Will better models make your current setup more valuable, or obsolete?

Every agent investment falls into one of two categories: things that get more valuable as models improve, and things that become irrelevant. Most teams put the bulk of their effort into the second category without realising it.

What compounds: context and permissions. Well-structured, discoverable information — CLAUDE.md files, knowledge bases, tool descriptions — becomes more valuable with every model upgrade because smarter models extract more from the same context. Permission infrastructure — what agents can access, what they’re allowed to do, how trust flows across boundaries — is durable because it reflects your organisation’s real constraints, not the model’s current limitations.

What decays: anything built to compensate for what the model can’t do yet. That changes with every release. Claude Code’s evolution is the clearest example — the team removed code with each model upgrade, not because they were simplifying, but because the problems their code solved had disappeared.

The question for your investment strategy: for every component in your agent stack, ask whether a smarter model makes it more useful or less necessary. That ratio tells you whether you’re compounding or resetting.

Go deeper: AI Agent Reliability Is Getting Easier. The Hard Part Is Shifting. traces how Claude Code’s architecture evolved — and why the real investment is context and permissions, not code.