I loved coding when I was a kid. It was a massive creative space where I could do whatever I wanted (with enough nested if statements thrown in). To me it was the most fantastic idea: imagine something, make something, have fun. I made games, NPC generators for Dungeons and Dragons, and databases to manage the scouts’ supply chain (yes, OK, I was a weird kid).
Later, as the companies I had founded grew from a scrappy crew of pirates into global enterprise software companies, I found out how real code had to be: debugging, object orientation, tests, infrastructure, git and the rest. I could understand all of it and did not want to do any of it, because it was not fun, and that was when I realised I suck as a software developer.
The distance from idea to iteration to payoff to scale was too long. Even though the companies I co-founded built massive enterprise systems, thanks to people far smarter than me, my coding sweet spot was always solving my own creative needs. I kept pushing myself to learn, to try new technologies, to fail and try again, because I am still a nerd at heart, but the days when I had AWS certificates for production on my laptop are long gone. The road from idea to production had become a mountain, and I knew my time was better used elsewhere.
Slowly at first, and now faster and faster, that has changed. Every six months I take whatever the state of the art in codegen is, make something with it and see how far I can go before the layer cake collapses. It is great fun, and a good way to understand how AI works and where its limits are. Recently the layer cake stopped imploding: it kept improving, and it worked. I could take even complex systems from idea to execution, and manage them. Instead of building infrastructure for debugging, deployment and all the boring parts, I found I had built an infrastructure for creativity: GitHub repositories for templates, Cursor and Cline for development, MCP servers for testing and code review, Docker for containers. The best part was that I rarely touched the code: with Composer in Cursor, Claude Code and the tools and agents I had built myself, I could finally get stuff done.
The developers on my team are some of the most brilliant I have ever worked with. In my excitement I started telling them about my new workflows: deep research to pull context for a new experiment, AI templates to manage projects and track progress, rules so that the agent would code the way I wanted, all with the purpose of taking the engineering out of software. They did not see it that way. Being engineers, and knowing how to code, they could never dream of going to such lengths not to code.
We are looking at what AI means for software development all wrong: we keep asking what it does to the people who can code, when for every hardcore developer with the mind and the stamina to go through the valley of real engineering (and I salute you) there are a hundred more like me. Software development is power, the power to turn your ideas into reality, and in a world where most products, services and value are virtual, that is a superpower. We are at the beginning of the tsunami of change, good and bad, that this will set off.
This is what we are building at Beyond Work: the ability for anyone in a business, the other weird kids who dream of solving enterprise problems, to build whatever they want for the problems they have, in a safe, compliant and secure manner, of course. The last time something like this happened was when Excel got macros.
Enterprises need their lovable software too.