Open reasoning loop
Rare. The model reasons in a self-directed loop but never touches a tool.
“Agent” isn’t a goal. Skip the label. Two simple independent questions help you decide what to build.
Most teams ask “should we build an agent?” but the word means too many things to answer cleanly, so skip it. Describe what you’re building along two axes - whether it acts, and who decides - and you’ll land in one of four kinds of systems, with what each one costs you. It’s the framework from the video, made clickable: useful whether you’re writing the code or just signing off on it.
▶ Watch the episodeDoes it take actions in the world (call tools, send messages, change data, spend money); or just produce text?
Can you write the steps down in advance, or does the path depend on each input in ways only the model can work out?
Answer both questions or try a worked example to see where this lands.
Start in the lowest quadrant that actually solves your problem. If a prompt does it, it’s a prompt. If a scripted pipeline does it, that’s a great place to stop. Climb only when the problem forces you to, and climb one question at a time.
If someone tells you they’re building you an agent, you now have two questions for them: can it act? and who’s driving?
Gartner predicts that more than 40% of agentic AI projects will be cancelled by the end of 2027 — driven by cost, unclear business value and weak risk controls, not by weak models. Those failure modes are exactly what the top-right quadrant buys you: autonomy and tools a problem never asked for. Choosing the lowest quadrant that solves the problem is how you stay out of that number.