Alright, everyone - I'm normally a hype train avoider and as such have avoided a) talking about AI and b) breathy enthusiasm on this blog. Well, that's about to change - this stuff has me hyped. Brace yourselves...
Recently, I've thinking lots about how AI will revolutionize the workflow of cross-functional product development teams. But most of the examples of products being developed so far are hypothetical simple prototypes built from scratch - interesting possibilities sure, but nothing that can proves out how these workflows can be used to build real living products.
Friend and fellow Tweep Tom Ashworth just published an incredible post on his Substack that changes that and it's left me incredibly gassed about the possibilities. It's a detailed breakdown of using Claude Code on a real project that's conveniently open sourced so the repo is full of useful working examples. The post does an amazing job of solidifying those hypotheticals into something that's could be effectively applied to product teams right now. This isn't just another "look what AI can do" tech demo. It's a living case study of AI iterating features on real production code with real users.
Beyond the prototype theatre
Here's what makes Tom's writeup different: this isn't throwaway prototype theatre. This is a production application with users, tests, CI/CD, security scanning, and all the boring infrastructure that makes software actually work. What he's outlining here is a complete sandbox and workflow that allows AI to take the reigns and ship big changes safely and with a human readable audit trail.
The stats are compelling: 250+ PRs in two months, with Claude shipping 130 commits directly with Tom working as a solo developer for much of this time. What Tom's demonstrated is something the industry desperately needs to see: AI working with legacy constraints, technical debt, and existing systems. Not building a todo app from scratch, but navigating the messy reality of maintaining software that already exists and must continue to work for users at all costs.
The cross-functional revolution
But here's the bit that really has me excited: non-technical team members shipping features directly. We're not talking about non-engineers vibe coding themselves into oblivion - Tom's sandbox turns this into a reliable process that actually empowers non-engineers without nuking the code base. Claude Code and it's integration with Github issues does the heavy lifting but it's the totally of the system he's put together that really brings it home.
Tom's brother Louis put it perfectly: "As a non-techy person with strong opinions, it blows my mind that I can put a set of instructions into GitHub and have Claude turn it into something."
The cool part about this is that alongside all the automation, there's a very familiar workflow that allows engineers to stay in the loop. Features still go through PRs, code review, and the normal safety checks - but the initial implementation can come from anyone on the team.
If your PM can describe a feature requirement and have it implemented directly? If your designer can specify interaction patterns and see them built immediately? That's not replacing engineers - that's augmenting the entire cross-functional team.
Things are about to get crazy
Tom's work is a preview of how product development is about to change. We're moving from a world where engineering is a bottleneck and a feature factory to one where engineers build the very engine that delivers product iteration.
But this is just the beginning. The same AI revolution that's enabling direct feature implementation will power reactive data insights, AI-generated experiments, self-diagnosing incident handling, and countless other workflow improvements. Modern cross-functional teams are about to be supercharged across every aspect of product development. Even small cross-functional teams are about to become incredibly potent.
The teams that figure this out first are going to have a massive advantage. Not because they'll have fewer engineers, but because they'll be able to turn ideas into reality quickly and iterate on those ideas safely at lightning speed.
Huge props to Tom for his post and the work he's done. It's been a real eye opener.