Anthropic, last ten weeks: February ToS change, April claw tax, April founder ban, April Pro-plan test. Four moves. Two stuck, two got caught and reversed.
If Claude is in your workflow - research, strategy, building, writing - this is your story too. Here's how I plan around it.
The timeline
February 20, 2026. Anthropic updates its Terms to prohibit subscription OAuth tokens in third-party tools. OpenClaw, Cline, and a handful of others are on notice.
April 4, 2026, 12:00 PT. The change is enforced. Subscriptions stop covering third-party CLIs. Users pay per-token via API on top of their Claude plan. Anthropic softens the blow with a one-time credit - roughly one month of your plan, up to $200 (around 170 EUR for Max 20x), redeemable until April 17. This is the "claw tax." It's still in effect.
April 10, 2026. OpenClaw's creator Peter Steinberger gets his Claude account briefly suspended. Reinstated within hours after his post goes viral. Anthropic says they never meant to ban OpenClaw users. The optics are bad and they know it.
April 21-22, 2026. Anthropic quietly updates the pricing page. Red X on the $20 Pro plan for Claude Code. Max-only, $100-200/month. The Internet Archive captures it. Reddit, Hacker News, and X light up. Reversed within hours. Head of Growth Amol Avasare on X: "small test on ~2% of new prosumer signups. Existing subscribers aren't affected." Simon Willison flags that he may shift teaching workloads to OpenAI Codex.
Four moves. Two months. Two stuck, two got caught and retreated.
It's not whiplash. It's directional pressure.
Read those four events together and the pattern is clean: Anthropic is figuring out its margin on serious users in public. They're testing what sticks. What can we charge for separately? Where do we draw the floor? How much will users tolerate? The moves that stuck (the claw tax, the ToS change) became the new baseline. The ones that got caught (Steinberger, the Pro page) got walked back - but the test happened.
This isn't hostile, it isn't personal. It's a frontier lab defending margins on the most expensive workloads. OpenAI does this. Google does it with Workspace tiers. The volatility is structural.
Coding is where Anthropic tests first because coding is loud, expensive, and visible. If they can nudge coders to $100/month without losing them, the lessons apply to everyone else next. My guess for the next move isn't another sticker price. It's tighter compute caps or session limits on the free tier and the $20 Pro plan - affecting every workload, not just code. Research sessions that used to run long get cut short. Agents that used to chain twelve steps chain six. Easier to A/B than a visible price move, less public backlash, same margin effect.
If you use Claude for anything serious, this is your story too. You're just not at the front of the queue yet.
How I actually use Claude
Not just for coding, despite what the news cycle this month suggests.
Morning: research. I read through Discord briefs first thing - Claude scans primary sources on CEE media market moves, competitor positioning, regulatory shifts, and surfaces what's different from last week. That setup is the Discord + Claude Code bot I wrote about here.
Working day: brainstorming, writing, decisions. Product directions, market entry moves, positioning angles run through a business-sharks skill before a pitch meeting. Wrote about how I built that here. Posts, briefings, internal memos get a second read - Claude catches where I'm being lazy or repetitive. Agents run analysis on traffic patterns across our sites and surface what moved. None of this replaces judgment. It sharpens it.
Evening: building. A Discord bot that runs Claude Code from my phone. Content pipelines. Automation for the Ringier product teams. Claude, Cursor, Discord - picked by where I am and what I'm doing, not by loyalty.
Any one of those streams breaking means my day works differently. That's the reason policy changes matter to me, not an abstract principle about vendor lock-in.
Four rules I follow
1. One primary provider, one warm backup.
I pay Max on Claude for primary work. I keep a cheap ChatGPT Pro sub and use GPT or Codex on something real once a month. Not to replace Claude. So the day Anthropic does the next thing, I'm one switch away, not one training cycle.
2. Engineer at the right abstraction layer.
OpenRouter is the right layer when you're building an app on top of a model - you abstract over providers and fail over on outage. It's the wrong layer for CLI or tool-integrated work, because the tool itself is locked to the provider. Adding abstractions that don't solve your actual problem is more surface area to maintain for no gain.
3. Don't build on the letter of the contract.
My Discord bot spawns the actual claude CLI as a subprocess. Not because I read the ToS carefully - because I don't trust any vendor enough to bet my workflow on their contract language. Subprocess stayed legal through the OAuth change. OAuth-in-wrapper didn't. The principle generalizes.
4. Treat lock-in as a tax, not loyalty.
You pay it for quality today. The day quality moves, you move too. Anyone telling you to pick a side and stay is selling a worldview, not a strategy.
Three real options if Claude runs your day
Pay Max ($100-200/month). This is where Anthropic is trying to get you. Going there voluntarily costs the same as being pushed there - and keeps every skill, hook, and tool you've built. Still exposed to the next move.
Add OpenRouter as a backup for apps you build. Good when you're assembling something on top of models. Wrong layer for CLI or tool-integrated work. See rule 2.
Switch to OpenAI (Codex for code, GPT for research and writing). The quality gap has narrowed on code. For research and writing it was never that far. Simon Willison is publicly considering it. Real option. But it's the same dependency with a different logo, and you'd rebuild a year of muscle memory for parity, not upside.
Closing
The four moves aren't the point. They'll keep coming. The point is whether your stack, your skills, and your decisions are organized for that or against it. Plan for the direction of travel. The individual move is always somebody else's decision.
Sources: The Register on the Pro plan test, Simon Willison's take, InfoWorld on the OpenClaw credits, TechCrunch on the Steinberger suspension, TechCrunch on the claw tax, The Register on the February OAuth policy.