What an AI assistant is good at
The shape of work where it adds real leverage. Lean here.

#Writing & rewriting
The strongest use case is the boring middle 80% of writing โ emails, drafts, briefs, internal docs, code comments, summaries. A good assistant can match tone if given a sample, translate between languages, summarize a long thread, or turn rough bullets into prose.
- Draft an email to a vendor / client / family member
- Turn meeting notes into action items + a status update
- Rewrite an explanation for a different audience (technical โ casual)
- Translate between languages keeping voice
- Summarize a 50-message thread into 5 bullets
#Multi-step research + cross-referencing
Give it a research question and it will go look things up, cross-reference sources, and report back with citations. The good ones stay skeptical when sources disagree and flag uncertainty rather than smoothing it over.
- "Verify that this email address is correct" โ checks the official site
- "Find the 5 best vendors for X with EU shipping" โ comparison table
- "What changed in the new API release?" โ diff against last version
#Code, scripts, and small tools
Strong at small-to-medium codebases, refactors with clear specs, debugging from stack traces, and building one-off scripts to automate something. Best when the assistant can run + test the code locally and read the errors.
- Write a script that pulls X from API Y and writes to a spreadsheet
- Refactor this file into smaller modules with these responsibilities
- Add tests to this function
- Fix this stack trace (paste it in)
#Orchestrating long-running work
With a runtime like OpenClaw or Hermes under the hood, the assistant can run scheduled jobs, spawn sub-agents for big tasks, watch inboxes, post to chat channels, sync data between apps. Useful for things that would otherwise need a Zapier + custom-code combination.
- Run a 15-min brainstorm every night and surface the 3 best ideas
- Watch a database and ping when a new row lands
- Every 30 min, sweep inboxes, mark routine email read, surface what needs attention
- Generate a daily report end-to-end
#Personal-assistant tasks
- Create a calendar event + send invite
- Reply to a thread (with approval)
- Find a contact, draft outreach
- Surface what needs attention from the last 24h across email + calendar
- Remind you of something at a specific time / date
#Strategy + sparring partner
A good assistant will push back when asked to. Drop a half-formed idea on it โ "should I ship X this week or wait?" โ and a well-prompted one will list the pros, cons, what would need to be true to commit either way, and where it disagrees with the framing. Better than a yes-man.
The strongest use: pair-think. Talk through a decision with the assistant like a colleague. Don't ask for the answer โ ask it to help you see the angles you're missing.