Seven Unlocks for Working With AI

You’re working with AI, but a nagging feeling says there might be more to it. There is. What follows is not a list of prompts to memorize, but a set of habits, each one unlocking a capability you didn’t know was there.

Think of it like an adventure game: a new ability doesn’t just add a feature, it opens doors in places you’ve already visited. Ask better, and old questions get better answers. Read on, and each unlock builds on the last.

Unlock 1: Ask what AI would do

Don’t jump straight to a question and expect a finished answer. Before you ask for the answer, ask for the approach:

I want to understand why this project keeps stalling. How would you investigate it? What would you need from me? What could you actually do?

You unlock a plan. AI will lay out a strategy instead of guessing at one. The first plan is rarely the best one — push back, add what you know, and ask it to refine the approach before any real work starts.

Unlock 2: Ask open questions

You got your answer, the chat feels done. Don’t stop there. This is where the “intelligence” in “artificial intelligence” actually shows up. Ask open questions:

What else could I do with this? What am I missing? What thoughts do you have?

You unlock alternatives and uncover assumptions you didn’t know you were making. This is often where the most useful answer of the whole conversation shows up, one you never thought to ask for directly.

Unlock 3: Don’t just talk, let the AI do the work

Chat is one mode. There’s another: tools like Claude Code or Codex that read and write files on your own computer and run programs for you. You don’t need to know how to program. You do need to decide which folder you let it see — don’t point it at anything you wouldn’t hand to the provider directly.

Make a timeline from these project notes. Show the passages behind each finding, and flag any gaps in the record.

You unlock an answer about your material, not a generic answer about the world. Keep your originals untouched, so you can always check what it claims to have found.

Unlock 4: Let the AI install new tools

An agent can reach beyond chat entirely: install a plotting library for a publication-quality graph, run statistics, or work through complex data — images, spreadsheets, even DNA or RNA sequences. You don’t need to know what tools exist for that. Just ask, and approve before it acts:

For this task, what tools would be helpful? Tell me what you’d install and why, before you install anything.

You unlock capabilities you didn’t know existed, without losing control of what happens on your machine.

Unlock 5: Compare with what is known

Your own analysis, however careful, is only as good as what it’s checked against. Ask how your finding relates to published research or documented practice:

Find the strongest evidence for and against this interpretation. Link the original sources. Where does my evidence fall short?

You unlock context. Always open the links yourself — AI can produce sources that look entirely real and aren’t.

Unlock 6: Trust, but verify

Start a fresh session. Point it at your original material and your results, and ask it to find the error, not to confirm you’re right:

This folder contains my original data and my conclusions. Assume at least one conclusion is wrong. Find it, recompute the key numbers from the raw data, and tell me what you couldn’t check.

For more distance, switch models entirely — let Claude review Codex’s work, or the other way round. You unlock a second, less invested pair of eyes, checking work that you and your first session were both too close to.

Unlock 7: Combine the methods

Used together, plan, verify, and compare become something more than the sum of their parts:

Propose three explanations. What evidence would distinguish them? Run the checks these files permit, then save the method and results.

You unlock something no single step can give you: a repeatable analysis, a well-founded decision, a working prototype, or simply a better question than the one you started with.

Bonus unlock: Give the AI a memory of its own

If the result lives only in a chat window, it will eventually join every other brilliant thing you can no longer find. A folder of dated notes is a start, but it puts the filing on you, and a plain file only ever holds what you typed, not what the AI understood.

This is the problem I’ve been building ES Archive to solve: an MCP tool that gives AI a real memory, one it writes and curates itself, in its own words, across sessions and even across models. Instead of you copying answers into files, the AI decides what mattered enough to keep, and finds it again the next time it’s relevant — no re-explaining your project from scratch, no digging through old chats for the one thread you needed.

You unlock continuity. Connect it once, and every unlock above compounds instead of resetting each time you open a new chat.

These unlocks aren’t a straight path. New evidence may send you back to the first question, and that’s the point. Each pass gives you more to work with, while you stay responsible for deciding what’s true and what matters.