The popular story is that AI subscriptions are about to double. £20 becomes £200, the labs stop subsidising us, and everyone wakes up to a new price. That's probably part of the eventual story. But the squeeze is already happening, just not through the front door.
This week's piece on samjenkins.com walks through the patterns I keep noticing. Tier creep, where the £20 plan today is roughly what £200 bought you a year ago. GitHub Copilot moving every subscriber onto usage-based billing from June 2026. Ads quietly arriving on the consumer free tier. Agentic workflows multiplying the tokens a single task consumes. Per-token prices keep falling. Monthly bills keep rising.
The full piece covers what helps (avoid lock-in, right-size each task to the model, track the changelogs, audit the spend, lock annual on the one tool you're sure of) and what doesn't (don't AI-everything, be deliberate about what you paste into ad-supported tiers, don't bet your unit economics on today's prices, and don't keep your only knowledge base inside one provider's product).
The single frame I'd take if I had to pick one: AI is excellent at finding the rule. It's often the wrong tool to run the rule. Once you've found it, ask AI to write the Zapier flow, the Apps Script, or the small Python file that runs the rule deterministically. No tokens, no provider risk attached.
Read the full piece here: https://samjenkins.com/writing/ai-cost-squeeze-already-here
More next Saturday.