If you've used Claude Code or a similar coding assistant recently, you've probably seen a setting called effort level, with an option at the top called ultra (sometimes labeled max or high, depending on the tool). It's easy to scroll past. It's also one of the more useful dials you have when you're using AI to write or fix code for your business.
What Ultra Code Effort Actually Does
Effort settings control how much the model thinks before it answers. On a low or default setting, the model takes a fast pass: read the prompt, generate a reasonable answer, move on. That's fine for simple stuff, like writing a short script or fixing an obvious typo in your code.
Ultra effort tells the model to slow down. It reasons through the problem in more depth, considers more edge cases, checks its own work more carefully, and often runs through multiple internal passes before giving you the final answer. Think of the difference between someone glancing at a bug and someone who sits down, reads the whole file, traces through the logic, and tests a few scenarios in their head before touching the keyboard.
The tradeoff is real: ultra effort takes longer to respond and costs more (in tokens, in API costs, or in your monthly usage allowance, depending on how you're billed). You're paying for more thinking, literally.
When Ultra Effort Is Worth the Cost
The rule of thumb is simple: match the effort to the difficulty of the problem.
Use ultra effort when:
- You're debugging something gnarly, like a bug that only shows up intermittently, or one that involves multiple systems interacting in a way that's hard to trace
- You're asking the model to design something from scratch, like the architecture for a new feature or a data pipeline, where a bad early decision gets expensive to fix later
- You're working with a large, unfamiliar codebase and need the model to actually understand how the pieces fit together before it changes anything
- The cost of a mistake is high, like code touching customer data, payments, or anything that runs in production without a human double-checking it first
In these cases, a few extra seconds (or a few extra cents) is nothing compared to the cost of shipping a bug that takes your team half a day to track down later.
When You Don't Need It
Most day-to-day coding work doesn't need ultra effort. If you're asking for a boilerplate function, a simple script, a small tweak to existing code, or anything where the answer is pretty obvious once you see it, the default or a medium effort setting will get there just as well, faster and cheaper.
Running everything on ultra effort is a bit like hiring your most senior, most expensive engineer to fix a broken link on your website. They can do it. It's a waste of their time and your money.
A good habit: start on a lower effort setting by default, and only bump it up to ultra when you notice the model is struggling. Signs it's struggling: it gives you an answer that clearly doesn't handle an obvious edge case, it contradicts itself, or it produces code that doesn't actually run. That's your cue to rerun the same prompt on ultra and let it think harder.
How This Plays Out for a Business Owner Who Isn't a Coder
You don't have to be technical to use this well. If you've got a developer, or a contractor, or you're using an AI coding tool yourself to build something in-house, the same logic applies: save the expensive setting for the expensive problems.
Say you're building a small internal tool to automate invoice processing. The part where you're generating a form or writing a simple export function doesn't need ultra effort. The part where you're deciding how the tool handles a duplicate invoice, or what happens if the payment API times out halfway through a transaction, that's where you want the model to actually think it through. Get that logic wrong and you're not looking at an ugly button, you're looking at double-billing a customer.
The setting exists because not all coding problems are equal, and treating them all the same wastes either time or money, sometimes both.
Talk to Level Up AI
If you're building or fixing tools with AI and want a second set of eyes on where you're spending too much (or too little) effort, we're happy to talk it through. Reach out to Level Up AI and we'll help you figure out what actually needs the expensive setting, and what doesn't.