About

What this is

Tech Talk Ideas is a working record of engineering practice. It covers how a codebase explains itself to whoever has to work on it next, how to reason about trust and security in systems that hold credentials and run code, how the tooling fits together across editors, build systems and CI, how to design the loops the work happens inside, and where the coding agents themselves actually differ once you are relying on one.

The deepest coverage right now is of AI coding agents, because that is where most of the recent hard problems have been and where the received wisdom is thinnest. The scope is wider than that, and the standard is the same wherever it goes: a claim you can check, drawn from a failure somebody actually hit.

It is written for engineers who have already run into the edges — the context file that made things worse, the dependency nobody reviewed, the unattended job that ran for forty minutes and produced nothing.

Who writes it

I am David Guzenburg. I write software, and I have spent enough time with this tooling to have opinions about which parts of it work. Everything here comes from practice or from published research, and where I am describing someone else's findings rather than my own experience, I say so.

How it is written

Three commitments, because they are the ones I would want from a site like this:

Corrections

This field moves quickly and some of what is here will age badly. If something is wrong, tell me and I will fix it and note the change.

Independence

No vendor sponsors this site, and nobody pays for coverage. Product and protocol names are the trademarks of their owners; nothing here is affiliated with or endorsed by any of them. The site carries display advertising, which has no bearing on what gets written.