The jumps in AI's autonomy

If you’ve raised kids, you know: they don’t develop in a smooth upward line. For weeks nothing seems to change — and then overnight they’re walking, or talking in full sentences, or suddenly grasping abstract concepts. Developmental psychologists call these growth spurts: periods where the brain reorganizes itself and a whole new level of capability clicks into place. The in-between stretches aren’t wasted: that’s when the groundwork is laid, but the visible breakthroughs come in bursts.

Claude Code vs OpenClaw: the gap analysis

One week ago I started TropicClaw: an analysis of the gaps between Claude Code and the new paradigm of OpenClaw. I wanted to understand what the secret sauce was, the combination of capabilities that made so many people go wild about it.

Estimating (LLM) bot traffic on my website

TL;DR: I configured local log analytics reporting on my site with goaccess and pforret/goax, a bash wrapper script, to analyse regular, SEO and LLM bot traffic. I then also extended my fail2ban configuration to ban vulnerability scanning bots.

Leaderboard for AI Terminal-based agents

TL;DR: I’ve created a daily updated AI terminal coding agent leaderboard (i.e. Claude Code and its competition), ranking popularity based on monthly downloads. It lives on https://terminalai.forret.com/.

The Mythical Agent-Month: Brooks's Law in the Age of Agentic Software Development

TL;DR: “Adding an autonomous AI agent to a late software project makes it later.”

A.I. Governance in Belgium

In the preparation of my PromptParlour meetings, I wanted to get an idea of who in Belgian government is responsible for understanding and forecasting the (exponential) progress of Artificial Intelligence, and its impact on Belgian business, innovation and society. Who do Belgian/Flemish ministers call when they want an expert opinion on legislation or government intervention?

Idea: A spectrogram for videos

I was driving home after work and I had a vague idea: what if I could represent a video with an image where the X axis was ‘time’ and the Y-axis was something that represented the images (it was this vague in my mind: “something with the colors”) in that video? Three hours of work later, spread out over the same evening, I got it working, assisted by A.I. every step of the way.

Transcode from C to Golang: Claude vs Codex vs Gemini

TL;DR: I compared three AI coding agents (Claude Code, Codex CLI, Gemini CLI) for transcoding an existing audio-processing C program to Golang. The clear winner in this case was OpenAI’s Codex. See the results on github.com/pforret/leqm-nrt.

Creating my first Jetbrains plugin with Junie and Kotlin

TL;DR: I created a Jetbrains plugin in Kotlin, a language I don’t master, with Jetbrains’ own Junie AI Agent. It went quite easy, and now I have my first published plugin in the Jetbrains marketplace. Source code: toolstud-io/LlmBrains

A static Docs site for every project

TL;DR: I have started using mkdox (which uses mkdocs, material and Docker) to create a static documentation site for every repository/project I’m working on. It’s easy, light and fast. It also invites me to document everything in Markdown, which is great for coding agents.