Why enterprise AI comes down to coordination and understanding.

If someone asked you to name the single biggest takeaway from the current wave of AI-driven software development, what would you say? Faster coding? Better automation? Smarter models?

Here's what I'd tell you, based on what we're seeing every day working with some of the largest enterprises in the world: none of that matters if you haven't solved for understanding and coordination at scale. Those are the two real challenges. Everything else is incremental.

And here's the thing, these aren't new problems. They've been around for decades.

The problem didn't start with AI

Let's go back to 1968. Southern Germany. A NATO conference. That's where the software development lifecycle (SDLC) was born, and it wasn't born out of ambition. It was born out of necessity.

Enterprise systems had grown too large for any individual to fully grasp. We're talking 10,000 lines of code, which sounds almost quaint today, but at the time it was genuinely overwhelming.

Languages like COBOL, PL/1, and assembly were cutting edge. And because these systems were so critical, engineers had to develop a coordinated, structured approach to building and maintaining them across teams.

Fast forward 58 years. The context problem hasn't gone away. It's just changed shape. Back then, it was a carbon-based context problem, a human being trying to hold too much in their head at once.

Today, it's a silicon-based one, an AI agent operating within a limited context window. And many enterprises are still reaching for that same 1968 solution, trying to coordinate their way around the problem through an SDLC that was designed for a fundamentally different era.

Simple LLM harnesses beat hand-crafted RCA agents

A new arXiv study pits simple LLM harnesses against five published root cause analysis agents. The plain setups won on accuracy and cost…

AI Accelerator InstituteAndrew Lovell

Why startups and enterprises aren't playing the same game

You've probably seen the headlines about a 19-year-old dropping out of Stanford, vibe coding their way to an MVP, raising $5 million, and posting about it on LinkedIn. And honestly? Good for them.

They really are building faster than ever before. Coding, for a greenfield project, is essentially a solved problem at this point. Go grab your favourite model after reading this and build something. You'll see.

But here's where I'd push back on the narrative: the reason that 19-year-old is moving fast has very little to do with being "AI fluent" or a brilliant prompter. It's because they're operating in a completely different problem space.

The architect who's been at a bank for 30 years isn't less capable. I work closely with these people. They're brilliant engineers. They're dealing with a fundamentally different set of constraints, and that difference matters enormously.

The two gaps that are actually holding enterprises back

Every large enterprise I've worked with is dealing with some version of two core problems.

The understanding gap. Take a 20 million line ecosystem for an insurance company. Car insurance, home insurance, life insurance, each with its own application, its own dependencies, its own history. A coding agent can typically see around 20,000 lines of code at a time.

And software is relational. You can't just grab isolated pieces and expect to understand how the whole thing behaves. The dependencies matter. The interdependencies between services, between teams, between legacy components and modern ones, those connections are where the real complexity lives.