The Prompting Wars Are Ending. The Specification Wars Are Just Starting.

If you’re running your own AI assistant, you’re probably wondering when you’ll get access to Mythos — the new, far more capable class of AI model Anthropic is starting to roll out. Ethan Mollick got early access to the first one, Claude 5 Fable, and his writeup is worth your time.

But “when do I get it” is the wrong question. The right one is: am I ready for how it changes the way I work?

Here’s what Mollick describes. He gave Fable a vague, ambitious instruction — build a researched, beautiful map of global travel times — and walked away. The model spun up its own swarm of cheaper sub-agents to dig through 2,200 flights, rail schedules from the TGV to the Shinkansen, and road-speed data from academic papers. While those ran, it started coding. Then it launched more agents to test and verify each other’s work. Nine and a half hours on one project. The result was real software that had never existed because it was never worth a human’s time to build.

His verdict: working with AI used to feel like being a wizard. You chanted the spell, you understood the incantation, something happened. Fable turns you into a patron instead. You commission the work, you sign off on the result, but you never set foot on the floor where the hundred small decisions get made.

That’s the shift. And here’s the part nobody’s saying out loud: the skill that matters is no longer coaxing. It’s clarity.

The old game was tricks — magic phrasings, “think step by step,” clever prompt hacks to squeeze competence out of a model that wasn’t quite there. A model that already understands you kills that game. What’s left is harder and far less gameable: do you actually know what you want? Can you write the brief? Can you tell, looking at the finished thing, what’s wrong with it?

The prompting wars are ending. The specification wars are just starting.

This is exactly why the relationship you build with your agent now matters more than which model is behind it. The people who’ll get the most out of Mythos aren’t the ones with the cleverest prompts. They’re the ones who already know how to brief an assistant, set the guardrails, and judge the output — because they’ve been doing it daily on the models we already have.

OpenClaw is a self-hosted AI assistant you run on your own server, with your own files, memory, and tools.

I teach two classes on setting up and getting the most from OpenClaw on Udemy: Easy OpenClaw and Get Real Work Done With an AI Assistant.