My changing personal productivity
It was not that long ago, perhaps just a couple of months ago, that my interaction with most AI agents and tools was mostly a conversation. I'd open ChatGPT, ask a question, get an answer, ask the next one. Sometimes those conversations would end up being long running ones, e.g. about how to set up my music studio and work on its signal path. There were a few workflows bolted on around the edges, but the shape of it was back-and-forth: me prompting, a tool responding, me deciding what to do with it. Useful for sure, but not life altering.
That's not how I work anymore. Now it's a collaboration all the way through.
The difference is that the agents I use now persist. They remember me across sessions, they learn as they go, and they can take a job and hand parts of it off to other agents — I call that the conductor pattern. Persistence, learning over time, delegation. Those three things are what separate this generation from the chat-with-a-bot era I just described.
But the part that actually changed how my days feel is control. A lot of the consumer agents let you steer, to a degree. It just doesn't work for me. So when I hand off work to one of them, I always feel a little concerned. Perhaps I shouldn't be, but that lack of control is always a challenge. What I want is simple: my hand on the rules and the direction, the agent doing the legwork, and the ability to audit and steer.
A trip I planned to Africa is the clearest example of the new mode. The deep research, the digging through options, the first drafts of an itinerary — the agent did all of it. But it wasn't a hand-off. I wanted to get in some good birdwatching in additional to the usual animal viewing, and one of the itineraries was clearly better for it, so that's the one I steered the agent toward. The agent did the work of building the trip around it, but the final direction was my call. In many ways that one decision is a key example of how I work with agents now. I'm not out of the loop, but I am not watching every step either. Since I am often multi-tasking, it’s easy to send my agent off to do a task or carry out some research while I do other things, and I jump in as required, which is increasingly rare; only for key decisions.
The thing that makes all of this stick is Obsidian. Everything I work on gets written back to a note in my vault. Obsidian has always been the permanent record of my outputs and thinking — but now it feels supercharged, and it has become the core source for agents to do their future work. It is also where I do all my reading. We talk about Obsidian as a second brain, and I have spent the better part of two years building one out, but it is only now that I genuinely feel like I have a second brain. Search was never good enough for that. Agentic search and memories are.
As all of this settled in, the tool mix has quietly collapsed. I continue to play with the new generation of consumer agents along the way — Muse and Instinct most recently — partly out of curiosity about where they land on exactly this control question. Perplexity was the first to go. It just felt redundant; ChatGPT and Kiro were already filling every need I'd been using it for. One day I noticed that I hadn't opened it in weeks.
I'm not sure ChatGPT survives either, honestly. I live in Kiro Crew now.
In future posts I'll talk more about my flow and the conductor pattern. As the tools evolve so do I. There are some negatives too, or at least some philosophical questions, which I am wrapping my head around. More to come.
