Greetings

Welcome to our little neighborhood software lab.

What you see here was inspired by a conundrum I faced in my day job: we were working to chart a course forward into the world of AI-augmented software development, but my personal experience with these tools was limited and I was relying on what I could find online and in newsletters. And online content, in particular, was overwhelmingly just AI slop and/or clearly unsubstantiated hype and/or possibly unsubstantiated anti-hype.

I started contacting friends in the industry to see what they had found to read and learn from, until one friend told me something that was the big unlock. “Dude”, he said, “just ask Claude.” It has access to a world of information that would take you weeks to sort through, it can even put together a lesson plan for you if you want to learn something of deep complexity.

So I asked Claude. One thing led to another and I built a whole agentic software lab in my spare hours. In the course of that I learned a lot of basics about agentic development and finally started to get my mind around the issues we were working on.

The follow-on problem at my day job was that, as we each learned more about agentic engineering, we weren’t learning from the same corpus. Without a shared understanding of what terms meant, how a system might work, and what was acheivable, we found oursleves talking past each other and spending a lot of time reconciling our basic ideas before we could build on them.

So I built another software lab, as a demonstration, on my employer’s systems, to show what my working mental model was. But this lab was necessarily very limited – I wasn’t going to use the codebase of an active production SaaS to work out, for instance, the pros and cons of overnight autonomous agents and different cost-containment methods.

Finally, I built out a much more complete system, at Convivy. Knowledge engineering, project management, agent materialization and control, policy and observation, all integrated into an agentic development environment with sessions that persist on a remote server.

The first thing I noticed when I started to really commit to working agentically: it increased my personal velocity as a software developer by something like 50x.

Your mileage will clearly vary; I was only doing development in my spare time so the baseline was really low. I’d been working on a multitenant SaaS as a side project, whenever I could make large enough blocks of time to focus, for a couple years, and realistically was no more than 10% of the way to having a shippable product.

Once I got facile in agentic development, I found it productive and satisfying to dip into development for ten minutes, set up and launch a process and return many hours later when I had time and mindspace to process the output. As a result, I was able to create a demo-able SaaS from scratch – it took me just a short while to decide it was better to greenfield the product agentically than to try to adapt what I had already developed – in about 3 months. I estimate the software will be production ready in another 2-3 months. And this is spending less time on it than I previously had been.

Meanwhile, if I encounter a need for a piece of software in my day-to-day life, at this point it’s easier for me to build exactly what I want than it is to try to see what’s available commercially and adapt it to my needs.

What I’m getting at here is this: I think these tools are demonstrating power to a degree that it is inevitable they will completely reshape the profession of software development. I don’t claim to know what the end shape will be, but if you want to explore some of the possibilities, I hope you find this site useful.