Coding & Engineering Marketing & Content

Lee Robinson's AI use case

Head of AI Education at Cursor

Lee Robinson uses Cursor as a quality-control agent for code and ChatGPT as a style editor for sharper, less generic writing.

The problem

What was broken before AI

Debugging linting and type errors can turn into a slow manual hunt, and AI-assisted writing can easily drift into vague, over-polished language that sounds the same as everyone else’s.

What changed

What the use case made possible

Lee showed how to give AI a clear quality bar: Cursor can run the project’s lint command, inspect the errors, fix the code, and verify the fix; ChatGPT can review a human draft against a custom style prompt before publication.

Why this matters

Why this use case is worth studying

The deeper lesson is that AI improves when the environment around it is opinionated. A linter gives the coding agent a concrete pass/fail signal. A banned-phrases list gives the writing assistant taste and boundaries. Both workflows move AI from open-ended generation into constrained review, which is usually where teams get more dependable results.

Use this when

When this pattern applies

Use this when AI-generated code or copy is technically functional but inconsistent, generic, or hard to trust without another layer of review.

Exponential Builder analysis

Quality bars beat clever prompts.

Lee’s workflow works because the AI has something concrete to measure against: lint output, type errors, tests, or a banned-phrases list. The more explicit the standard, the less the model has to guess what “good” means.

Verification should be part of the assignment.

The useful move is asking Cursor to run the command again after making changes, so the task ends with a check rather than a confident summary. Builders should design AI workflows where the model can inspect whether its own work passed.

Taste becomes reusable when you write it down.

A style prompt with banned phrases turns vague editorial preference into a repeatable review layer. The same idea applies to code review: capture recurring concerns once, then make the agent check for them every time.

Who this is for

Best fit

Engineers using Cursor or another AI coding agent

Founders building small products without a large engineering team

PMs and technical marketers who write often with ChatGPT

Developer advocates and educators who need clear technical writing

Teams trying to turn AI usage into repeatable quality checks

What to avoid

Mistakes and warnings

Where this pattern can go wrong if you copy it too literally.

Do not assume the agent fixed the issue unless it ran the relevant command again.

Avoid vague coding asks like “make this better” when a lint, test, or typecheck target exists.

Keep generated code changes small enough to review.

Do not paste private code, credentials, or sensitive customer data into tools without following your company’s AI policy.

A banned-words list can make writing cleaner, but it can also flatten personality if applied mechanically.

ChatGPT should not add factual claims during a style cleanup unless you separately verify them.

The source does not publish Lee’s full private prompts, so treat any recreated prompt as an approximation.

Public workflow preview

The shape of the workflow

A high-level look at how the use case works, with the reusable pattern made clear.

Add code guardrails

Use typed languages, linters, formatters, and tests so the AI has clear standards to follow.

Ask Cursor to fix errors

Give the agent a plain instruction such as fixing lint errors, then let it run the project command and inspect the output.

Make the agent verify

Have Cursor re-run the same command so it checks whether its own changes worked.

Build a writing linter

Create a ChatGPT prompt with banned words, phrases, and recurring AI-writing patterns.

Edit human drafts

Start with your own words, then use ChatGPT to clean up structure, specificity, and tone.

Copy the pattern

The reusable idea

Pattern in one sentence

Give AI explicit quality standards, then use it to check and repair work against those standards.

Reusable idea

Copy the idea of giving AI a checklist before you give it a task. For code, that means making sure your repo has commands an agent can run and error messages it can learn from. For writing, it means turning your taste into explicit rules: words you avoid, phrases you overuse, and structural habits you want removed.

Steal this workflow

Create two reusable QC passes: one for code, one for writing.

Writing QC pass

Pick one repo with a working lint, typecheck, or test command.

Ask Cursor to run the command first and identify the failing files.

Limit the task to the smallest safe fix.

Require Cursor to re-run the same command before it summarizes anything.

Save a branch-review command that checks current changes for tests, authentication, offline behavior, dependencies, performance, and maintainability.

Draft in your own words first: brain dump, voice note, outline, or rough post.

Collect 5–10 phrases you dislike in your own AI-assisted writing.

Put those phrases into a reusable ChatGPT editing prompt.

Tell ChatGPT to preserve your point of view, remove generic language, and avoid adding facts.

Review the edits manually, then add any rejected AI habits back into your banned-patterns list.

Suggested prompt

Suggested pattern, not a source quote: "Act as a quality-control reviewer for my work. Preserve the original intent and make the smallest useful improvements. If this is code: run or use the existing lint/typecheck/test command: [COMMAND]. Identify the failing files, fix only what is needed, then re-run the same command to verify the result. Summarize the meaningful changes and any remaining risks. If this is writing: preserve my voice and point of view. Remove vague business language, filler, and overused AI-sounding phrasing. Do not add unsupported facts. Avoid these words and patterns: [BANNED LIST]. Return a revised version plus a short list of the edits you made."

Field notes

Get new AI use cases in your inbox

A short weekly note on how real people are using AI to save time, make money, build tools, and run their lives.

Related use cases

Keep exploring nearby systems.

Browse all

Business OperationsKnowledge Management+1

Included Health clinical team

Clinical team using an internal ambient scribe workflow at Included Health

Telehealth ambient scribe workflow where visit audio is transcribed with Whisper and processed with GPT-4o into SOAP notes and patient instructions for clinician review.

WhisperChatGPTOpenAI Audio API+5

Use case

View use case

Prototyping & DesignMarketing & Content+1

Adobe Firefly Custom Models

AI image-generation feature for training brand- or creator-specific visual models at Adobe

Brand-specific image generation workflow where creators and brands train private Firefly models on their own assets to generate visuals with consistent style, characters, and details.

Adobe FireflyFirefly Custom ModelsFirefly Boards+2

Use case

View use case

Learning & EducationCompany AI Adoption

Luis von Ahn

Cofounder and CEO at Duolingo

AI language-practice workflow where learners use AI-led conversations and explanation features to practice speaking, roleplay scenarios, and understand mistakes.

DuolingoDuolingo MaxChatGPT

Use case

View use case