This is your life. You only get one. Automate it at your peril.
Pedro Franceschi is 29 years old, and having sold Brex to Capital One for $5.15 billion, he’s a good deal more successful than I am ever likely to be.
On the Core Memory podcast, he walked through his daily operating system: a “signal ingestion pipeline” screens his email, his Slack, various Google Docs, and WhatsApp, and then filters everything through the 25 people he actually needs to track. Granola (meeting note recorder) sits on every meeting, pushes the transcripts into the same pipeline, and generates action items automatically.
The system pulls context from the original meeting // drafts the follow-up (Slack, email, text etc), and Pedro clicks approve. He has a virtual recruiter named “Jim” who lives in Slack with his own email address. There’s a security layer called “Crab Trap” designed to intercept all agent network traffic through an LLM proxy, a sort of ~second AI watching the first, in real time.
This is how a $5 billion CEO runs his life.
And my first reaction was total exhaustion.
Reader - I am tired.
I don’t doubt, not for a moment, that Pedro’s system ~works. But his setup is the ceiling of what’s possible right now, and somehow it’s becoming repackaged by various influencers as the new floor, the new baseline, the minimum viable human. Scroll Tech Twitter for 4 minutes and you’ll find 40 posts telling you that if you’re not running an agentic pipeline across your communication channels, you’re falling behind, leaving leverage on the table, about to get eaten by the people who figured it out first, about to fall into the permanent underclass...
Automation used to require a justification.
No longer.
Mario Zechner, who built the libGDX game development framework and whose work sits underneath a good deal of the tooling these agents are built on, including everyone’s favourite Lobster Cult, published a post recently called “Thoughts on slowing the fuck down.” He spent a year watching agentic coding happen and wrote about what he found: codebases compounding errors faster than humans could review them, teams of 2 replicating the technical debt of a 200-person enterprise in weeks, tests written so thoroughly by agents that they were equally untrustworthy as the code underneath.
Agents do not ~learn.
A human makes the same mistake a few times and eventually someone screams at them. The agent just keeps going.
“We have basically given up all discipline and agency for a sort of addiction, where your highest goal is to produce the largest amount of code in the shortest amount of time.”
That word, addiction.
I think we throw it around a good deal too easily - as a continuously recovering alcoholic, I may be a little sensitive to it - but I think it applies here. And that makes me somewhat uncomfortable.
This week, I’m planning on moving my writing, my work and my business off Google and onto Proton. Email, calendar, storage, all of it. Privacy is part of it; but mostly I don’t want the automation, or even the pull toward it.
Google’s ecosystem is an automation machine. Everything connects to everything - your email connects to your calendar connects to your docs connects to your contacts connects to whatever AI product they’ve bolted on this quarter. It’s frictionless by design, and it’s practically self-installing. You can hook your inbox to a chatbot and that chatbot to a spreadsheet and that spreadsheet to a notification and before you’ve made a single conscious decision, you’ve built a fragile little parasitic nervous system that runs on your behalf.
Proton doesn’t do that. The tools are solid but the integrations are sparse. There’s no “connect everything” button. If you want something automated, you have to go looking for it, which means deciding you actually want it, which means ~thinking.
That friction is the point.
I think, on some level, I crave it.
Architects talk about affordances: the handle on a door that tells you to pull, or the flat plate that tells you to push. The shape of Google’s product says: let me handle this. Proton’s says: what do you need?
Moving ecosystems is a clean break, but it’s not the only option. What I’m pursuing is the conviction and the committment to staying in the decision itself. About what gets automated, what doesn’t, what connects to what, who has access to my attention.
Pedro may have made all those calls deliberately. He may have thought through, considered, mapped out and stress-tested the 25 people he “tracks,” the specific filters, the explicit approval step before any message goes out. I can’t say for sure, one way or another.
But I do know that most people copying his setup are likely to skip that part. They’ll install the pipeline before they’ve worked out what they’re trying to do with it.
Something Zechner wrote has stayed with me: let the agents do the boring stuff, the stuff that won’t teach you anything. Then look at what they came up with, take what’s reasonable, and finish it yourself. The agent is the fast part, but you’re still the one who knows what ~good looks like.
That version works, but it asks something of you. You need a view going in // some sense of what you’re building // what you’re genuinely willing to hand off versus what you need to stay close to // whether you’ll actually look at the output or just ship it. “Just automate everything and see what happens” collapses because you’ve removed yourself from the loop before you understood what the loop was for.
We’re in a moment of excess. You can connect your email to a chatbot to another chatbot to a Slack notification to a CRM update to an invoice to a calendar block. The plumbing exists, and the APIs are documented, and the tutorials are sitting on YouTube. There are no technical barriers left, and the only thing standing between you and a fully automated business you don’t understand is a decision you have to actually make.
What you do and do not automate is a real decision, and it deserves the same deliberation you’d give any other choice about how your business, your life, your brain runs.
For some things the answer is obvious. Schedule coordination, status updates, the boilerplate that eats 2 hours a week and requires no thought - fine. But for anything that touches how you see your business, how you relate to customers, the automation removes information you didn’t know you were collecting. You lose whatever lives in the friction.
But sitting with friction is how you learn the system.
If it breaks, you need to be the one who can fix it.