Automate the tasks that repeat and quietly cost you attention: the status updates, the follow-up emails, the meeting prep, the "where are we on this" digging. When you've only got a few hours, a good chief of staff doesn't try to fix everything. They find the two or three recurring drains eating your week and hand those off. Think of AI as your junior chief of staff. Your job is to point it at the right work.
Most people do the opposite. They automate whatever's shiny, or whatever they saw in some video, and end up with clever tricks that save four minutes on a thing they do twice a month. That's not a rescue. That's a hobby.
Think like a chief of staff, not a power user
A good chief of staff has one job on a bad week: protect your time and attention. They don't ask "what's fun to optimize?" They ask "what keeps landing on your desk that shouldn't need you?"
Borrow that lens. Before you write a single prompt, look at your last five working days and find the tasks that are both frequent and low-judgment. Frequent means it shows up most days or every week. Low-judgment means a competent assistant could handle it with clear instructions. That overlap is where AI buys you real hours.
The four buckets to sort your week into
- Delegate to AI now — repetitive, rules-based, no real stakes if a first draft needs a tweak. Drafting replies, summarizing threads, turning notes into a task list.
- Delegate with a check — AI drafts, you approve. Client emails, anything with your name on the outcome.
- Keep for yourself — high-judgment, relationship, or strategy work. This is what you're freeing time for.
- Delete or defer — the stuff you're only doing out of habit. A chief of staff kills these quietly.
Sort honestly and you'll usually find your "delegate to AI now" bucket is small but mighty. Three or four task types, done constantly. Those are your targets.
The step-by-step triage
Step 1: Empty the week onto one page
List every recurring task from the past week. Not projects — tasks. "Reply to the vendor," not "vendor relationship." Keep it granular. You're hunting for the small repeating motions, because those are what quietly swallow a day.
Step 2: Mark frequency and judgment
Next to each item, note how often it happens and whether it needs your specific judgment. A quick two-column pass does it. The items that are high-frequency and low-judgment float to the top. That's your automation shortlist.
Step 3: Write one clear instruction, not a clever one
For each shortlisted task, aim for a prompt that captures how you want it done, so the output barely needs editing. Bake in three things: the context (who it's for, what tone), the format you want back, and the guardrails (what to never do). The difference between a prompt that saves time and one that creates cleanup is specificity about your standards. Vague in, vague out — and then you're doing the task anyway, plus reading the AI's version first.
Step 4: Run it, then fix the prompt, not the output
The first run will be roughly right but not perfect. Don't just edit the answer and move on. Edit the prompt so next time it's better. This is the part people skip, and it's the whole game. A prompt you improve once and reuse fifty times is an asset. A one-off correction is just more work.
Step 5: Stop at three or four
You're doing a rescue, not a rebuild. Get three or four high-frequency tasks running well and you'll feel the week loosen. Add more later, once these run on autopilot.
The honest shortcut
Writing tight, reusable prompts for your top time-drains is the slow part — you're basically teaching an assistant your standards from scratch. If you'd rather start from prompts already built for the common rescue tasks (thread summaries, follow-ups, meeting prep, turning a messy brain-dump into a plan), that's exactly what our 5-Hour Rescue AI Micro-Automation Prompt Pack is for. It's the shortlist done and worded, so you spend your few hours running the rescue instead of designing it. Worth it if your bottleneck is time, not curiosity.
What good looks like a week later
You should notice the same tasks landing and not flinch, because they now take two minutes of review instead of twenty of doing. The chief-of-staff test is simple: has your "keep for yourself" bucket actually gotten more of your attention? If yes, the triage worked. If you just made busywork faster, revisit which tasks you picked.
FAQ
How many tasks should I automate in one sitting?
Three or four. Pick the highest-frequency, lowest-judgment tasks and get them genuinely reliable before adding more. A few solid automations beat a dozen half-working ones you don't trust.
What if I'm not technical?
You don't need to be. This all happens in plain language inside a normal AI chat tool. The skill is describing how you want a task done clearly — closer to briefing an assistant than to coding.
How do I know a task is safe to hand to AI?
Ask whether a bad first draft would cause real damage. If a quick review catches any issue, it's safe to delegate. If a mistake would go straight to a client or cost money, keep it in the "draft, then approve" bucket.
How is this different from just using AI here and there?
Ad hoc use saves minutes. Triage saves hours because you deliberately target the recurring drains and reuse the same tuned prompts every time. It's the difference between a trick and a system.