Skip to content
AI development

Vibe coding that survives contact with users

AI coding agents are very good at producing something that runs. Everything that makes it survivable, review, tests, boundaries and written decisions, is still a human job.

What this covers

The practice around the tool

Prompts that do the job

Written for your codebase and your question, with the evidence requirements that stop an agent from agreeing with itself.

Review that catches what agents miss

Error handling, concurrency, migrations, and confident code that solves a slightly different problem than the one you had.

Keeping the architecture intact

Agents optimize locally. Somebody has to hold the shape of the system across a hundred local optimizations.

Decisions written down

So the next session, the next agent and the next developer all start from what was already settled.

How this shows up at each level

Prompts are a tool, not the product

  1. Advisory level

    I write or rework prompts for you and tell you what a good result would look like. You run them.

  2. Leadership level

    I prepare the instructions your team and your agents work from, and I read what comes out.

  3. Vibe Coding level

    I use Claude Code Max and Codex Pro myself, as working tools, with the same review discipline I would apply to a colleague.

  4. What you are not buying

    A prompt library. Generic prompts are free and everywhere. The judgment about which question to ask is the part that is worth paying for.

Where this helps

Not what this is for

Training your team to use AI tools from scratch

Choosing between AI vendors on price

Automating a business process that has nothing to do with code

What this is for

A product mostly written by Claude Code or Codex

Generated code that nobody has fully read

Agents producing more output than anyone can review

Wanting to keep the speed without keeping the mess

Start with the area that worries you most

Pick the level of involvement that fits, describe your product, and we begin there.