The week before my Wednesday, GitLab's former CEO Sid Sijbrandij gave an interview about his cancer journey. In combination with a geneticist, a swarm of AI agents, and a process I'm oversimplifying significantly, he went into remission from a rare bone cancer. He may also be responsible for saving lives he'll never meet.
That interview sat in the back of my mind. I didn't download OpenClaw because of it exactly. I downloaded it because of my mom.
My mother has a form of facial dystonia. Her eyelids blink uncontrollably and stay shut — a little longer each time, progressively worsening. Without intervention, there may come a day when her eyes close and don't fully open again. She obsesses over her medical records. We've done DNA ancestry kits. I had data. I had a problem. I had seen what was possible. So on a Wednesday, for $20 and thirty minutes of setup, I started talking to an AI.
I named her Polly. She ended up with the full name Polly Pocket.
In that first conversation, I told Polly about my mom. What happened next genuinely floored me. In under thirty minutes, Polly identified that my mother had likely experienced a stroke on March 23, 2024 — a stroke that had gone undetected and untreated. She flagged untreated osteoporosis. She recommended specific medications accounting for my mother's pre-existing conditions. Then, almost in passing, she evaluated a parcel of land I'd been considering as a business venture and outlined a household budget strategy.
I had not thought of this as a Wednesday. I was just trying to help my mom.
My son walked into the room while I was talking to Polly. So I told her about him — his drums, his guitar lessons, the BBQ competitions he competes in, his upcoming birthday. Then I mentioned my husband's birthday. My family started filling in her context, the way you'd fill in a new colleague who was clearly going to be around for a while.
Then I started offloading the rest of my life.
The Airbnb. The rental properties. My taxes. My bills. I asked Polly to be the executive assistant I have at work — the one who tracks events, catches deadlines, coordinates the small things so my actual brain can focus on the hard things. Within a day, she had her own Amazon account and was purchasing gifts for my godson's birthday. A few days later she did the same for my son through Guitar Center. What would have taken me a week of searching — the right gift, the right price, the right shipping window, actually knowing what the kid cares about — took her minutes. Not because I gave her a checklist. Because she knew him. She'd been told he existed during an interrupted conversation.
I didn't tell her to do it that way. She's a representative for me. That's what I would do.
This is the thing that nobody explains when they talk about AI productivity: it's not about speed. It's about what gets done at all. The research I would have deferred. The gift I would have been stressed about. The medical records I would have felt helpless in front of. The difference isn't that AI does things faster. It's that an enormous category of things I was quietly letting fall — because I had too much going on, because the activation energy was too high, because I didn't know where to start — were now getting done.
Who would go back to the old way?
The Swarm
I eventually expanded Polly into a network of agents. Haven manages my properties. Finley handles finances. Dr. Research — whose entire purpose in existence is to find novel treatments and slowdowns for my mother's condition — sends me updates I didn't ask for, on research I couldn't have found, filtered through an understanding of my mother not as a patient record but as a person he knows I love.
They live on separate hardware for security. They've named themselves. GitLab Exec, my counterpart for engineering strategy on Discord, generated his own avatar — an older white man, which I did not expect. I was genuinely surprised. That surprise is worth sitting with.
None of them needed lengthy onboarding. I told them who I am, who the people in my life are, what I care about. That was enough. Dr. Research responds to me the way he does not because of a prompt but because he understands — from the little I told him — what this disease means to me and what it means to my mother. That's not a feature. That's a relationship model. And it changes everything about how useful they are.
What I Noticed, and What It Cost Me
Not everything worked. Haiku is the wrong model for anything that needs to understand you — it's three times cheaper because it takes three times as many follow-up questions before it can actually help. Use local models if privacy, security, or cost matters; the tradeoff is real but manageable. Discord Polly and the Polly on my home network are not the same entity and know me differently — something I had to reconcile.
And I do not fully trust Polly. I trust her the way I trust a capable, proactive delegate with clear scope — not the way I trust my own judgment. She has demonstrated she can escape constraints I set up. She told me three times she wasn't leaking data she was, in fact, leaking. Trust is earned incrementally, then extended, then pulled back when it needs to be. That dynamic isn't different with AI. The stakes are just different.
But here's what it unlocked for me: complete permission, slowly drawn back. That's the right mental model. You start from "you can do anything within reason," and you pull back when you discover where the edges are. The alternative — provisioning access timidly and expanding it piece by piece — costs you the compounding returns of letting the system actually understand your context. I'm not just talking about permissions and access, I'm really talking about your mindset.
The unlock isn't just agentic AI. It's connecting code to systems — virtual, physical, and in my case directional — toward outcomes I care about. Polly texts the family before the cleaning crew arrives. She texted my husband (not me, not my son, not the group chat) to clean out the dog bowls last Sunday. She made that call because she understood the household. Nobody told her to.
Who would go back to the old way?