Anthropic called it a 25% increase. Claude Code users are getting a 17% cut — what it means for teams budgeting on AI coding tools
Anthropic announced a 'permanent' 25% increase to Claude Code's weekly usage limits from 14 September 2026, then had to admit — after deleting the original post — that it's actually a 17% reduction from today's temporarily boosted allowance, a reminder that AI coding tool capacity isn't a fixed cost.
4 September 2026
On 29 August 2026, Anthropic announced what it called a “permanent” 25% increase to Claude Code’s weekly usage limits for Pro, Max, Team and seat-based Enterprise plans, effective 14 September. The framing didn’t survive contact with its own user base. Claude Code has been running on a temporary 50% capacity boost since earlier this summer; when that expires on 13 September, the new “increased” limit of 125% (relative to the pre-boost baseline of 100%) is actually a step down from the 150% teams have been working with for weeks. Anthropic deleted the original post after the backlash and posted a clarification admitting the plain arithmetic: “Compared to today, this works out to a 17% reduction in weekly limits on Claude Code.”
Why this matters beyond one company’s messaging misstep
This is the second time in three months Claude Code’s weekly caps have moved (a similar boost-then-taper pattern played out around 13 July). For any team that has built AI-assisted development into its actual delivery workflow — not as a novelty, but as a load-bearing part of how code gets shipped — usage limits are now a real capacity planning variable, on the same footing as seat licences or cloud compute budgets. Anthropic hasn’t published a reason for the tightening, which reads as compute allocation catching up with demand rather than a pricing change, but the practical effect is the same: teams relying on a single AI coding vendor’s default plan can see their effective throughput drop with little warning.
So what
If Claude Code (or any single AI coding assistant) is central to your delivery velocity, this is worth treating as a signal to diversify tooling and plan headroom rather than assume today’s limits are permanent — the vendor landscape is moving fast enough that lock-in to one tool’s capacity ceiling is a real delivery risk. It’s also a reminder that “AI-assisted” doesn’t mean “AI-replaced”: teams that pair these tools with senior engineering judgement are far less exposed to a vendor’s capacity decisions than teams treating the AI as the whole workflow. See how we run AI-assisted development with that balance built in, or get in touch to talk through how it affects a build you’re planning.