Stop keeping up with AI. Build the system that keeps up for you.

The idea

The setup is one project that knows you: your role, your organisation, the tools you already pay for. Have the AI interview you, drop in the strategic plan and last year's impact report, and that context becomes the engine everything else gets read against.

Then the habit. Anything you come across, a video, a release note, a new model, goes in with one question: based on everything you know about my work, my tools and my role, does this apply to me, and where would it give me value?

The best answer is often no. Do not bother watching it, there is nothing in there you do not already do: 45 minutes handed back. And when something does matter, the system does not summarise it. It walks the idea into your own work, which no summary ever did.

Four honest verdicts. One of them is ignore. Every source gets read against your work rather than against the general public, and comes back with a decision: ignore, park, test or adopt. The ignores are what make the system pay for itself.

A summary tells you what was said. This does something else. It takes the idea and walks it into your week, your tools and the report you already have to write.

How it works

1. Make the learning projectOne folder, on your computer or in your AI of choice. This is the container the habit lives in.
2. Get interviewedLet the AI build the context by asking you questions (role, organisation, tools, tech stack) rather than you writing documents about yourself.
3. Feed it what you findTranscripts, articles, release notes. The system reads them against you, not against the general public.
4. Force a decisionFour honest verdicts: ignore when it adds no value now; park when it may matter later; test when it looks useful but is not proven in your work; adopt only when the test has already earned a place in your routine.
5. Run the smallest safe testIf the verdict is test, ask for one contained next step. Name the real task, what success looks like, how much effort you will allow, what evidence you need and which risks still need a human check.
6. Package what workedOnce the result is right, ask the AI to turn the finished process into a repeatable prompt or skill. Save the method you refined, not the first prompt you happened to type.
7. Schedule the stable partIf your tool supports scheduled tasks, let it find or review new material on a sensible cadence and run the same decision process. Keep the final decision and any action with a person.
8. Share what workedThe organisational version is a loop: someone finds something, shares it in the team channel, and everyone runs it through their own system. A self-reinforcing system.

The loop

Work through one real source. The loop only keeps what you choose to keep, and your answers stay on this device.

1. ContextReview what it knows. Correct what it assumes. Add what is missing. Tap a chip and write one line.
2. DecidePick what arrived, answer honestly, and the loop tells you what it has earned. A useful answer can be ignore.
3. TestYou review the plan. AI does not run the test or change your tools. Thirty days is a ceiling, not a requirement.
4. ReuseA skill is a saved set of instructions for one repeatable job. Start with a prompt that worked, name the steps and checks that should stay the same, save them as a skill or project instruction.
5. ReviewRun the test and collect real evidence. Choose. Update the system only with what you approve.

Using the information already available in this project or workspace, create a Personal AI Context Profile that will help you evaluate new AI tools and techniques against my real work.

Include my role and organisation, current priorities and projects, recurring tasks and workflows, tools and systems, constraints, policy or privacy boundaries, working preferences, and success measures.

Separate confirmed facts from assumptions. Flag anything uncertain or outdated. Ask no more than five questions to fill the most important gaps. Do not invent details. Show me the draft before saving it.

You have context about me and you are working inside my self-learning system. Analyse the source I have provided against my role, priorities, recurring tasks and current tools. Do not just summarise it. Tell me what is relevant, which tasks it could improve, whether a tool I already use covers the same need, and what evidence, effort or risk I should check. End with one recommendation: ignore, park, test or adopt. Explain why in plain language. If the answer is test, suggest one small, safe next step. Point me to the exact timestamps, pages or sections worth reading when the source provides them.

Turn this recommendation into the smallest useful pilot for my real work. Define the task I will test it on, the current way I do that task, what I will change, what success looks like, what evidence I should collect, the time limit, the risks and the points where a human must review the work. Keep the scope small enough that I can stop without disrupting anyone. End with the decision I should make after the pilot: discard it, change the test, keep it for occasional use or adopt it into my routine.

Review this conversation from the original task through to the final result. Turn the process that worked into a reusable prompt or skill. Include its purpose, when to use it, the context it needs, the steps to follow, the required human checks, the output format and one example of what good looks like. Remove dead ends and one-off details. Ask me to confirm the final version before treating it as reusable.

Help me set up a weekly scheduled review for this topic. Find only new material from credible sources, run it through my ignore, park, test or adopt decision process, and return no more than five items. For each item, give me the source, why it matters to my role, the task it could improve, the evidence or risk I should check and one recommended decision. Do not take external action or change my files. End by asking which recommendations I want to keep, test or discard so the next review can improve.

The sentences to type

Run both inside a project that holds context about you.

Based on everything you can find in this project, I'd like you to coach me in adopting AI. Ask me one question at a time until you have enough context about who I am and the work that I do. Then use everything you have access to to design a personalised AI adoption plan for me and my organisation.

This document is how I've been using AI. Pick up the patterns, pick up the gaps, and tell me what I should be doing next, and which one repetitive workflow in here is most worth turning into an automation.