Capture the model's discipline

“discipline makes the astronomer; the telescope only helps”

Put the strongest model available in the hands of a beginner, and give a seasoned engineer a model 2 years older: the engineer still ships the better result. Nate’s field notes open with that thought experiment. If the practitioner beats the model upgrade, the advantage was never the model; it is the instructions, systems, and loops around it. This guide is about extracting how the best model you can rent works, so the method survives when the model is priced up, capped, or taken away.

“You can't keep the model's intelligence, but you can keep its process.”

Nate (AI Automation), “How I Make Opus Think Like Fable (5 easy steps)” · 1:13

01The judgment lives in the plan

Nate ran the same multi-agent workflows three ways: a top model directing equally strong workers, the same planner over mid-tier workers, and a strong planner over the cheapest tier available. Results were roughly interchangeable; costs were not. The cheap-worker runs landed around 3 times cheaper for the same output. When the expensive model scopes, anticipates what could go wrong, and reviews what comes back, the workers only execute a well-specified plan, and that is exactly what cheap models are good at.

So treat the top model as a senior engineer about to rotate off the team: have it review your setups, improve your skills, and explain its own reasoning while you still have it.

02Turn the discipline into a skill

When a top model hands you a deliverable you love but cannot quite explain, reopen the session and interrogate it: what did you consider, how did you verify it, why does this output work. Then have the model condense the answers into an installable skill file any model can run, the format covered in Expertise you can install. The interrogation costs about 15 minutes; the discipline it captures outlives the model. Nate’s version gates work through five stages:

Gate	What it forces
Scope	Devil’s advocate before any steps are listed
Evidence	Gather before reasoning
Adversarial reasoning	Attack your own draft answer
Verify	Evidence in hand before declaring done
Report	An honest read on how sure it is, not optimism

Scoping does the most work. A plan lists steps; scoping plays devil’s advocate first: which assumptions might be false, which unknowns the plan silently walks past.

The verify gate runs on two rules. Recognizing something from training does not mean it is current knowledge, and a file implied to exist is not a file verified to exist: audit every claim against what the tools actually returned. That habit is the backbone of Guardrails for hours-long agent runs. Encoded as a skill, the discipline improves a mid-tier model and runs on anything, including models you host yourself.

03Route by table, spend by effort

Decide per task how much intelligence to rent. Two settings matter.

The model. Keep a routing table for your toolkit, one row per model, scored for cost, intelligence (reasoning strength), and taste (creativity, interface judgment, framing). The two scores do not move together, and a table that collapses them routes design work to a strong-but-tasteless model and hard debugging to a tasteful-but-weak one. Hand the table to your orchestrator and let it delegate against it. When the pressure is a subscription limit rather than an invoice, Stop hitting your token limit turns the same routing into budget relief.

The effort. Score-versus-cost curves show effort levels overlapping across tiers (a top model on low effort lands near the previous tier on high), so the model-and-effort pairing is the real unit of choice. Past a point, maximum effort buys longer runs, second-guessing, and worse output than the tier below on high. The tell is a response that argues against its own correct first answer. When you see it, route down.

“We don't own these models. So, what we can own is our processes, our systems.”

Nate (AI Automation), “How I Make Opus Think Like Fable (5 easy steps)” · 9:36

Further reading

• Nate (AI Automation) · How I Make Opus Think Like Fable (5 easy steps), the field notes this guide distills
• Anthropic · building effective agents, the orchestrator-worker pattern from the vendor side
• Anthropic · Claude models overview, the tiers a routing table scores
• Wikipedia · mixture of experts, the same routing intuition inside the models themselves
• Yao et al. · “ReAct: Synergizing Reasoning and Acting” (arXiv), the plan-act-observe loop the gates formalize

# Capture the model's discipline Consult when deciding which model runs a task, when you want to keep an expensive model's working discipline after it is priced up or capped, or when structuring a planner-and-worker split. The method survives; the model does not. For live model ids, pricing, and tiers, use the `claude-api` skill rather than restating version-pinned facts. ## Put the judgment in the plan, not the workers Quality lives at the plan-and-verify layer. A strong planner over cheap workers matched a strong planner over strong workers in Nate's runs, for roughly one third the cost, because executing a well-specified step needs far less intelligence than designing it. - The expensive model scopes, anticipates failure modes, and reviews each report, redesigning the next step from what came back.- Cheap models execute the well-specified steps and report.- Treat the top model as a senior engineer about to rotate off: have it review your setups, improve your skills, and explain its own reasoning while you still have access. ## Extract the discipline into a skill file When a top model hands you a deliverable you love but cannot explain, do not just save the deliverable. Reopen the session and interrogate it: what did you consider, how did you verify, why does this work. Have it condense the answers into an installable skill file any model can run. Costs about 10 to 15 minutes at the end of a session; the discipline outlives the model.