How to pick your AI tool stack with Claude

What this audit actually does

Most tool advice starts from the tool and works backwards to you. This starts from your week and works forwards. You describe how your last seven days really went, Claude groups the repetitive parts into a small number of jobs, and then each job gets exactly one primary tool and one backup. Everything else on your list becomes a deliberate no, which is the part that gives you your attention back.

The whole thing takes about twenty minutes in one Claude session. You do not need a paid plan to run it. Claude's free tier handles all four prompts, and Projects (where you will save the finished stack doc) are available to everyone, with free accounts able to create up to five (Claude help centre).

The four-step run

1. 👩‍💻 Open Claude and start a fresh chat. Go to claude.com or open the Claude app and sign in. A brand new chat matters here, because you want the audit reasoning to be clean rather than tangled into whatever you were working on before.
2. 🧠 Dump your real week. Run prompt 1 below. Be unglamorous about it: the client emails, the thumbnail you remade four times, the meeting you re-typed notes from. The audit is only as good as the honesty of the input, and vague inputs produce a vague stack.
3. 💡 Get three real tools per job. Run prompt 2. You are asking for a shortlist with reasons and costs, not a list of names, and you are explicitly asking Claude to flag where it is unsure so you can check the pricing page yourself.
4. 🌱 Write the stack doc and save it. Run prompt 3, then save the output somewhere you will actually see it. Prompt 4 is the one you run in thirty days to keep the stack from quietly growing back.

Prompt 1: audit my actual week

This is the one people mean when they ask for "the audit prompt". It turns a messy description of your week into a small set of named jobs, ranked by how much of your time they eat.

You are helping me design a deliberately small AI tool stack. I want to start from my real week, not from a list of tools.

Here is my last seven days, written out honestly: [describe your week in plain language. Include the tasks you repeated, the ones you dreaded, the ones that took longer than they should have, roughly how long each took, and what you currently use for them, including "nothing" and "manually".]

Do this in order. First, group everything I described into no more than six repeating JOBS, where a job is a category of work that recurs (for example: writing, research, images, video, transcription and notes, admin and automation). Name each job in my words, not in tool-marketing language. Second, for each job estimate how many hours a week it costs me and how much of that is genuinely repetitive versus genuinely judgement-based. Third, rank the jobs by how much time I would get back if that job were half as slow. Finally, tell me which one or two jobs are worth solving with a tool right now, and which ones I should deliberately leave alone this month because the time saved would not justify learning another app.

Ask me up to four clarifying questions before you answer if my description is too vague to rank properly.

Prompt 2: shortlist three real tools per job

Run this straight after prompt 1, in the same chat, so Claude still has your jobs in context. The instructions about pricing honesty and knowledge cutoffs matter, because tool pricing changes constantly and a confident wrong price is worse than a gap.

Now shortlist tools for the top jobs you identified. For each job, give me exactly three real, currently available tools, ordered from the one you would recommend to a beginner down to the specialist option.

For every tool include: what it is genuinely best at in one sentence, what it is bad at or where people get frustrated with it, whether it has a real free tier or only a trial, and the entry paid price if you know it. Where a tool overlaps with something I already pay for, say so directly and tell me which one to keep.

Be honest about your own limits. Pricing and free tiers change often, so mark any price you are not confident about as "check the pricing page" and give me the URL rather than guessing a number. If web search is available to you, use it and cite what you find.

Then give me the one-line rule for choosing between the three: what would have to be true about my situation for me to pick option two or three instead of option one. End with the tools I should NOT add this month, and why.

Prompt 3: write my stack doc

This is the output you keep. The point is not the list, it is that the decision is made and recorded, so tomorrow morning you open a document instead of opening a comparison article.

Write my personal AI stack doc based on everything above. Write it as a document I will re-read in three months, not as a chat answer.

Structure it like this. A one-paragraph summary of how I work and what my stack is optimised for. Then one section per job containing: the job in my words, my primary tool, my backup tool, the specific trigger that tells me to reach for it, and the monthly cost. Then a short "deliberately not using" list naming the popular tools I have decided against and the one-line reason for each, so future me does not re-litigate it. Then a total monthly spend figure. Finally, three things that would make me revisit this stack, written as specific triggers rather than vague conditions.

Keep it plain and specific. No marketing language, no hedging, no "it depends". Then create it as a downloadable document so I can keep it outside this chat.

Where to keep the stack doc so it survives

1. 📁 Make a Claude Project for it. In Claude, open Projects, create one called "my stack", and put the stack doc in the project knowledge. Every chat you start inside that project already knows your tools and your constraints, so you stop re-explaining your setup. Projects are available on all plans, with a maximum of five on free accounts.
2. ✍️ Paste the summary into your Claude settings too. Under Customize, add a short line about how you work and what you use, so even chats outside the project stop suggesting tools you have already ruled out.
3. 📌 Put the total monthly figure somewhere you see it. The single most useful number in the whole document is the spend total. When it moves, you notice, which is the entire mechanism that stops a stack from creeping.

Six jobs, three tools each

This is the shortlist I would hand a beginner today, with honest pricing. Treat it as a sanity check against whatever Claude gives you, not as the answer. Prices move, so the links go straight to each pricing page.

How to actually close the other tabs

The audit only pays off if something leaves. This is the sequence that makes it painless, and the reason it works is that nothing is permanent: every step is reversible, so you are never risking anything by trying it.

• 🌱 Cancel, do not delete. Cancelling keeps your account and your files until the end of the period. You lose nothing by cancelling a month early and resubscribing later if you were wrong, and you save a year of quiet renewals if you were right.
• 💡 Screenshot the ones you are unsure about. Most of what you are attached to is one or two outputs, not the tool. Export those, then cancel.
• 👩‍💻 Unsubscribe from the newsletters that sell you tools. Tool directories and roundup newsletters are usually affiliate-monetised, which means their incentive is a longer list, not a shorter one. That is not a conspiracy, it is a business model, and knowing it changes how you read them.
• 🧠 Give the winners one real task each this week. A tool you have used once on real work is worth ten you bookmarked. This is also how you find out that your shortlist was wrong, which is useful and cheap to fix now.

Prompt 4: the review that stops the creep

Put a recurring reminder in your calendar for a month from now and run this in the same Claude project. It takes about four minutes and it is the difference between a stack and a collection.

Here is my stack doc from last month, and here is what actually happened since then: [paste the stack doc, then describe honestly which tools you opened, which you never touched, what you added, and what frustrated you].

Review it like an operator, not a fan. For every tool, tell me whether the evidence supports keeping it, and be specific: did I use it on real shipped work, or did I only intend to? Name anything I added since the last review and ask whether it duplicates something I already had, because duplication is where the spend hides.

Then update the stack doc: same structure, revised choices, revised monthly total, and a one-line changelog at the top explaining what changed and why. If nothing should change, say so plainly instead of inventing improvements. Finally, name the single biggest remaining time sink in my week and tell me whether it is actually a tool problem or a process problem.

Two things worth knowing before you trust the output

Proof

Harvard Business Review, how much time we waste toggling between applications
Claude Projects, plan availability
Create and edit files with Claude, available on all plans
Claude pricing