“A remarkable amount of things that at one point we felt very proud of building and were in fact necessary are now hobbling the agent.” Hex estimates 5+ examples per week of once-necessary, now-harmful systems. Concrete example: they built a complex ID-mapping/reference-registry system because earlier models hallucinated cell IDs in notebooks with 50+ cells. Current models don’t need it — but the compensation code introduces bugs, complexity, and behavioral constraints that actively degrade the agent.
This extends Scaffolding is tech debt against the next model — the bitter lesson applied to product building from a forward-looking caution to a retroactive reality: the debt isn’t hypothetical, it’s accruing right now at production scale. The implication is that Build for the model six months from now, not the model of today isn’t just about targeting future capabilities — it’s about actively auditing and removing past compensations. Hex’s experience suggests a regular “compensation audit” practice: systematically testing whether old workarounds still help or hurt. The rate — 5+ per week — indicates this is a continuous maintenance task, not a one-time cleanup. The failure mode is Tool design is continuous observation — see like an agent: tools designed for weaker models constrain stronger ones.