A list for spare tokens

Sometimes a very practical question occurs to me:

once we pay for AI / agentic tools, the tokens or allowance are a kind of resource. If they aren’t used up this cycle, they expire.

So is leaving them unused a kind of waste?

My answer now is:

yes. But don’t force yourself to use AI just to “use up the allowance.”

A better way: keep a “list of low-risk background tasks” for yourself.

These tasks have a few things in common:

they aren’t urgent; they don’t need decisions right away; one wrong step by AI won’t cause big losses; but over time they reduce chaos and save future attention.

These days I spend spare tokens on these five kinds of things.

1. Digital housekeeping

For example, cleaning up Google Drive, Downloads, the desktop, screenshots and project folders.

But the point isn’t letting AI delete things automatically. A safer way is to have it do a read-only inventory first.

You can use this prompt directly:

Please do a housekeeping pass on this folder. Read-only analysis: don’t delete, move or rename anything.

Goals:

1. find files that could probably be deleted

2. find files I should archive

3. find files that look sensitive

4. find files that need my judgment

Please output a table: file name / location / type / why it’s flagged / risk level / suggested action

I think cleaning up folders is a great use of spare tokens. Because it’s important but not urgent, and so easy to keep putting off.

2. Organizing your knowledge base

Many notes, PDFs and Google Docs aren’t worthless; they’ve just sunk out of sight.

You can have AI help you index them:

Please help me organize these notes/documents. Only summarize and categorize.

For each one, please output:

1. what this content is about

2. in what situations it might be useful later

3. whether it’s worth archiving

4. whether it could become an article, a project, résumé material or a case study

5. suggested tags

These tasks don’t necessarily produce anything that day. But they make material you couldn’t find, couldn’t remember or couldn’t use become usable again.

3. Project retrospectives

Many projects don’t fail because you can’t do them, but because once they stall, the cost of getting back in is too high.

You can have AI look at a project folder, chat history or product experiment, and ask:

Please do a project retrospective for me.

Output:

1. where the project stands now

2. what’s been done

3. where it’s still stuck

4. what the biggest risk is

5. if next time I only have 30 minutes, what the smallest next step is

6. what’s worth archiving or writing up as a case study

This one is especially good for spare tokens. Because it lowers the future cost of restarting a project.

4. Read-only codebase audits

With a coding agent, you don’t always have to have it change code.

Spare allowance can go to read-only checks:

Please do a read-only audit of this codebase. Don’t modify any files.

Please check:

1. TODO / FIXME

2. dependencies that may be outdated

3. duplicated logic

4. gaps in testing

5. places where the docs and code don’t match

6. suspicious large files or temporary files

Please output a list of candidates by priority; don’t make changes directly.

That’s much safer than “just optimize my code a bit.”

5. Repurposing content

I think this is the most easily underrated kind.

We have lots of chats, notes and experiment records that could actually become:

RedNote drafts; Substack posts; résumé project material; portfolio case studies; personal retrospectives.

You can use this prompt:

Please extract reusable content from the record below.

Output:

1. 3 topics that could become RedNote posts

2. 2 angles that could become Substack posts

3. material that could go into my résumé / portfolio

4. which parts are just process notes and not worth publishing

5. which one you recommend writing first, and why

A lot of the time it isn’t that there’s no content, but that it hasn’t been shaped into something publishable.

My judgment now is:

spare tokens aren’t well suited to “forced generation.” They’re better suited to low-risk maintenance like:

organizing, taking inventory, archiving, reviewing, auditing, extracting, generating candidate lists

Especially when files, photos, privacy or code are involved, I stick to one principle:

AI can make the candidate list first, but the right to delete and the final decision stay with a person.

A fairly safe process:

• In the evening / when idle: AI does a read-only inventory
• The next day: I confirm what to delete and what to archive
• After that: AI executes the confirmed operations in small batches

So I no longer see “unused AI allowance” only as waste. It’s more like a block of time you can use for background maintenance.

Don’t use it just to use it up. But if you already have a pile of important but non-urgent organizing to do, agentic tools really are good at doing a first round for you.

In one sentence:

spare tokens are best spent reducing future cognitive debt, not creating new to-dos.