Models & thinking

By the end of the last chapter the agent finally knew your project: the CLAUDE.md hierarchy meant it picked up your conventions, your stack, and the shape of the payments service without you re-explaining a thing. That settled what the agent knows. This chapter settles a different dial — how hard it thinks — and the whole skill is realising that the answer changes from one task to the next.

Here’s the trap most people fall into. They find a setting that works, leave it there, and pay for it on every prompt. Either they run the most capable model with the deepest reasoning on everything — and burn budget and latency reformatting an error string — or they leave it light and fast and then watch it flail at a problem that actually needed reasoning. Both are the same mistake: a fixed setting against a variable workload. The operator’s move is to shift gears, matching the model and the thinking budget to the task instead of to habit.

The day we’ll follow

Same payments service, but today the work is deliberately lumpy — it spans the whole difficulty range in a single afternoon, which is exactly what makes the gear-shifting visible:

• A copy tweak first thing: a charge-failure error message reads badly to customers and needs one string rewritten. Trivial. The agent shouldn’t be reaching for its deepest reasoning to do this — switch to a lighter, faster model and move on.
• Then the one that ruins afternoons: a concurrency bug in the charge ledger. Two requests race, the running balance ends up wrong, and it only reproduces under load. This is not a string edit — it needs the agent to hold the whole interleaving in its head and reason about it. Turn the thinking up and let it earn the cost.
• And underneath both, the discipline that ties the day together: spending model and reasoning where the context is genuinely hard, and pulling both back down the moment the work goes mechanical again — including when raw speed is worth paying for.

The through-line is the same one the whole course keeps circling: context engineering is about putting effort where it pays. You learned to spend the context window deliberately in Sessions & context. This is the same instinct applied to the model’s reasoning — the gnarly problem gets the heavy gear, the copy tweak gets the light one, and you stop paying max for a workload that’s mostly mechanical.

Start with the easy one — the copy tweak, and the model switch that fits it.