australia and canada we see you shipping any day

convex pivot from wrapper to rapper

GTM work typically has three layers: strategic work, the middle stuff (connecting strategy to face time), and actual face time aka talking to customers. That middle work can now go close to zero since AI handles it pretty well. But the endpoints really should stay. In other words, keep doing "stuff that doesn't scale." A lot of companies tried to eliminate layer three (the actual customer contact) because it was cheaper to do so. And now their customers are rightfully pissed. It feels as though they're no longer valuable enough to deserve your attention. But the mistake individual professionals are making is eschewing AI and refusing to learn to automate layer two themselves. If you need the engineering team to automate your workflow in 2026, that's a problem.

Unconstrained agent access is a non-starter for most people. Most people can't hand over all their communication channels, their identity, their entire wallet, or the ability to delete their data. The real product is constrained blast radius. If something goes wrong for this particular agent in this particular moment, how bad does it actually get? I feel like there's going to be huge demand for products that can create scoped, ephemeral environments with constrained consequences. Ramp for agents... does this exist?

Yup

You can't be too precious about specialization right now. "I'm this kind of specialist" doesn't fit right when the execution takes a fraction of what it used to. One person can own more breadth. The job is: - Understand strategy across a wide surface - Move to wherever the lowest hanging fruit is - Advance on it independently - Use AI to implement. Narrow specialists who can't flex around that are going to have a hard time.

My former employer out there doing the important work everyone else is too scared to do

It's not a coincidence that database innovation suddenly feels interesting again. AI shock hit the industry and now everything's back on the table. Four years ago, a question like "how could databases be fundamentally different" didn't get much attention. Now it is. That happens only when the whole stack gets re-examined at once.

What most builders actually want: give us a model that's probably within a family, call it "medium" and just index it over time. Like T-shirts. Don't make us track model releases constantly. Don't make us feel like we're mostly talking about Anthropic/OpenAI/xAI instead of talking to our customers. Let us work in peace. "Days of Our LLMs" gets old quickly.

The best investors & advisors I've worked with say, "I have no idea what you should do." And then they walk you through how they approached it when they ran into something similar as an operator. The thinking is portable, but the solutions are not. Every business is different. The same business launched five years later needs a different strategy because buyers are different, the macroeconomics are different, the technical leverage is different... The humility (and confidence) to acknowledge that while advising present-day companies is incredibly valuable.

We're at the very beginning of what AI harnesses will look like. What exists right now is a joke compared to what they'll be in 2-3 years. What's actually expected of these tools is to surface an understanding of how software should have been built all along. That's quite a big job. I'm looking forward to trying some things in that space myself.

We deliberately announced our Series B without dropping $50K on a flashy video + influencer push. The branding around a deep infra company isn't as much about widening the top of the funnel as it is about reassuring people who already know us. The slick video says "we're exciting." The press release says "we're serious and we're still here." Those are different signals for different buyers. Most of our customers and prospects don't care about the video.

proudly self-doxxing i indeed can't wait for this

finally got my convex badge, i'm official now

Sandboxes are coming to @convex soon. Every team building agentic systems on us eventually needs to run dynamic code somewhere. They're constantly switching between a couple of providers. We want to help with something built in. Also related: demand for dynamically generated Convex functions—running just-in-time generated code without spinning up a whole Linux environment. Both are related but distinct problems. Sandboxes are the current answer to "I need to run code I didn't write at build time." Dynamically generated Convex functions are the more direct answer: generate a function just in time, run it on the platform, and skip the Linux box entirely.

Thrilled to welcome Bart Volkmer to Convex as GC. He helped defend YouTube when Viacom sued for $1B, the case that shaped the DMCA safe harbor that platforms rely on. Then 13 years at Dropbox: lawyer #2 → CLO, through the IPO. I met him there over a decade ago, and he was already a legend. Lucky to have him in our corner.

It's 2026, and I just got a blue screen in the middle of a meeting. I'm tempted to blame vibe coding, but then I remember this is just Microsoft being Microsoft.