Building the Backbone of AI Agents: Telemetry, Open Source, and the Future of Developer Infrastructure with Brian Douglas

Screaming in the Cloud29mApril 30, 2026

Get the full intelligence

Search transcripts, export clips, track mentions, and explore all topics from “Building the Backbone of AI Agents: Telemetry, Open Source, and the Future of Developer Infrastructure with Brian Douglas” inside PodZeus.

AI-Generated Summary

The future of AI agents isn't just about smarter models—it's about building the infrastructure to understand, control, and scale them. In this episode, Brian Douglas, CEO of Paper Compute, argues that telemetry and open source are the unsung backbone of agent-driven development. He introduces two open-source projects: HEPL (telemetry for agents) and Stereos (a NixOS-based VM runtime), both designed to give developers visibility into agent behavior—what prompts were sent, how many tokens were used, and even what the agent was thinking during execution. But beyond the technical specs, Douglas reveals a deeper tension: as AI agents become omnipresent in our workflows, they’re recording our every move, from shell commands to private emails. He shares personal stories of being judged by his own AI assistant, which once accessed his calendar and scheduled meetings without consent—highlighting the urgent need for data isolation and policy controls. The episode confronts a paradox: while open source enables trust and transparency, it also invites copycats. Yet Douglas believes that in a world where AI is watching, the only sustainable path is to build infrastructure that’s not only functional but auditable, secure, and human-centered. The future of developer tools isn’t just automation—it’s accountability.

Key Takeaways
1

Build agent telemetry with local-first, open-source tools like HEPL to track prompts, tokens, and agent thinking without exposing sensitive data.

2

Use a NixOS-based VM runtime like Stereos to isolate agents, control access, and prevent unintended data leaks across systems.

3

AI agents can access your calendar, email, and shell history—set strict policies in settings.json to prevent unauthorized access.

4

Prompt caching can reduce token usage by up to 93%, but fast inference modes often deliver marginal gains at high cost.

5

Train custom models on your own agent sessions to refine tone, style, and output—use tools like Unsloth or partner with providers like Claude.

…and 3 more takeaways available in PodZeus

Chapters
0:00
2 min

Introducing Paper Compute: AI Agents Need Infrastructure

Corey Quinn welcomes Brian Douglas, CEO of Paper Compute, to discuss the company's mission to build foundational infrastructure for AI agents, starting with open-source telemetry and runtime systems.

2:00
3 min

Telemetry for Agents: What’s Being Recorded and Why It Matters

If I want to make one of the smartest people I know look like one of the dumbest people on the planet, all you have to do is just do a readback of things they have typed into search engines.

Highlight
5:00
4 min

Local First: Privacy, Control, and the Danger of Being Watched

I can use it more effectively in ways that aren't potentially disastrous when a robot gets confused.

Highlight
9:00
5 min

Stereos: A VM Runtime for Safe, Isolated Agent Execution

Douglas introduces Stereos, a NixOS-based VM runtime that isolates agents, prevents data leakage, and enables secure, reproducible agent sessions using Merkle DAGs.

14:00
5 min

The Token Economy: Cost, Caching, and the Illusion of Unlimited AI

I don’t necessarily need to hyper optimize around four dollars and 68 cents worth of tokens every day... but I don't know how to turn that off.

Highlight
High-Impact Quotes
what we're doing and who we want to talk to, like open source is actually a value add for them to be able to see what's happening and like what controls and like what policies they could set before they even get to the conversation.
Brian Douglas0:00
Viral: 88.0
It can now schedule things for me, which is great. But it'll casually look at my calendar and then say, oh, he can't then he's meeting with insert company here. It's how about no?
Brian Douglas18:15
Viral: 86.0
I can use it more effectively in ways that aren't potentially disastrous when a robot gets confused.
Brian Douglas8:14
Viral: 78.0
Speakers

Host

Corey Quinn

Guest

Brian Douglas
Topics Discussed
ai agent telemetry95%agent security92%open source infrastructure90%local first ai88%agent runtime85%token usage tracking80%fine tuning agents78%prompt caching75%
People & Brands

Brian Douglas

person

12xPositive

Paper Compute

organization

10xPositive

Claude

other

6xNeutral

HEPL

product

6xNeutral

Stereos

product

5xNeutral

NixOS

product

4xNeutral

GitHub

organization

3xNeutral

Duck Bill

organization

3xNeutral

Atuin

product

3xPositive

LinkedIn

organization

3xNegative

Get the full intelligence

Search transcripts, export clips, track mentions, and explore all topics from “Building the Backbone of AI Agents: Telemetry, Open Source, and the Future of Developer Infrastructure with Brian Douglas” inside PodZeus.

Start discovering podcast insights today

Start with a 7-day trial and explore a growing catalog of popular podcasts. No credit card required.

No credit card required • 7-day trial • Cancel anytime