The End of Syntax: Allocating Agentic Compute

The Definition of a Developer Has Fractured

GitHub recently stated their goal is to support a billion software developers. When I first read that, the operator in me scoffed. There are only about 30 million professional software engineers on Earth right now. To hit a billion, you don't just train more people — you have to fundamentally redefine what the word "developer" means.

Look at how work is actually getting done today. A recent write-up from Czechia detailed a user managing a complete IT project assigned by voice from a train, entirely through AI agents. No IDE. No mechanical keyboard. Just natural language and intent, streamed over a mobile connection while moving at 150 kilometers per hour.

This is not a cute parlor trick. This is an order-of-magnitude shift in leverage.

When I was running engineering teams at ShopBack and building out infrastructure across APAC, our bottleneck was always translation — translating business logic into architecture, and architecture into syntax. That translation layer is collapsing. A developer is no longer someone who writes syntax; a developer is someone who allocates agentic compute. The billion developers GitHub is talking about aren't going to be writing Python. They are going to be managing fleets of autonomous workers.

Infrastructure Trumps Intelligence

Here is a hard truth about the current AI landscape: everyone is obsessed with the intelligence of the model, but the bottleneck has already shifted to the infrastructure hosting the agent.

A demo on a MacBook is easy. You string together an LLM, a local vector database, and some Python scripts, and it looks like magic. But deploying that to production? Handling state, managing timeouts, dealing with rate limits, and ensuring an autonomous agent doesn't bankrupt your AWS account in a recursive loop? That is an entirely different sport.

This is why the news that Cloudflare is expanding its Agent Cloud for AI software agents matters far more than whatever benchmark a new foundational model just barely beat.

Agents need to live somewhere. They need to run close to the edge, they need to be secure by default, and they need infrastructure that understands they are fundamentally different from traditional stateless microservices. If your AI strategy relies on long-running scripts on an isolated EC2 instance, you do not have a strategy — you have a liability. The winners in this next cycle won't necessarily be the companies building the smartest agents. They will be the companies that figure out how to run thousands of average agents reliably, concurrently, and cheaply.

The Deployment Hangover

We are finally entering the trough of disillusionment, and it is about time.

You can see the sentiment shifting at the very top. Reports are surfacing about Sam Altman’s second thoughts regarding the growing AI backlash. The hype machine promised that AI would seamlessly integrate into every enterprise overnight. The reality? Legacy systems are messy, data is siloed, and enterprise compliance moves at the speed of a glacier.

I spent years at Amazon and Digital China watching enterprises try to adopt "the next big thing." The pattern is always the same. Executives buy the vision, middle management panics, and engineering teams are left to stitch together a prototype that inevitably fails to scale. The backlash we are seeing now isn't because the technology doesn't work. It is because the market confused a technological breakthrough with a product lifecycle.

A foundational model is a raw material. It is not a product.

We have three domains in this life: career, family, and finance. Time is the ultimate constraint across all of them. The operators who compound their value over the next decade won't be the ones chasing every new model release on Twitter. They will be the ones who ignore the hype, accept the backlash as a necessary market correction, and put their heads down to do the unglamorous work of actual integration.

Stop building toys that look good in a boardroom. Build systems that survive contact with the real world.