How Good Are Coding Agents Now?

Tuesday, October 13th, 2026

The last couple of times I tried to use a coding agent in anger to come to grips with a large existing C or C++ codebase for a desktop application, they failed miserably. Either the code base was too large or too complex, or there wasn’t enough C and C++ training data in the dataset for the LLM, or for some other reason, it couldn’t even begin to do something reasonable. Most recently, when I tried to use a coding agent to vibe code a macOS desktop app and a simple one at that, it failed miserably. It could not produce anything that could even build and run one simple test. But that was about four months ago, and that’s a long time in the world of LLMs. So I figured I’d give it another try with another problem that has recently come to my attention.

Specifically, I’m going to try to see if any available coding agent can analyze https://github.com/transmission/transmission/issues/8401 to figure out why the Transmission UI routinely goes into a spinning beach ball state on the Mac. Let’s begin.
(more…)

Perplexity beats Gemini Pro

Friday, February 20th, 2026

Over the last couple of days I was working on a relatively complex software integration project for image generation with stable diffusion. I had dozens of different pieces from multiple websites to download, configure, integrate, and debug. None of it was particularly well documented. In the past, I would have read what documentation existed, most of which was out of date, experimented a lot, and asked a number of questions on Reddit or other forums. This time, instead of bothering a lot of people online, I decided to use LLMs and chatbots to assist me and explain to me in detail what I needed to do when and how to accomplish my task. The results weren’t pretty.
(more…)