Execution or judgment? How to actually pick an AI effort level

Most AI tools now have a setting like Codex’s Effort dropdown: Light, Medium, High, Extra High, Ultra. Most people leave it on whatever it defaults to and never think about it again. Here’s the test I use, and what it actually looks like across a normal cross-functional workweek.

The test

Two separate settings are doing two different jobs. The model is the engine, that’s the raw capability, how powerful it is under the hood. Effort is how hard you press the gas for this specific question, independent of which model you’re using.

So the question isn’t “how important is this topic.” It’s: do I already know what I want, or am I asking the tool to help me actually think?

• You know what you want. You’re asking for execution: draft this, organize this, summarize this. Stay Light or Medium.
• There’s real ambiguity, no clean right answer. You need the tool to actually weigh things and help you land on a judgment. Go High or Extra High.
• It’s rare, and getting it wrong has real consequences. Save Ultra for that. It burns through usage limits faster, so it’s not a default, it’s a deliberate choice.

What that looks like across a cross-functional week

Light — quick, low-stakes, you already know the shape of the answer.

• A quick reply to someone on a different team
• Cleaning up the tone on a message to someone you don’t work with closely
• A fast summary of a document another team sent over

Medium — normal day-to-day execution, still no real ambiguity.

• Prepping for a meeting with people outside your immediate team
• Turning cross-functional meeting notes into next steps everyone can actually act on
• Drafting an update that has to make sense to people who don’t share your context

High — real judgment starts here. There’s a decision, not just a task.

• Working through a disagreement between two groups who both think they’re right
• Deciding how to frame a hard message to a team whose priorities differ from yours
• Thinking through a decision that affects people outside your own function

Extra High — same kind of judgment call, more moving parts at once.

• Weighing tradeoffs across multiple stakeholders who each want something different
• Untangling a problem that touches more than one function, with more than one plausible cause

Ultra — rare, and you’ll know it when you see it.

• Something where the outcome affects the business in a real, hard-to-reverse way

A decision template you can save

Before you touch the dropdown, ask yourself these two questions in order:

1. Do I already know what I want here, and am I just asking for help producing it?
   → Yes: Light or Medium.

2. Is there real ambiguity, no clean answer, and do I need help actually weighing it?
   → Yes: High or Extra High.

3. Is this rare, and does getting it wrong have real consequences?
   → Yes, and only then: Ultra.

Screenshot it, keep it next to your AI tool, and stop defaulting to whatever the dropdown last landed on.

Try it this week

Pick one task you’d normally throw at the default setting without thinking. Run the two questions above on it first. See if you land somewhere different than where you would have on autopilot. That’s the whole habit, one deliberate choice at a time, which is the same reasoning-partner approach the course AI, Finally is built around if you want the fuller version of this.