Most descriptions of AI agents are either abstract ("autonomous task execution") or fantasy ("it handles everything so you can sip cocktails"). Neither is useful.
Here's what actually happens in 24 hours with a Stomme agent. This is a real workflow, based on a Personal tier setup for someone who runs a small consultancy.
06:45 — The briefing lands
Before the alarm goes off, your agent has already:
- Scanned your inbox (23 new emails overnight)
- Sorted them: 4 need a response, 6 are informational, 13 are noise
- Checked your calendar for the day: 3 meetings, one conflict
- Pulled together a summary
You open Telegram over coffee and read:
Morning briefing — Tuesday
Action needed:
- Client invoice from Andersson & Co (€4,200) — attached, awaiting approval
- Meeting request from Lena Berger — conflicts with your 14:00, suggested rescheduling options below
- Contract draft from legal team — review requested by Thursday
- Newsletter subscription renewal — expires Friday
Meetings today:
- 10:00 — Weekly team sync (Google Meet)
- 14:00 — Project review with Marcus (in person) ⚠️ conflicts with Lena's request
- 16:30 — Quick call with accountant (15 min)
Calendar note: You're free from 11:30–13:45. Suggested: use for contract review.
Two minutes of reading. Your day is mapped.
07:30 — Email triage continues
You reply to the briefing: "Approve Andersson invoice, reschedule Lena to Thursday morning, and draft a reply to the legal team saying I'll review the contract by Wednesday."
Your agent:
- Confirms the invoice approval (flags it in your inbox, doesn't send money)
- Sends Lena three Thursday morning options from your calendar
- Drafts a reply to legal: professional, specific, in your writing style
The draft appears in Telegram for your approval. You read it, change one sentence, reply "send." Done.
09:15 — Research while you prep
You're heading into the weekly sync and remember you need competitor pricing for a slide. You message: "Find current pricing for the top 5 project management tools for small teams. Table format."
By the time you've poured your second coffee, the research is back:
Tool Free tier Paid (per user/mo) Notes Asana Yes (limited) €10.99 Popular, feature-heavy Monday.com Yes (2 users) €9/seat Visual, marketing-friendly Linear Yes (limited) €8/user Developer-focused Notion Yes (limited) €8/user Flexible, docs + PM ClickUp Yes (limited) €7/user Everything-app approach
Sourced from current websites, not training data. You paste it into your deck.
10:00 — Meeting prep
Two minutes before the weekly sync, your agent sends a quick note:
Meeting prep — Team sync
Last week's action items (from your notes):
- Marcus: deliver revised timeline ✓ (received Thursday)
- You: approve design mockups ✗ (still in inbox)
- Anna: client feedback summary ✓ (shared Monday)
Suggested talking point: Design mockups are 3 days overdue from your side. Might want to address that.
It's not telling you what to do. It's making sure you're not blindsided.
12:00 — Document work
After the sync, you forward a 14-page PDF — a partnership proposal from a potential collaborator. You message: "Summarise this. Highlight anything that looks unusual in the financial terms."
Eight minutes later:
Summary: Nordic Digital Collective — Partnership Proposal
They're proposing a 60/40 revenue split on joint projects (60% to them), with a 12-month exclusivity clause for the Nordic market. The revenue terms are standard for this kind of arrangement, but the exclusivity scope is broad — it covers "all digital consultancy services," not just the joint project.
Flag: Section 4.3 defines "competing services" very widely. Under this clause, your existing client work with Andersson & Co could technically be considered a conflict.
Recommendation: Worth having legal review Section 4.3 before signing. The revenue split itself is fair.
You didn't ask for a recommendation. Your agent learned from previous interactions that when you ask it to flag unusual terms, you want an opinion too.
14:30 — Quiet hours
Your agent knows you're in a meeting (it read your calendar). No messages sent during 14:00–15:30. Emails still triaged in the background, but nothing interrupts you.
When you come out, there's a brief catch-up:
While you were out:
- Lena confirmed Thursday 09:30 ✓
- 2 new emails flagged for response (neither urgent)
- Newsletter renewal reminder — do you want to renew? (€79/year)
16:45 — End of day
After your accountant call, you message: "Anything I need to deal with before tomorrow?"
End of day check:
- Contract review for legal — due Wednesday (you have a 2-hour block tomorrow morning)
- Design mockups — still unapproved. Want me to draft an apology note to the team?
- No urgent emails remaining
Tomorrow's schedule: 1 meeting (11:00, client call). Otherwise clear.
You approve the apology draft (it's two sentences, honest, no excuses). Day done.
23:00 — Overnight
You're asleep. Your agent isn't.
- New emails arrive and get sorted
- A calendar invite for Friday comes in — your agent accepts it (it learned your preference: accept unless it conflicts)
- A research task you queued yesterday ("find three potential venues for the December team retreat, Stockholm area, budget under 15k SEK") finishes running
- Tomorrow's briefing starts compiling
What this isn't
This isn't autopilot. Your agent didn't send a single external message without your approval. It didn't make a financial decision. It didn't access anything you didn't give it permission to access.
What it did: it read, sorted, summarised, researched, drafted, and scheduled — the work that fills the gaps between the work that actually matters.
That's 3–4 hours of admin handled. Not in theory. In practice, on your Mac, on a Tuesday.
Stomme AI agents run locally on your Mac with cloud reasoning via Anthropic's Claude API. Your files and conversations stay on your device. Learn more about how we handle your data →