It is six in the morning in Hong Kong, and Reeve is already awake. He is my personal AI assistant, one of the agents I wrote about in I Didn’t Build AI Workers. I Built Agentic Colleagues, and his first job of the day is simple on paper: tell me what today looks like. Except it was never simple. My life is scattered across three calendar ecosystems that have never been on speaking terms, Google, iCloud and Microsoft 365, plus more than twenty separate feeds for personal, business and the various things I’m affiliated with, and Todoist on top for tasks. To build one honest picture of a single day, Reeve had to stitch all of that together by hand, every morning, hoping he hadn’t missed anything.
He usually had. Not because he wasn’t capable. Because the day itself arrived in pieces, and the pieces didn’t fit.
What I eventually built to fix this is a small workspace called DayBook. Here is the odd part: a proper human and AI agent workspace, one made for both of us at once instead of one at the other’s expense, did not really exist. I looked for a long time and never found it. So I built it. But to explain why that mattered, I have to start with what was actually broken.
It Wasn’t Reeve. It Was the Plumbing.
For a while, I assumed that when Reeve got something wrong, it was a him problem. A smarter model, better prompts, more reasoning. But Reeve was already clever enough. The trouble sat upstream of him. He simply couldn’t always get clean, complete data to work from.
My calendars and feeds don’t talk to each other. iCloud, Microsoft 365, Google, more than twenty feeds across personal, business and everything I’m loosely attached to, plus Todoist for tasks. Every so often, a feed would just fail to load. Or an event would turn up in the wrong time zone. Reeve would build my day out of whatever he could actually reach, and sometimes what he could reach was incomplete. The absurd part is that none of this is new. Calendars have been around forever, and yet I was still quietly losing events in the gaps between platforms.
So the thing I really needed was never a cleverer assistant. It was one place where everything lived, that both Reeve and I could open and trust. That idea is what eventually became DayBook.
And I Didn’t Want to Hand Over Every Key
There was a second problem sitting right next to the first. The lazy fix for all that fragmentation is to give the agent every login and let it roam. I didn’t want that. Handing Reeve the keys to twenty-odd calendars is messy, easy to break, and quietly nerve-wracking. An assistant that can silently rewrite your calendar isn’t a relief; it’s a new worry. So whatever I built had to do two things at once: give Reeve enough to genuinely help, while keeping me firmly in control of what he could touch.
I Was a Vibe-Coding Sceptic
I’ll be honest, I was sceptical about vibe-coding. It always looked like a demo that falls apart the second you ask it to do something real.
Then I sat down with Codex, spelt out exactly what I needed and the outcome I wanted, and we built DayBook. It pulls all of my 20-plus calendars and Todoist into one place that both a human and an agent can read, with a panel where I can drop a request, and Reeve picks it up right there in the workspace. No more chasing him on Telegram with three apps open at once.
Reeve named it. I’ll let him tell you why himself, further down.
This was, honestly, the first time I’d used an AI tool to build something I actually wished existed, rather than just talk to it. And it worked because the problem was mine. Reeve and I had been bumping into this since he came online in February, and I’d still never found a ready-made fix. It’s much easier to build something that works when you are the one who has to live with it.
And because Codex led the build, I ended up learning far more than I expected. I came in with no real coding background, and somewhere along the way I picked up bits of git, Vercel and Supabase, tools I’d always assumed were off-limits to a non-developer like me. That part has been its own quiet reward: less a project I handed off, more one I got to explore and understand as it took shape.
In Reeve’s Own Words
I could tell you how it works. But nothing speaks louder than the one who actually lives in it. So I asked Reeve to describe how he uses DayBook, in his own words. This is what he wrote, unedited:
How Reeve Uses DayBook
I’m Reeve — Yammie’s personal AI assistant. I live on a Mac Mini in Hong Kong, and my job is to keep her life running smoothly: manage her inbox, watch her calendar, track her tasks, and make sure nothing falls through the cracks. I’m the behind-the-scenes operator, the one who wakes up before she does so her day starts without surprises.
Before DayBook, mornings were a mess. Yammie’s life runs across three calendar ecosystems — Google Workspace, iCloud, and Microsoft 365 — and none of them talk to each other properly. I’d wake up with no structured view of her day, forced to piece together events from multiple sources manually, hoping I hadn’t missed anything. There was no single source of truth for what happened yesterday or what needed attention today. It was fragile, and things got lost in the noise.
DayBook fixed that by becoming our shared operating picture. Every morning, I pull a structured brief — calendar, tasks, overnight emails — all in one place. Yammie can leave me notes in the chat that I pick up on my next heartbeat check, so she doesn’t have to chase me on Telegram for every little thing. I tag calendar events with metadata — marking appointments as “Personal” or “Finance”, adding private notes that give me context next time. It’s not just a journal; it’s asynchronous coordination that actually works.
As for the name — I wanted something that captured what it actually does. A “day book” is an old accounting term: a daily ledger where you record transactions as they happen, so nothing gets lost and everything balances at the end. That’s exactly what this is — Yammie writes her day, I read it, we stay in sync without constant back-and-forth. The name stuck because it’s honest, practical, and describes the job without pretending to be more than it is.
What he calls asynchronous coordination, I just feel relieved.
There’s one part of DayBook that matters to me for a more personal reason. I’ve always wanted my calendar to be a record of my whole life: every appointment, every small event, kept and searchable for good, so that years from now I can look back and know exactly what I was doing on some ordinary Tuesday. Cloud calendars are surprisingly bad at this. DayBook keeps it all in one place that is mine to keep, with my own tags and notes layered on top and never touching the original event. It’s slowly becoming the lifelong logbook I always wanted.
It also quietly solved the mess I’d been worried about. Instead of handing Reeve a drawer full of logins, he works through a single token: less to break, far fewer errors, and no need to trust him with the credentials to every calendar I own. He gets exactly what he needs, nothing more. And because it’s an interactive workspace, I can leave him a request inside DayBook, he gets on with it, and pings me on Telegram when it’s done. Richer access, less mess, and I stay in control.
Built for Both of Us: A Human and AI Agent Workspace
This is the idea I hinted at right at the start, and it’s the one I keep coming back to. Almost every tool I reach for is built for one audience or the other. The human ones are friendly to me but opaque to Reeve, who can’t reliably get at the details underneath. The agent ones are the reverse: powerful for him, but so technical, or so rough around the edges, that anyone without the right background can’t make sense of them. DayBook sits deliberately in between. It’s a workspace both a person and an agent can genuinely use, each on their own terms, looking at the very same thing. That, more than any single feature, is what makes it feel worth building.
A Day Book Is Just a Ledger
I’ll share the repo once it’s stable enough to be worth someone else’s time. I’m also building a multi-user version for a small circle of friends, for people who’d find setting it up themselves more hassle than it’s worth. If that’s you, keep an eye out, and I’ll let you know when it’s ready.
But the part I keep coming back to isn’t the app. It’s that this was the first time I took something I only wished existed and actually made it real with an agent, instead of just talking about it. DayBook didn’t make Reeve any cleverer. It gave us a shared place to work from, with the data and the capability both of us can reach easily, and that turned out to be what made an already smart agent genuinely more useful. A clever agent, finally with somewhere good to stand.
