I went to San Francisco in October to fill my tank and reset my direction. I had just finished a large AI consulting engagement with a New York firm, and I needed to choose what comes next.
Direction is hard right now. Things move so fast that the line between my work and what is suddenly possible keeps blurring. Every week there is a new model, a new tool, a new thing I could build. I needed to step out and look at the whole board.
San Francisco is a wild city for that. I love visiting. For years I made the trip for the J.P. Morgan Healthcare Conference, or at least for the healthcare week around it. The concentration of brains, talent and ambition is enormous, and you can feel it on the street and on the billboards.
This time I got off the plane and walked into Cloudflare Connect: about 10,000 people who all came to the same place for the same three days. It’s bottled inspiration.
Where I have been
This site went quiet in March, so first, a short update.
In the spring I left Payquad after five years as Head of Product and Growth. Thank you to that team, and to how fearlessly it adopted AI, both in how we built and in the product itself. It also builds heavily on Laravel and its newest packages, which is how I came to know the framework.
Then my AI consulting practice picked up, and I joined Maigrate, a firm moving American business to AI, for a private engagement. I led go-to-market, product and strategy while we put AI agents in front of real customers. The engagement was full time, so I paused my #ShippedWithAI series. Thank you to the Maigrate team for the trust and the pace.
Since that engagement ended, I’ve gone back to a mix of clients. Connect was my first week back in public, and a chance to see where things are going.
Why I went: to test a thesis
I’m a Cloudflare investor, and I’m bullish. One reason I went was to check parts of the thesis behind that, about Cloudflare and about the wider inference market (inference is the work of running a trained model to get an answer), where everyone is racing for a share.
Cloudflare has been described to me in three acts:
- Act one: the network. The layer behind the internet. DNS, routing, keeping sites fast and up. Mostly invisible, but a foundation of the internet we use every day.
- Act two: security. Zero Trust (check every user and device on every request, instead of trusting the office network), VPN replacement and sign-in for companies.
- Act three: compute at the edge. Running code, inference and agents close to the people who use them.
Here’s how I look at a company like this. Its future eats its present. The products that pay the bills today get overtaken by the thing the company is chasing. So the question isn’t how big act three is today. It’s how committed the company is to chasing it. That’s the core of my Cloudflare thesis.
So I went to see how the company splits its attention, and whether act three is a side project or the plan.
Act three is the plan
I didn’t meet one person who doubted it. Not the engineers, and not the people who sell act two. Even people whose job is selling security to large companies told me the company is fully committed to act three, and that the growth is there.
That surprised me. Big companies usually split into camps: the part that pays the bills and the part that gets the press. At Cloudflare I couldn’t find the split. An aircraft carrier is turning, and the captain and every sailor are pulling in the same direction.
You could see it on stage too. Cloudflare now builds its own models. Clef, a decision model, shipped at the start of October. Then, the week of the conference, Michelle Chen and the team announced Clairic, a new “System One” class of extraction model. The name borrows from the fast, automatic kind of thinking, and the model is fast, cheap and built for one job: pulling structured facts out of messy input. And agents came up in almost every session.
The people
People matter in any company, and even more in one that has to turn the ship and sometimes cannibalize its own products. The people I met were exceptional.
They’re curious, and they try things. Almost every engineer I met had their own project site, and an experiment they couldn’t wait to show me. Demos on phones. Side projects running on the platform they build. Half-finished ideas they were happy to talk through with a stranger.
They also get room to do it. Nobody seemed to be waiting for permission or a roadmap slot. I think that room is the reason Cloudflare ships as fast as it does. Fast companies aren’t fast because people work harder. They’re fast because people are allowed to try.
The Laravel team
Connect wasn’t the only reason for the trip. I also went to the Laravel meetup and met the core team: Taylor Otwell, Tom Crary, Andrew Culver and the marketing team. Shout-out to the team at Resend too. They ran Resend Forward, their first conference, the same week, and hosted many teams, the Laravel team among them.
Laravel is one of my bets. I build most of my products on it, and I use its AI SDK and MCP packages every day. (MCP, the Model Context Protocol, is the standard way AI tools connect to apps and data.) I’ve contributed in small ways too: failover for out-of-credit errors in the AI SDK, a fix in Boost, and my own Workers AI provider for Laravel. From the outside, the pace is hard to believe. New releases land constantly, and they’re good.
In person it made sense. It’s an ambitious team that moves fast, and the same thing I saw at Cloudflare holds here: people who build, and the room to do it.
What excites me is where Laravel Cloud is going. Email just shipped. Nightwatch, their monitoring product, is folding into other parts of the platform. I have a fuller thesis on where they could and should go, and where I think they will. One guess: inference as a service comes next. Put those together and you get your code, how it runs and its logs, all in one place. The framework can add the hooks, and an AI client on your own machine can ask questions across all of it through MCP.
I’m writing a full post on why I’m betting on Laravel. More soon.
Inspired, so I built
I came home inspired, in how I think and in what I want to build.
There’s an image from meditation that has stood out to me. You close your eyes and watch your thoughts. They drift across the sky of your mind like clouds. They show up as randomly as clouds do, and they pass just as easily. You can watch one go by. You can try to puff it away. Or you can hold on to it and follow it across the sky.
Ideas without building are like that. They pass. Building is how I hold on to one.
Here are the ones I held on to.
Callozum. At Connect I watched a webcam charades demo built on Clef. One line in its code held a whole pattern: a fast model gives probabilities, simple code holds a short memory, and an action fires only when the answer is stable. Put that on a robot or camera glasses and every step has a best place to run, on the device, at the edge or in the cloud, and that place changes from second to second. Callozum, named after the corpus callosum, is a router that makes that call. I wrote up the idea in Inference That Knows Where to Run, and the bigger picture in Where Physical AI Runs.
The Adaptive Factory. Agents now write most of the code. What stays expensive is proving the new version is better. The Adaptive Factory rebuilds a system from a record of what it must do, runs the rebuild in shadow on real traffic (it sees every request, but its answers aren’t used), and promotes it only on evidence. I’m building the harness that does the proving. The case is in The Adaptive Factory.
Connections for Cloudflare OS. A dozen agent and bot platforms are coming out, and every large provider has one. Cloudflare’s approach is different: you own it, you build it and you maintain it. That has real benefits, and real friction. I felt the friction in the hands-on Cloudflare OS workshop with Adrian Hall and Patrick Gryczka (thank you both for the time). Each service needed its own OAuth app (the setup that lets one service sign in to another for you), secrets copied by hand and a redeploy. I wanted it to feel like what WorkOS built for sign-in, so I built one-click sign-in and connectors for Cloudflare OS. Walkthrough here.
A new personal site. And for fun, I rebuilt this site, to tell my story better and to put my ideas inside it. That story is its own post.
Thank you
To everyone at Connect, at the Laravel meetup and at Resend who gave me their time and their ideas: thank you. Most of what’s in this post started in a conversation with one of you.
If you’re building on Cloudflare or Laravel, I’d like to talk.