Hi friends,
One night, as I was falling asleep, I remembered something I needed to buy. I sent Muse a message asking it to remind me the next day. By morning, it had gotten to the checkout page and was waiting for my approval. A quick yes, and it was done.
There is almost no chance I would have sent that request to another person at 11:30 at night. At best, I would have written myself a note. More likely, I would have forgotten until the next evening, when I’d repeat the vicious cycle.
Over the past several weeks, I’ve been giving Instinct and Meta’s Muse access to more of my life and seeing what happens. Before this, I had tried OpenClaw and Hermes. After just a few weeks, I compared the time I spent with these products. I asked both how much we’d been talking. Their counts came back at roughly 3,500 messages with Instinct over 30 days and another 1,600 with Muse over about two weeks. Even for someone who spends his days investing in AI, that felt like a lot.
Some of this has taught me a lot about how I delegate. I have a full-time chief of staff. There are probably plenty of things I should already be handing over. But I feel weird sending another human every half-formed thought that passes through my head, especially when it’s something small or personal. With an agent, I don’t hesitate. Now, work that would have stayed in my head has somewhere to go.
The agent also picks up things I haven’t explicitly handed over. After one recent call, Muse read my Granola notes and noticed I’d promised a founder an introduction. It came back with the email drafted. During our fundraise, my agent has been updating our CRM as investors respond, bringing the context from our calls into the record. Those little follow-ups accumulate quickly when I’m switching between fundraising, portfolio work, and preparing our annual meeting.
As I started handing over more, though, using two personal agents at once created a surprising amount of work. Every task came with another decision about where to send it. I tried separating my work and personal life. Then a meeting change affected when I could pick up my kids, and I wasn’t sure which agent should take that request. For now, I’ve found it easier to spend a stretch using one agent for almost everything.
That has made me pay more attention to how each product handles complicated work. Muse has felt faster, and its separate conversations have helped me keep projects straight. For our annual meeting script, I’ve been piecing together my own process. I send voice notes, have the agent turn them into a draft, review it, and work through revisions before moving a section into Google Docs. Every section still needs careful checking. It is a lot to manage, and I doubt we’ve found the best way to do all this yet.
I also need to be careful with small tasks sometimes. I recently asked Muse to reorder something I usually buy on Amazon. It couldn’t complete the Amazon purchase and suggested buying directly from the vendor. The total was $31.78. When I checked Amazon myself, it was $18.50. Finding the soap and getting it to checkout still left me with an important detail to catch.
Giving these agents more access also means thinking about what they might do with it. One of my early instructions to Instinct was to check with me before taking action based on an incoming email. Someone else’s message could contain instructions I would never authorize. I can’t assume that one rule covers every way things could go wrong. A lot of people don’t have the time to spend their evenings figuring that out. The products have to earn that trust.
We’re working through some of these questions inside Flybridge, too. Daniel and the team have made our internal agent more useful by having it surface developments across the portfolio on its own, including signs that a company might need help with fundraising. That creates an opportunity to call a founder before they’ve had to chase us. Daniel wrote about that work here.
Working through this as both a user and an investor is shaping how I think about opportunities at Flybridge. Even with today’s models, there is an enormous amount of work left in making agents reliable and easy enough to hand things over to. Using them every day makes those problems very specific. I can feel the overhead when I’m managing two assistants or checking whether one has understood what I actually care about.
For now, I’m still tinkering. Some experiments cost me time, and others get things done while I’m in a meeting or with my family. I love my new mornings when I wake up, and the agent has gone through my email, prepared a couple of replies, and pointed me toward the one task that needs my attention. Being able to start there feels pretty good.
If you or someone you know is building something early in the AI agent space, I’d love to hear what they’re working on and what you’re learning. And if someone comes to mind who I should meet, shoot me an intro.
Jesse
