If you mostly use chatbots and assume the next jump in AI agents is just a better model, ai-agent-book has a more useful reset: the model is only one third of the system [C002]. Miss that, and you can waste time, budget, and attention chasing the wrong upgrade.
bojieli / ai-agent-book [C001] is the kind of repo you almost scroll past, then reopen because it changes a beginner's next decision. A new AI update is worth reading only if it changes your next decision. This one does: stop treating model choice as almost the whole game.
The README cuts the recipe into three parts: model, context, and tools. Not "stronger model = stronger agent." If you only know chatbots, that is the plain-English version: the answer depends on what the model is given and what it can use, not just the model name.
Chapter 1 has the real reversal. Better models do not erase the need for rules, checks, and correction around them. They raise it. The setup around the model does not get less important as models improve; it gets more important.
Chapter 2 pushes the harder point: context quality sets the ceiling. The repo's claim is that a mid-tier model with the right instructions and files can beat a top model working half-blind. Better models still matter. This is a reweighting, not a denial [C002].
So if you are deciding what to learn next, spend less time chasing model rankings and more time learning context, tools, and self-checks. Share this with the person who still thinks "agent upgrade" mostly means "swap the model." Boundary: repo read only, from the README plus Chapters 1 and 2; no live run.