The Value Isn't in the Model Anymore — It's in What Surrounds It

Watching what's been happening on the dev tooling side these past months, there's one observation that won't leave my head: a newly released model isn't news on its own anymore. The interesting part is what's being built around that model.

A few months ago, when a new frontier model dropped, my reflex was "what did it score on the benchmarks, how much does it cost, which API is it on?" That reflex doesn't fire anymore. Because the gap between model releases looks small next to what's happening in the ecosystem around them.

Where the signal is coming from

Look at the first half of May. At Code w/ Claude, Anthropic didn't announce a new model — they said "today is about the products working better." Instead they talked about Code Review, Routines, Auto-fix — i.e., the pipeline the model sits in. Even in the PwC partnership announcement, the thing actually being sold isn't the model's raw power; it's how the model fits into insurance underwriting, M&A diligence, month-end close.

Over on GitHub, mattpocock/skills climbed into trending. The repo isn't "a new agent framework" — it's just one engineer's personal .claude/skills/ folder. Standalone SKILL.md files for TDD, triage, diagnose. That this is what's getting attention is the same signal: people aren't asking "did the model get stronger" — they're asking "how do I channel the model."

What this changes

As a frontend developer, this is where I see it most concretely:

When starting a new project, picking the model is no longer one of the first three decisions. Instead: what does this project's SKILL.md layer look like? Which routines should run automatically overnight? Will the agent do code review, or will I?

The tool's brand is becoming meaningless. A single SKILL.md file works in Claude Code, Cursor, and OpenCode alike. Even if the model changes, you get the feeling of "I have one context layer, it's portable." That's already past last year's "which IDE?" question.

The industry is pricing this in too. Look at Windsurf 2.0 — it's stepping away from the editor + sidekick model toward Spaces and a Kanban command center. Cursor has its parallel agent architecture. The shared story is the same: being the smartest thing in the room isn't enough for the model anymore; you have to bend the model tightly into the flow of the work.

What this is telling me

Models still matter, and when there are real jumps they'll get talked about again. But day to day, a new model means "slightly better output"; a tooling shift is a completely different way of living — "I leave a routine before closing the laptop, the PR is ready by morning."

That's why what's been on my mind lately isn't "Opus or Sonnet." It's "what can I build around this work so I don't keep redoing it?" — that question. Models progress on their own. The real work is designing where the model lands.

Will this argument still hold a year from now? Not sure. A model leap might sweep it all away at some point. But where I stand today, the slogan "the value isn't in the model anymore — it's in what's around the model" still feels right to me.