Free preview5 min

The $9 Prototype and the $200 Game

Reviewed by Human ยท Updated September 21, 2026

One team handed a top AI model a task and spent $9. Another team handed the exact same model the exact same task and spent $200. The cheap run failed. The expensive run worked.

Why Does the Same Model Produce Wildly Different Results?

Because the model is only one part of the system. Drop a capable model into an empty repo with a one-line prompt, and it improvises. Give that same model instructions, a memory of what it's done, and a way to check its own work, and it behaves like a careful engineer. **Harness**: the environment, files, and checks built around a model so its output is reliable, not just its instructions. The model decides *what* code to write. The harness governs *when, where, and how* it writes it. Same brain, different scaffolding, opposite outcome.

The harness doesn't make the model smarter. It makes the model's output reliable.

0

The course grounds this in a controlled experiment Anthropic ran: same model (Opus 4.5), same prompt to build a 2D retro game editor. Without a harness it spent $9 in 20 minutes and produced something that didn't work. With a full harness (planner, generator, evaluator) it spent $200 in 6 hours and built a game you could actually play. Sources: the course README and Anthropic's 'Effective harnesses for long-running agents.'

Your turn

Think about the last time a coding agent let you down. Write one sentence describing what it did wrong. Then label it: was the model incapable of the task, or was the model missing something in its environment?

Reflect

Most 'the AI is bad at this' moments are actually 'the AI had nothing around it to catch the mistake' moments.

The unharnessed run and the harnessed run used the same model and the same prompt. What actually changed the outcome?

end of module

You've finished this module.

Mark it complete to earn your XP and keep your streak alive.

Progress saved locally ยท Sign up to earn XP