Here's something that'll change how you use AI from today.
Most people treat AI like a one-shot tool. You type a prompt, you get an output, and you hope it’s good enough.
If it’s not, you tweak the prompt slightly and try again. If it’s still not right, you either settle for it or give up.
That's the one-shot approach. And it's why most AI output feels average.
The people actually getting useful results aren't writing better prompts.
They're building loops.
A prompt is one conversation. A loop is a system that keeps going until the output is actually right.
Here's how a loop works:
You give AI a clear spec of what you want built or written. The AI does the work. Then instead of handing it back to you and calling it done, it grades its own output against what you asked for, requirement by requirement, lists everything that failed, fixes it, and runs again. It keeps going until the output passes every single check you defined upfront.
You walk away. You come back to something that's actually right.
The reason most people have never done this is because nobody showed them how to set it up. It sounds technical. It isn't.
In Claude Code you build three skills that run the loop automatically.
One skill turns your idea into a detailed spec before anything gets built.
A second skill executes from that spec exactly as written.
A third skill grades the output against the spec line by line, names every gap, and sends it back to the build skill until everything passes clean.
The first time you run this and come back to an output that actually does what you asked, it feels like you hired someone competent.
Which is a feeling most people haven't had from AI yet.
Here's how to set it up:
Step 1 Build the /spec skill
Paste this into Claude Code:
Create a Claude Code skill called "spec". When I run /spec, interview me about the feature or app I want to build. Ask one focused question at a time until you fully understand the goal, the must-have requirements, the constraints, and what "done" looks like. Do not start building. When you have enough, write a clear, detailed spec and save it to specs/<name>.md. The spec must include: the objective, the exact requirements, the edge cases to handle, and a concrete definition of done.Step 2 Build the /build skill
Paste this into Claude Code:
Create a Claude Code skill called "build". When I run /build, read the spec in specs/<name>.md and build exactly what it describes. Do not add features, refactor unrelated code, or invent requirements that aren't in the spec. When you finish, list which spec requirements you covered so the review step can check them.*Step 3 Build the /review skill
This is the secret sauce.
Create a Claude Code skill called "review". When I run /review, compare the current build against specs/<name>.md. Go requirement by requirement and list every gap, bug, or missing piece, naming the exact spec item each one fails. If anything fails, write the specific fixes needed and hand them back so /build can address them. Only pass the build when every requirement in the spec is fully met.*Step 4 Run the loop
Run /spec and answer the questions. Then paste this and walk away:
Loop /build and /review: build from the spec, review the build against the spec, fix whatever fails, then repeat until the review passes clean. Keep going on your own until it passes.
Not using Claude Code?
Paste this at the end of any AI output you want to improve. Works in Claude, ChatGPT, anywhere:
Score your own work 0-100 against a clear rubric you write first. List the weakest parts and why they cost points. Rewrite to fix the top weaknesses while keeping what scored well. Repeat until the score stops improving by a real margin, then give me the best version and the score trajectory.
Cold emails
Proposals
Landing page copy
Ad copy.
Hiring posts
Any output that needs to be great rather than just acceptable.
Most people are on their first draft wondering why it doesn't convert. Loop users are on draft six before they even read it.
This is the kind of thinking we bring to every client at Build With Nav
We help businesses design and implement custom AI workflows that fit the way they already operate, so AI becomes part of the business instead of another tool sitting in a browser tab.
If you're curious what that could look like in your business, click here to apply.
Nav
