A higher level of abstraction

Why this conversation

Agents now write code faster than people can review it. Benjamin Hindman created Apache Mesos and helped set up the CNCF, and today he is founder and CEO of Reboot, a framework built for that problem. In this conversation he compared the Agentic AI Foundation with the CNCF’s early years and explained how Reboot keeps an application readable when agents write it. It is his second conversation on Udienza, after the first at AGNTCon + MCPCon Europe in Amsterdam. The host recorded with him at the conference’s studio on the show floor in San Jose on 22 October 2026, the first day of AGNTCon + MCPCon North America. The rest of that day is in AGNTCon + MCPCon North America 2026: day one.

From Mesos to the CNCF

Apache Mesos came out of PhD work he was doing at Berkeley, and it was one of the first container orchestrators, “if not the first public container orchestrator”. Twitter adopted it, then Netflix, Uber, PayPal and many more companies in Silicon Valley, and he went on to found Mesosphere, which helped kickstart the cloud native movement. He helped set up the CNCF with the Linux Foundation and sat on its initial Technical Oversight Committee. That was about ten years ago, when the world was totally different.

The same pattern, ten years later

He sees a lot of similarities in the Agentic AI Foundation. The Linux Foundation, he said, is good at recognising the technologies and movements that benefit from being in a foundation, and much of what is being built here is going to proliferate in everyone’s stack and become foundational, as it did with the CNCF. One lesson from his time on the TOC still applies: let a thousand flowers bloom, because you want lots of things to happen, but find a way for it not to become a nightmare for the organisations that have to bring all that software in-house and manage it. Everything moves faster now that agents write the code, and that makes it even more challenging. Agents may manage it all, “but at the end of the day, I still think that human oversight is really, really important.”

The domain object

That is the problem Reboot solves. Agents write more than humans can really review, and when people do review, they often do not feel they are moving much faster. Reboot’s answer is a new framework to write the code with in the first place: a backend, full-stack framework, with Python as its first language, so that instead of Django or FastAPI the agent writes the backend in Reboot. Its key primitive is the domain object, the thing a team puts on the whiteboard when it designs an application: this object, its relationship with that one, the interface between them.

Many programmers today have their agents write a markdown specification of the API and the behaviours, and then still inspect the code to check that it matches. With Reboot, the agent builds the code and Reboot extracts the domain objects from it, so there are not two places to harmonise, and any change to the code shows up as a change to the domain objects. He compared it with the move from assembly language to higher-level languages: “We need to give a higher level of abstraction to humans, still let the agents write as much software as they want”.

Building is not gone

To try it, go to reboot.dev and install the Claude or ChatGPT plugin, then work with your agent to build the app, 100 percent locally on your laptop or in the cloud, and ask the agent to show you the dashboard of your application. As a developer, he still makes the decisions he would make writing the code himself, about the data architecture or how services call each other, only much faster, because he can see right away what the agents are building. And because Reboot is a new framework, its apps are MCP servers out of the box: tell it to make some domain objects tools, and they immediately become tools.

When it comes out

The episode is edited and in the release order. Like every episode it goes out on YouTube first, then on Apple Podcasts, Spotify and Amazon Music.