token consumption rather than a flat seat fee. A day later, Microsoft is set to unveil its own coding models at Build 2026. Together, these moves reprice AI-assisted development and turn it into a variable cost businesses now have to manage.
For three years, the AI coding assistant was sold like any other SaaS seat: a predictable monthly fee per developer. That era is ending. The tools are not getting more expensive overnight, but the way you pay for them is changing in a way that rewards measurement and punishes inattention.
What Actually Changed This Week
Two developments landed within 48 hours of each other, and they point in the same direction.
First, GitHub Copilot moved to usage-based billing on June 1, 2026, replacing the old system of premium request units. Under the new model, every plan includes a monthly allotment of GitHub AI Credits, and additional usage is billed by consumption. According to GitHub's billing documentation, credits draw down based on token usage, counting input, output, and cached tokens at each model's published API rate.
Second, Microsoft is preparing to unveil a suite of homegrown MAI models at its Build conference on June 2 and 3, 2026, including a coding model aimed at strengthening GitHub Copilot. Multiple outlets frame the launch as a direct effort to reduce Microsoft's dependence on OpenAI and Anthropic, taking aim at the same coding-tool market where Anthropic's Claude Code has pulled ahead.
One story is about price. The other is about who controls the model underneath the price. For businesses that have quietly made AI coding tools part of their development stack, both matter.
How the New Copilot Pricing Works
The headline prices did not move. Copilot Pro remains $10 per month, Pro+ stays at $39, Business is $19 per user, and Enterprise is $39 per user, according to GitHub's pricing. What changed is what those dollars buy.
Each plan now includes that same dollar figure in monthly AI Credits, where one credit equals one cent. A $19 Business seat comes with $19 in credits. Once a developer exhausts the monthly allotment, additional usage is billed against a budget you set, at per-token rates that vary by model. Routine code completions and next-edit suggestions remain free and do not consume credits, per GitHub's documentation. The credits get spent on the heavier agentic work: multi-file edits, chat with large context, and autonomous task execution.
That distinction is the whole story. The cheap, ambient autocomplete that defined Copilot's first era stays bundled. The expensive, agentic capability that defines its current era is now metered. Some developers reacted to the announcement bluntly, telling Visual Studio Magazine the change means you "get less, but pay the same price" at the entry tiers. Whether that is fair depends entirely on how heavily your team leans on agentic features.
Why Microsoft Is Building Its Own Coding Model
The pricing change does not happen in a vacuum. The model underneath these tools is expensive to run, and Microsoft has decided it would rather own that cost than rent it.
Microsoft's own engineers tell the story. The company let thousands of staff use Claude Code internally, and it became popular enough that Microsoft is now phasing out internal Claude Code licenses by the end of June and pushing teams to its Copilot CLI, even though many engineers preferred the Anthropic tool. The new MAI coding model is the supply-side answer to that demand: if developers want frontier coding assistance, Microsoft would rather serve it from a model it controls than pay a competitor per token to do it.
Our take: This is the same vendor-concentration dynamic reshaping the rest of the AI market, now playing out inside the developer tooling layer. We covered the broader version of this story in our analysis of the vendor power shift among frontier AI labs. The coding-tool version has a sharper edge for buyers, because the tool, the model, and the cloud bill increasingly come from the same company, and your switching costs rise with every workflow you wire to one vendor's runtime.
The Uber Warning: When AI Coding Costs Run Away
If you want to understand why usage-based pricing demands attention, look at what happened at Uber. The company exhausted its entire 2026 AI tools budget in roughly four months, according to Fortune, after rolling out tools including Claude Code to around 5,000 engineers. Reported monthly costs ran from $500 to $2,000 per engineer as usage climbed, and an internal usage leaderboard helped drive adoption from 32% to 84%.
The most instructive detail is not the overspend. It is that Uber's president and COO, Andrew Macdonald, reportedly questioned whether the spending connected to innovations that actually serve customers, even as roughly 70% of committed code involved AI. High adoption and high spend did not automatically translate into a clear return. That gap between usage and value is exactly the AI ROI measurement problem that catches so many organizations: it is far easier to measure how much AI you are buying than how much value it produces.
Under flat seat pricing, an over-enthusiastic team costs you nothing extra. Under usage-based pricing, the same enthusiasm shows up directly on the invoice. The Uber episode is what unmanaged usage-based AI spend looks like at scale, and it arrived just as the entire market moved toward that pricing model.
What This Means for Your Business
The strategic implication is not "AI coding tools got expensive." It is that the cost is now elastic, and elastic costs reward governance.
Organizations that treat AI coding spend as an unmanaged utility tend to get surprised by it. The ones that do not have usually put clear cost-governance and vendor strategy in place before scaling seats: per-team budgets, visibility into cost per engineer, and a deliberate choice about which work runs on premium models versus cheaper ones. That discipline is the same lever we described in what the DeepSeek effect means for your AI budget, applied to the developer tooling line rather than the inference line.
There is also a portability dimension. As Microsoft pulls its coding model in-house and GitHub meters the agentic features, the tooling, the model, and the billing converge on single vendors. That makes each tool stickier and raises the cost of changing your mind later. The hedge is the same one that works elsewhere in the AI stack: keep your prompts, your workflows, and your evaluation criteria portable enough that you can move a workload if the economics shift.
How to Budget for Usage-Based AI Coding Tools
You do not need a procurement overhaul. You need four habits.
- Measure your real token consumption first. Before scaling seats, run a small pilot with full usage tracking. The number that matters is cost per engineer per month at realistic usage, not the headline seat price.
- Set budgets and alerts per team. Usage-based billing without spend limits is how Uber-style surprises happen. Cap budgets, route overages through approval, and review the meter weekly during rollout.
- Match the model to the task. Premium models for hard, multi-file refactors; cheaper or bundled features for routine autocomplete. Most teams overspend by sending trivial work to expensive models out of habit.
- Tie spend to output, not adoption. Track a productivity signal you actually trust, such as cycle time or throughput, against the AI bill. Rising usage is not the goal; rising value per dollar is.
Common Mistakes to Avoid
The first mistake is assuming the flat seat price is the whole cost. With agentic features metered, a $19 seat can cost several times that for a heavy user. The second is letting adoption metrics stand in for value, the trap Uber's leadership flagged: 84% adoption tells you nothing about return on its own. The third is wiring critical workflows to a single vendor's agent runtime without an exit plan, just as that vendor pulls the model in-house and gains pricing leverage. Each mistake is avoidable with measurement and a portability mindset.
Key Takeaways
- GitHub Copilot moved every plan to usage-based billing on June 1, 2026, replacing premium request units with AI Credits priced at one cent each and billed by token consumption; code completions stay free.
- Microsoft is unveiling homegrown MAI coding models at Build on June 2 and 3, 2026, to reduce reliance on OpenAI and Anthropic, signaling that the tool, model, and bill increasingly come from one vendor.
- Uber exhausted its 2026 AI tools budget in roughly four months at $500 to $2,000 per engineer monthly, a preview of what unmanaged usage-based spend looks like.
- The strategic response is governance, not avoidance: measure real consumption, set per-team budgets, match models to tasks, and tie spend to output rather than adoption.
Navigating the new economics of AI coding tools does not have to be a solo effort. Book a free discovery call and let's map out what the shift to usage-based pricing means for your engineering budget and your roadmap.