Living in a Micro-Conway world
Conway’s Law: “organizations which design systems … are constrained to produce designs which are copies of the communication structures of these organizations.”
Here is what I see happening to software engineering organizations in 2026: systems increasingly get built out of subsystems the size of one human and their Claude.
Call it the Micro-Conway effect.
Conway’s Law: “organizations which design systems … are constrained to produce designs which are copies of the communication structures of these organizations.”
Here is what I see happening to software engineering organizations in 2026: systems increasingly get built out of subsystems the size of one human and their Claude.
Call it the Micro-Conway effect.
One human, one Claude#
The communication channel between a human and their Claude is thick. It’s easy to type into (actually, please don’t type, speak to it—much faster and richer still). What you type doesn’t have to be perfect. It can be a draft. It can be an imperfect version, and you can have several going in parallel.
Coordinating with other humans, by comparison, is just as hard as it ever was.
So, all other things being equal, subcomponents that fit one human are now more likely. I do not think Mr. Conway is surprised.
Is it better?#
I have spent much of my career and research on Rails applications that got too large (I wrote two books about them: Component-Based Rails Applications and Gradual Modularization for Ruby and Rails). From that perspective, I want to say yes. More people building smaller pieces that fit together, instead of one large piece nobody can tear apart, is a welcome change.
But I suspect it’s not ideal.
Think of the monolith whose boundaries nobody ever thinks about. Now think of the system built only out of human-plus-Claude-sized things. In neither case have we designed for the needs of the system. Its structure is a side effect of the affordances we had while creating it.
I’m pretty sure neither is ideal. (If you want to know what I’d aim for instead, I wrote about the destination of a modularization journey a while back.)
Collaborating in public#
Which brings me back to the more collaborative development environments I have been excited about: a coding agent in Slack that one person, or several people, can use in a public channel. People watching each other work with the agent learn to collaborate on writing code.
If the communication structure changes, Conway says the artifacts should change with it.
What will those look like?
Mention this post from your site: