How to use AI to automate tasks
A practical playbook for using AI to handle the repeating work that drains your day — email triage, scheduling, reminders, follow-ups, and briefings. Frameworks, prompts, and real workflows you can steal.
What counts as a task worth automating
Not every task should be handed to an AI. The best candidates share three traits: they repeat (daily, weekly, or every time X happens), they follow a roughly predictable pattern, and the cost of a small mistake is low or reversible.
Good candidates: sorting incoming email, drafting replies to common requests, scheduling recurring check-ins, sending "still on for tomorrow?" nudges, summarizing long threads, compiling a morning briefing, logging tasks from conversations. Bad candidates: one-off legal decisions, terminations, anything with irreversible financial consequences.
The 3-step automation framework
Step 1
Observe your week for 3 days
Before automating anything, list every task you touched more than once. Circle the ones that took under 15 minutes but interrupted deep work. Those are your first candidates — small, frequent, high-friction.
Step 2
Describe it in one sentence
Write the trigger, the action, and the escalation in plain English. Example: "Every weekday at 7am, summarize my calendar and any unread priority email; if anything needs a reply today, draft it for me to approve." An AI assistant like Argus turns that sentence directly into a running routine.
Step 3
Start with approval, graduate to autonomy
Run every new automation in "draft for my review" mode for a week. Once you're approving 9 out of 10 outputs unchanged, let it send automatically. This is how you build trust without gambling.
7 workflows to automate first
These are the workflows Argus users automate in their first two weeks. Pick one or two — not all seven.
Morning email triage
A 7am summary of what came in overnight, sorted by whether it needs a reply, is just FYI, or can be archived. Drafts prepared for the ones that matter.
Meeting prep briefings
15 minutes before every meeting, get a one-paragraph brief: who's attending, the last thread you exchanged, open action items, and 3 talking points.
Task capture from chat
Anything you say to your assistant like "remind me to call the plumber Thursday" becomes a scheduled reminder without you opening a task app.
Follow-up nudges
A daily sweep of threads where you're waiting on a reply. The assistant drafts a gentle nudge for anything older than 3 days.
Scheduling handoff
When someone asks "when are you free?", the assistant proposes 3 slots that respect your working hours, existing focus blocks, and travel time.
Daily plan generation
Every morning, a proposed order for your day that fits meetings, tasks, and energy level. You approve or reshuffle in 30 seconds.
End-of-day recap
At 6pm, a summary of what you finished, what slipped, and what carries into tomorrow — automatically rolled into tomorrow's plan.
Copy-paste prompts and routines
Paste any of these into an AI assistant with email and calendar access. Adjust the times to your schedule.
Morning briefing
"Every weekday at 7:00am, send me a briefing with today's meetings, any prep I need, unread email that looks important, and the top 3 tasks I said I'd do today. Keep it under 200 words."
Reply drafting
"When a new email arrives from someone I've replied to before, draft a response in my voice and save it as a draft. Don't send anything without my approval."
Follow-up sweep
"Every Tuesday and Thursday at 4pm, find email threads where I sent the last message more than 4 days ago and the recipient hasn't replied. Draft a short follow-up for each."
End-of-day recap
"At 6pm every weekday, summarize what I completed today, what I didn't, and what should carry over to tomorrow's plan."
Building trust with an AI that acts
The reason most people abandon AI automation isn't accuracy — it's that they gave it too much power on day one, got burned once, and pulled the plug. Ramp up in layers:
- Read-only. Let the assistant see your inbox and calendar and produce briefings only. No actions.
- Draft-only. Allow it to create drafts, proposed events, and task suggestions — but everything sits in a review queue.
- Send-with-confirm. Route sends through a one-tap approval on your phone.
- Full autonomy on low-risk lanes. Auto-archive newsletters, auto-schedule internal check-ins, auto-send follow-ups to threads you've explicitly tagged.
For anything sensitive — client-facing email, external commitments, money — keep confirmation on indefinitely. Speed matters less than not embarrassing yourself. Read more about how Argus handles this in our security guide.
Common mistakes to avoid
- Automating everything at once. Start with one workflow and live with it for a week.
- Skipping the approval phase. Even a great assistant needs a week of calibration in your voice before it sends on its own.
- Vague prompts. "Help with email" produces nothing useful. "Draft a two-sentence reply thanking them and proposing Thursday at 2pm" produces exactly what you want.
- No escalation rule. Every automation needs a "when in doubt, ask me" fallback. Otherwise you'll get confidently wrong actions.
- Ignoring the recap. Read what your assistant did each day for the first two weeks. That's how you catch and correct patterns.
Frequently asked questions
What tasks can AI actually automate for me?
The highest-ROI targets are repeating, rules-based tasks with clear inputs: triaging and drafting email, scheduling meetings, sending reminders, preparing for calls, following up on threads, logging expenses, and compiling daily or weekly briefings. Anything you do more than twice a week is a candidate.
Do I need to know how to code to automate tasks with AI?
No. Modern AI assistants like Argus use plain-English routines and pre-built connections. You describe what you want in a sentence — 'every morning at 7am, brief me on today's meetings and unread priority email' — and the assistant runs it on schedule.
How is AI automation different from Zapier or IFTTT?
Zapier and IFTTT run rigid if-this-then-that recipes. AI automation understands context: it can read an email, decide whether it matters, draft a reply in your voice, and only bother you when a human decision is required. It handles ambiguity that traditional automation can't.
Where should I start?
Start with one workflow that costs you time every single day. For most people that's the morning email + calendar triage. Automate that first, live with it for a week, then add a second routine like meeting prep or end-of-day recap.
Is it safe to let AI act on my accounts?
Use tools with scoped OAuth permissions, no model training on your data, and an explicit confirmation step for anything that sends or deletes. Start read-only, then graduate to send-with-approval, then full autonomy for low-risk tasks.
Let Argus run these workflows for you
Argus is a personal AI assistant that connects to your email, calendar, and tasks — and runs the routines above on autopilot.