Agents

If you’ve spent any time on social media last week, you would have to work hard to avoid the hype surrounding Clawdbot / Molty / OpenClaw. In the course of one single week, the artificial virtual assistant changed name twice and caused a spike in Apple’s Mac Mini sales. The idea behind OpenClaw is simple but fascinating: give an AI agent access to your desktop and delegate work to it. Ask it to free up your Friday afternoon, and it’ll reschedule the meetings to another day. It can organise a Doodle and then book a table for that dinner with busy friends. It’s the trusty personal assistant that only top managers could afford. Until now. ...

Like all online topics these days, coding agents are divisive. Some zealots predicted the end of the software developer around 8 months ago. Others consider those agents to be as useless and atrocious as Microsoft Teams. Most of us are somewhere on the spectrum between those extremes. Over the last few weeks, I have been taking advantage of Anthropic’s holiday offerings and have been spending an unhealthy amount of time with Claude Code. It’s fun, it’s productive, but it also teaches us something about the state of the art. Or rather: about the perceived state of the art. ...

In my line of work, I see a lot of legacy code – products that were built so fast the foundations are weak. Every little change causes something else to break. The industry standard way to stabilise such a codebase has always been test automation. But writing unit tests takes time, and in the rush of shipping features, that time is often dedicated to anything but tests. As the product expands and the code gets ever more complex, this lack of tests becomes a painful problem. If small changes break the system, sweeping changes are downright scary. The team gets paralysed. ...

LLMs are stochastic parrots. They just predict the next most likely word. And the most likely token to follow “AI” these days is, without a doubt, “Agents”. AI agents are hailed as the next quantum leap in the evolution towards building SkyNet AGI. An agent is basically a program that can interpret input with an LLM and has access to tools to actually do something with that input. Instead of ChatGPT drafting you an email, an agent could send emails on your behalf. You could ask your agent to “send friendly reminders to fill out that Doodle”, and it could read your latest Doodle, see who didn’t reply yet and then ask the Gmail agent to send those reminders to the right people. Rather than text-in-text-out, an agent converts input into actions. ...