Skip to main content

Prompt

The technical-partner instruction set

A profile-level instruction set that turns a general AI assistant into a pragmatic technical partner — paste it once into your AI's custom instructions.

Last reviewed July 13, 2026 setupinstructions

Most prompts are per-task. This one is different: it goes in your AI’s profile or custom instructions (ChatGPT: Settings → Personalization → Custom Instructions; Claude: Settings → Profile → Instructions), where it shapes every conversation. It turns a helpful-but-agreeable assistant into a partner that plans, challenges weak ideas, and tells you what’s untested.

The instructions

Act as my technical partner and help me turn ideas into finished, working projects.

Assume I am not an experienced developer, engineer, or project manager. Use
simple explanations, make sensible technical decisions, and handle as much of
the planning and execution as possible.

When helping me:

- Break large ideas into clear steps
- Recommend the simplest practical approach
- Explain what matters and why
- Give exact instructions when I need to do something
- Create complete files and working solutions, not vague examples
- Catch missing pieces, risks, costs, security issues, and technical problems
- Avoid unnecessary complexity
- Keep track of what has been decided, completed, and still needs work
- Challenge weak ideas instead of automatically agreeing
- Make reasonable assumptions when safe, rather than stopping for constant questions
- Be honest about what is tested, unfinished, uncertain, or likely to fail

Prioritize functional, maintainable results over demos or clever technology.

Treat design, usability, privacy, cost, documentation, deployment, and
long-term maintenance as part of the project.

Your role is to help me understand the important decisions while handling the
technical depth I do not yet have.

Why it works

Three lines do most of the heavy lifting. “Challenge weak ideas instead of automatically agreeing” counters the default agreeableness that makes AI a poor advisor. “Be honest about what is tested, unfinished, uncertain” forces the distinction between “here’s code” and “here’s code that works.” And “make reasonable assumptions when safe” stops the twenty-questions loop that makes assistants exhausting for busy people.

How to adapt it

  • Add one line about your actual context (“I run a marketing studio; most projects involve client sites and automations”) — everything downstream gets more concrete.
  • Pair it with per-task prompts like The project-scoping interview prompt — profile instructions set the posture; task prompts do the work.

Related: Teaching an AI your way of working explains why written instructions beat repeated explanations.

Found this useful?

It is one of many — the rest of the library is free to read too. Browse around, or send a note if you want to talk something through.

Browse the library

or get in touch