MD Michael Dominick Michael Dominick Software developer and podcaster covering the art and business of software development, from Rust and game development to AI coding tools and development practi...

“LLMs and AI coding agents and assistants seem to suffer from technical debt as well. … As your project gets bigger, as your to-dos and your documentation drift, just like a new developer on the team, the LLM spends more tokens, more context, trying to wade through the cruft than actually doing anything, making it less efficient, potentially sending it down blind alleys.”

Coder Radio · Hands-On Engineering Podcasts · November 2026

“LLMs and AI coding agents and assistants seem to suffer from technical debt as well. … As your project gets bigger, as your to-dos and your documentation drift, just like a new developer on the team, the LLM spends more tokens, more context, trying to wade through the cruft than actually doing anything, making it less efficient, potentially sending it down blind alleys.” — Michael Dominick, Coder Radio

Context

Mike was summarizing a paper by a Google engineer suggesting coding agents slow down on cluttered, out-of-date repositories, and compared it to stale docs he finds in client projects.

Transcript

Coder Radio Around 00:00 into the episode
Michael Dominick

…law. So yeah, big deal. Big deal. And I think, you know, this is one of those things where I'm sure by the time this show actually hits the feeds, the story is going to be a little deeper. Other news. So a Google senior staff engineer named Sam Saccone. I hope I said your name right. I actually wrote a really good paper entitled A Great Distraction. So basically, I'm going to TLDR it for you. LLMs and AI coding agents and assistants seem to suffer from technical debt as well. And he actually basically, he proves it out in the paper. So what that means is as your project gets bigger, as your to-dos and your documentation drift, just like a new developer on the team, the LLM spends more token, more context, trying to kind of wade through the cruft than actually doing. Anything, making it less efficient, potentially sending it down blind alleys. I mean, this is all the…

Speaker 2

Hey, well, first of all, thank you for having us today, Michael. Yeah, I'll give you a little bit of background on kind of where the companies come from and where we're going. I'll start with where we're going first. So MZero is the developer of what we call an autonomous cloud control plane. And…

Ben Vernays

Yes, thanks for that introduction, Chris. Depends on how you define employee. I'm up there in terms of seniority. But from a perspective of CloudQuery, it started off as an ETL tool focused on cloud security and cloud infrastructure. We were sort of the first to look at security and infrastructure…

Michael Dominick

So interesting. All right. So the first company, or I'm first is maybe wrong, with the one that Chris mentioned first, that merged and became Ev0, was basically doing, to put a plainly abstraction for Terraform and stuff like that.

Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.

More from Coder Radio