Most “AI assistant” content is a staged demo or a feature checklist. Nobody shows you the actual, unglamorous back-and-forth.
So here’s the opposite: the real conversation I had with alfred_ on a normal Monday, doing real work for my real company. Real names, real emails, real product mess. These are screenshots straight out of the app, with a note before each one so you can see what I was watching for. Lightly trimmed for length, nothing material changed.
The job I actually gave it
I had a note for my two engineers, Pranav and Dinesh, about what to focus on this week. The problem: it read like an engineering spec, exact fields, exact flows, exact UI. I wanted to hand them problems, not instructions. So I started here.
It read my note and told me the truth
This is where most tools fold. Ask them what they think and they flatter you. alfred_ said my note was good thinking wrapped in too much prescription, and it was specific about where.
I said don’t cut anything, and it kept all of it
I didn’t want a tidier list. I wanted the same problems, reframed. It stripped the how from each, folded the overlapping ones together, pulled out the two that were mine, and split the rest between the two of them, with an owner tag on each section.
The moment I’d normally cut from a demo
Being honest, since this is a real session: mid-edit, the note looked like it had vanished from my widget. That’s a heart-stop. alfred_ restored the full thing, but a restore isn’t the same as never scaring you, and I’m leaving it in.
It drafted, and it stopped
It drafted the email and stopped, nothing sent until I said so. Here’s the draft it composed:
Hey Pranav and Dinesh,
Wanted to share what I think we should be focused on this week. Our focus is to begin turning alfred_ into more human than software, starting with proactive reminders and notifications. Dinesh, please look at the Important to Watch step in the new onboarding process and the data we're capturing. I think this is the perfect chance to hook users into features without making them answer a questionnaire, so it just works without them knowing how.
From a product standpoint, Dinesh, you've been doing a great job and the product experience is truly great. The tasks for your area should only further that experience and create true delight for our users within SMS and web chat.
From a data and infrastructure standpoint, Pranav, you've really cleaned up Donna, which sets us up for big success. Let's wrap up Donna and go deeper on the infrastructure so we can start implementing the end-to-end tasks we talked about. I've attached a doc of what I think alfred_ could handle end to end. Go through it and flag which items you think we could genuinely tackle end-to-end, like we discussed last week.
Dinesh, your areas this week:
- Proactive Reminders
- Em Dashes
- Important to Watch
- Smart Reminders (includes Doctor's appointments and Flights as reminder contexts)
Pranav, your areas this week:
- Extract To-Dos from Donna (and make sure Donna is fully updated here)
- Auto-Add Events to Calendar
- Protect 2FA/OTP
Let me know if you have questions.
It caught a nuance I didn’t spell out
Then I asked what else was waiting. It read my cofounder’s threads and summarized them straight, no invented status. One was a half-formed product thought I’d been chewing on: notifications only ever say “To: me,” even when I’m just CC’d, and a group thread should read differently from a 1:1. It turned that into a clear, threaded reply to Pranav (with Dinesh on CC), the kind of judgment I’d expect from someone who had been in the meeting.
Hey Pranav,
One more thing for the notification UX thread: we need to surface recipient context better.
Right now notifications only show "To: me" even when I'm CC'd, which is confusing. If an email is to Dinesh and I'm CC'd, the notification should reflect that, not make it look like the email came straight to me. Whether you're the primary recipient or just copied changes how you read the urgency entirely.
A few thoughts on how this should work:
- If I'm in To, show "To: me"
- If I'm CC'd, show "CC'd" or "To: Dinesh, CC: me" so it's clear I'm not the primary recipient
- If several people are on the thread, don't list everyone, cap it at two or three names then "+N others"
- Group communication feels different from a 1:1, and the notification should make that obvious at a glance
Let me know what you think, happy to talk it through.
Then it closed the loop
Out of that whole notification thread, it pulled the concrete follow-ups, set priorities, and logged them so nothing falls through, the boring part that’s the first thing I actually drop.
What this actually proves
None of this is a single feature. It’s one continuous thread of judgment: reframing without losing substance, getting the people right, reading my inbox without inventing anything, catching a nuance I didn’t spell out, and never sending a thing on my behalf.
That’s the difference between a tool that drafts emails and an assistant you would actually hand your Monday to. This was a Monday.