Peter Zakin

From a year ago. I still think this is underemphasized by coding agent companies....

A small thread of thought I think is worth pulling on: Working with agents is a new type of coordination problem. Large companies face a lot of coordination problems and build out lots of scaffolding to mitigate them. If we’re trying to answer questions like what’s the best…

Claude can code, but most setups trap it in a chat window. We made it real: a @Daytona sandbox behind a @Cloudflare Worker, running a live React + Vite project. Claude edits files directly, sees previews update live, and keeps state across sessions. No backend. No…

I make it a point to notice when someone is living in the future. Very few people whose judgment on this subject I trust as much as Ben’s.

Agent wrappers are the new model wrappers.

Feels like we're entering a new chapter for AI coding assistance: agent orchestration. Agent orchestrators are not IDEs. They don't augment writing code so much as making it possible to manage a fleet of agent labor. Essentially, they turn devs into eng managers. The main…

It feels like we're entering a new phase when it comes to AI coding tools. If the last phase was marked by a transition from copilots in VS Code to supervised agents in VS Code forks like Cursor + Windsurf, my sense is that the next phase will be marked by a move towards agent…

🤓Calling all ML nerds! 🤓 Join me at ICML for the @askalphaxiv happy hour! We have food, drinks, and a panel. An amazing panel. Really the best panel, with @FazlBarez, @pzakin, and @tdietterich! Topic: "The science singularity." lu.ma/n0yavto0

Incredible execution from @ivanburazin and @JukicVedran. The cohort of founders who built cloud IDEs in 2010s continues to do very very well (bolt, replit, daytona).

"The problem with spec-driven development is that the default, naive implementation is high-friction. Compared to the iterative approach that we see in Lovable etc... SDD takes a lot of time and effort. I don't think Convert-Spec-to-Code is a viable product direction." This…

I wrote this back in 2019 when I was trying to figure out what my next startup was going to be. One of the more surprising realizations I had from the ensuing startup journey: I wasn't nearly as mercenary as I thought I was. After Macro hit a wall and we went through a pivot…

I've written a lot over the last year about the future of spec-driven development. A few quick (slightly updated) thoughts: - Coding agents becoming more reliable means that devs can gain leverage from tools that help them communicate more effectively to their agent…

We're looking to bring on a researcher-in-residence. Really unique opportunity for someone who has spent time working in AI research and wants to work at a venture firm, perhaps with the aim of transitioning into VC or starting a company. This person would work closely with me…

The biggest threat to SaaS is not that software becomes very cheap, but that certain business processes are largely obviated by a shift to agent labor. E.g. the future looks grim for HR software.

A few thoughts: - I have in mind the faint glimmer of an AI productivity tool that can help you overthink the hell out of something. Branches of thought; complex ideas shrunk down or expanded; conjectures with supporting or opposing researched arguments... - I think we will…

Please RT! We're seeking data analysts using AI in their workflows. We would love to observe your process and learn what has been exciting or challenging. Your insights will help us build better tools. If you are willing to sit down with us (on zoom) for a couple of hours…

I continue to be intensely interested in finding the team that figures out how to productize spec driven development.