05 — The Anthropic Ban and the Counter-Move

What happened

Anthropic silently blocked using Claude subscriptions inside OpenCode (a practice many third-party tools used), dropping the block at ~9 p.m. with no communication. The user base erupted; OpenCode barely commented 21:34.
Dax's read: Anthropic is "new to working with developers" — protecting the business is fine, but a phased, communicated rollout over a month would have avoided the concentrated moment of outrage 21:57.
Meta-point about AI-speed leverage: a fast-growing company forgets how much leverage it has — one small action ripples through millions. OpenCode lived this too: a light-mode bug "flashbanged" about a million users in a week 22:19.

They knew this day would come and felt excited when it did. They also knew only a small subset of their 650k MAUs actually had $200 Claude Max subscriptions — the Twitter bubble exaggerated the damage 23:20.
Crucially, they had spent the prior weeks closing official-support deals: Microsoft had already agreed to bless GitHub Copilot in OpenCode, with more in the works — all unannounced. The big one they hadn't approached: OpenAI 24:05.
The night of the ban Dax messaged OpenAI: "tomorrow everyone's going to be angry at Anthropic — you can score a PR win by officially supporting OpenCode." OpenAI confirmed next morning; the integration shipped and was announced by end of day 24:25, 24:52.

"The strategy that we know how to play really well is to pick one temporary bad guy and then galvanize all their competitors to push something forward against them" — same as OpenNext vs. Vercel; this time Anthropic was in the bad-guy seat 25:14.
Even a small company can make billion-dollar companies move if it applies pressure in the right places; the outcome (more people able to use the tools they want) is the point of competition 27:12, 27:30.
And the roles can rotate: "at some point maybe OpenAI becomes a bad guy that we have to galvanize everyone around" 27:12.

Dev tools that become standards are adopted bottom-up, which requires thinking like a consumer company — and "programmers are horrible at B2C products"; they don't even realize dev tools are B2C 28:13, 28:41.
OpenCode optimized the first impression: they built a ground-up terminal rendering framework (unlike Claude Code and others, which used Ink) so the product feels different from the first launch — "you walk away thinking the people that did this are competent" 29:05.
Then they minimized friction to first prompt, including edge cases like locked-down enterprise laptops 29:28.
The inverted strategy: their harness "wasn't very good for the first five months — but it was good enough that most people couldn't tell the difference." They won share on experience first, then went back to make the harness good. Everyone else bet on "smartest harness wins" 29:55, 30:17.