Professor Oak’s gig economy
Let's talk about Professor Oak. Oak's entire operation runs on other people's effort. He gives trainers a starter Pokémon, hands them a Pokédex, a device that sends data back to him, and sends them into the world to catch every species in existence.
The trainers think they're on an adventure. They are.
They're also running Oak's distributed research infrastructure at their own expense while he stays in Pallet Town updating his notes.
A generation of VC-backed startups raised a combined fortune to build the harnesses that made Claude work in production. On February 20th, Anthropic updated their Terms of Service. On April 4th, they cut third-party access. On April 8th, they launched their own infrastructure.
The Pokédex was always the product. The trainers were always the infrastructure. Professor Oak just got a $0.08/session-hour pricing model.
The commentary landed immediately on the obvious frame: Anthropic's "AWS moment." Same playbook as Amazon, 2006. Let third parties build the production layer, then commoditize it.
When Amazon competed with its own ecosystem, the incentive was visible and the anger was proportional. Anthropic abstracted infrastructure and trust governance in the same move. Their engineering post reaches explicitly for the OS analogy.
They call Managed Agents a "meta-harness" — the layer beneath all other harnesses. An OS doesn't just run programs. It decides what can run, controls what they can access, and becomes the governance layer for everything above it. The harness harnesses the harness.
"Harnesses encode assumptions that go stale as models improve." Their argument for Managed Agents: don't bake today's model behavior into your infrastructure because models will improve and those assumptions become dead weight. That logic works in every direction, including toward Anthropic's own managed infrastructure.
When Amazon competed with its customers, you could leave. You could go to Azure, Google Cloud, to a server in a basement in New Jersey.
When Anthropic bundles the infrastructure play with the safety brand, leaving requires making a different argument entirely, that trusted governance, scoped permissions, and execution tracing are things you don't want. Nobody wants to hold that position.
When The Register documented developer backlash, when Hacker News flagged the lock-in concerns, Anthropic didn't respond. OpenAI did, by explicitly endorsing third-party harness use of Codex subscriptions the same week.
One company running the AWS playbook. The other running something that doesn't have a name yet, because the last time someone bundled safety with infrastructure at this scale, we called it an operating system and moved on.
Knowing this and caring about it are, it turns out, different things...
Claude is my Charizard, chosen, loved, occasionally lying down on the battlefield while I'm mid-argument.
Ash sent Charizard into the Indigo League.
Charizard took a nap.
Ash kept him anyway.
So did I.