Stop Wasting Time on the Wrong AI Model

I wasted three hours yesterday. And I have nobody to blame but myself.

I was deep in NetSuite scripting, trying to track down a bug that had me going in circles. I had Claude Sonnet running on low effort because, honestly, that's my default for most things. It's fast, it's cheap on tokens, and 90% of the time it gets the job done.

This was the other 10%.

The Three-Hour Loop

Sonnet kept giving me answers that were almost right. Close enough to keep me engaged, but wrong enough to send me down rabbit holes. I'd try one fix, get a new error, ask again, get another plausible-sounding suggestion, try that, hit another wall. You know the cycle.

The frustrating part is that each individual response looked reasonable. It wasn't hallucinating or giving me garbage. It was just... not seeing the full picture. It kept treating symptoms instead of diagnosing the root cause.

The 15-Minute Fix

After three hours of this, I finally did what I should have done from the start. I switched to Claude Opus on max effort.

Fifteen minutes. That's all it took.

Opus didn't just fix the immediate error. It identified the underlying architectural issue, explained why my previous approaches kept failing, and gave me a solution that actually addressed the root problem. It connected dots that Sonnet kept missing.

And here's the thing that surprised me: the responses felt less compressed. Sonnet on low effort has this tendency to compact its answers, giving you the minimum viable response. Opus on max effort actually unpacked its reasoning, showed its work, and gave me the context I needed to understand why the fix worked.

The Real Cost of "Saving Tokens"

I thought I was being efficient by defaulting to Sonnet on low effort. In reality, I burned three hours of my time to save a few cents in compute. The math doesn't work.

Here's my new mental model for picking AI models:

Use the lighter model when:

• The task is straightforward (formatting, simple edits, quick lookups)
• You already know the answer and just need it written out
• The problem has one clear solution path

Use the heavier model when:

• You're debugging something complex
• The problem involves multiple interacting systems
• You've been going back and forth for more than 15 minutes without progress
• You need the AI to reason about architecture, not just syntax

The 15-Minute Rule

I'm adopting a new rule for myself: if I've been stuck on the same problem with a lighter model for more than 15 minutes, I switch to the most capable model available and crank the effort to max. No exceptions.

The cost difference between models is measured in cents. The cost difference in my time is measured in hours. It's not even close.

What This Means for Your Workflow

If you're using AI for development, whether it's NetSuite, Python, JavaScript, or anything else, don't let "model efficiency" become a trap. The most efficient model is the one that actually solves your problem.

Start light when the task is light. But the moment you feel that frustrating loop of almost-right answers? Escalate immediately. Your time is worth more than the token savings.

---

We help small businesses build smarter workflows with AI. If you're curious about how AI tools can save you time (and when to invest in the good stuff), let's talk.