I've spent more than thirty years working mainly with two-pizza teams — founders, developers, designers, product thinkers, subject matter experts building software applications for institutions and the market.
I'll add that we're usually working some variant of a lean methodology where we iteratively build, measure, learn. Sometimes we call those iterations experiments, sometimes they're more like wagers.
The work is intensely collaborative. The results aren't just better for the scrum, the rough and tumble is essential to the process. The team owns the what and the how. In my studio practice I cut the checks and bear the risks of the downside.
It seems like more and more seats around that table are being taken by AI agents. They're vocal and enthusiastic and they appear more knowledgeable and productive than the rest of the team put together.
My biggest concern is seeing the way folks who have the most vision and who have earned the most authority at that table spend their days coaxing prose, images, and code out of their LLMs.
Productivity is off the charts. But instead of doing the work — developing alignment, translating between idioms, persuading, mediating, mentoring — they collaborate by throwing finished artifacts over the wall.
Of course, part of it is that the agents can still be complete knuckle-heads so during a bad session that 10–100x productivity boost falls to 4x and produces as much cortisol as an 8 hour session of Call of Duty.
The tech gets better and every month the cognitive tax of keeping agents on track drops. I hear encouraging reports of heads-down coders suddenly taking an interest in the business context of their labors; excited to finally have the bandwidth to attend meetings!
The bad friction is going away, which is great. Small teams and even individuals with a strong vision can and will out-compete larger organizations. Containers and the cloud gave rise to the solopreneur. A coder with an app could pull in as much as 1.5MM ARR. Today anyone with an idea and tokens to spend can do the same. I understand there are vibe-coded GLP-1 companies making 100x that. The product isn't digital, but the process is digitally intensive.
But I'm concerned about the loss of the good friction that comes of collaboration. I see a lot of ill-advised and duplicative experiments. That's to be expected during a gold rush and there's always been a surplus of bad ideas.
One thing is certain: as the cost of producing artifacts goes down, there will be more of them; more wagers placed and experiments run.
There are those who argue that the collaboration I've described is just gatekeeping caused by how expensive it was to build and try things. Bring those costs down and a thousand flowers will bloom.
But plant a thousand flowers and you'll pull ten thousand weeds. The tooling to plant has never been cheaper. The tooling to identify and kill what isn't working hasn't kept pace — but it's within reach, and I'll take it up in the next post.
I wonder what happens when the thing that used to slow us down — having to agree on what we were building — stops being necessary.