XY@xydotdot
Agentic Programming Thoughts...
I have a few predictions about where vibecoding is going, and it starts with a simple observation.
Language is becoming the programming layer.
As vibecoding and its derivatives mature, the ability to deploy software will converge toward the ability to express intent in English. When that happens, the cost of building applications trends toward zero, and the time from idea to deployment compresses to minutes.
That switch removes the main constraint that shaped software for decades...COST
When building was expensive, solutions had to be bundled.
Once building becomes cheap, bundling stops making sense.
My first prediction is that this creates an explosion of micro-apps.
See, when creation is cheap, specialization becomes rational.
The issue with this premise is that it will lead to fragmentation. Each micro-app works well in isolation, but the surface area between them will grow faster than anyone can manage.
And that will be a huge issue.
The limiting factor now becomes coordination.
In a world of millions of micro-apps, value accrues to whatever reduces coordination cost. The hard problem becomes getting small tools to understand each other, trust each other, and operate inside a shared system without constant manual stitching.
My second prediction is that interoperability becomes the dominant source of power in this cycle.
Someone will have to build interoperability infrastructure as the top layer, one that defines how identity works across apps, how permissions are enforced, how context is passed, how memory persists, how actions are sequenced, and how failures are handled.
It will basically be a set of architectural rules that micro-apps must follow to participate in a larger workflow.
This layer will be insanely valuable and will be the next big tech that garners power-law outcomes.
Each new micro-app that integrates increases the value of the system for every other app. Over time, the interoperability layer captures the network effects that application companies used to capture.
These predictions naturally funneled me to LLM concentration and forkability.
If a small number of models generate most software, then code becomes easy to reproduce. Forking apps becomes trivial.
So what isn’t trivial?
Reproducing context, distribution, trust, and integration.
Those assets accumulate at the interoperability layer. You can fork an app. You cannot easily fork an ecosystem with shared identity, shared permissions, shared memory, and established coordination norms.
My third prediction is that most durable value will not sit at the application layer.
Millions of useful micro-apps will exist. Very few will matter on their own.
The big winner will be the systems that everything routes through.
The layer that makes micro-apps composable.
When software creation becomes universal, control moves upstream.