Drying the Moat
The $1 trillion lesson in what happens when AI can finally read your code.
A few years ago, a computer science student asked me for career advice. I shared the bigger picture, based on my experience working for tech giants in Silicon Valley, Singapore, Australia, and China: follow your curiosity, challenge everything, understand the business side, and build, build, build. But I also shared a quick hack: “If you want to make money quickly, learn COBOL.”
It seemed like safe advice. An estimated 95% of ATM transactions in the US still run on COBOL. So do pension systems, tax processing, and welfare payments across dozens of countries. My driver’s licence is a record stored in a system that looks like a hacker screen from Terminator 2. The language, born in 1960, still powers the backbone of banking, insurance, airlines, and government. And the people who understand it are retiring faster than they’re being replaced. Supply: low; demand: high. Easy money.
I’m glad she didn’t listen.
On Monday, IBM lost $40 billion in market value, its worst day in 25 years, after Anthropic demonstrated that its AI can map, analyse, and help modernise COBOL systems. The kind of work that once took consulting teams months can now be accelerated to weeks by AI.
Here’s the thing that puzzles me, though: Anthropic didn’t launch a new capability. They simply showed one that’s been possible for months.
So, what’s happening? The tools haven’t improved, but awareness of them has. And that might be the most important lesson here. Most people don’t grasp what AI can do until someone demonstrates it on something they recognise. COBOL is just the demo that clicked. They’re pointing to the moon. Don’t focus on the finger.
There’s so much more to discover with these tools. I kid you not: I’ve recently been walking around like a zombie, because I’ve been spending entire evenings, way past my normal sleep time, building tools that extract and bring together my personal information from literally everywhere. I am extracting my personal data from everywhere. And it works. Walled gardens? Shmalled gardens. But I am digressing.
Or maybe I am not digressing at all: the headlines are missing the bigger story. IBM’s moat was never really COBOL. It was the incomprehensibility of COBOL. The real lock-in across enterprise software has always been information asymmetry: your vendor and their consultants understood your systems better than you did. They knew how to access and work with your data. You didn’t (you know I am right, you just had a meeting with a vendor to understand what your enterprise system is capable of).
Migration was expensive because the old system was impossible to read.
I have a name for what’s happening right now: drying the moat. (No references to any current presidents here). A business model for many enterprise companies has been to fill a ditch with water and charge you for the bridge. AI can dry any ditch almost overnight. The castle is still standing, but now everyone can see it’s just a building. And the moat was just a ditch.
Welcome to the DRY MOAT world
COBOL is only the beginning. SAP? 250+ million lines of ABAP code, understood by a shrinking pool of specialists (and to prove the point, SAP now has an AI agent to help, actual kudos for being proactive). Oracle’s PL/SQL, Salesforce’s Apex, and ServiceNow’s platform logic are all moats built on the same assumption: switching costs would always exceed staying costs. Over $1 trillion has evaporated from enterprise software stocks in 2026. The “SaaSpocalypse” is a repricing of what complexity is worth when AI can decode it.
Now, translating code isn’t the same as modernising a system. Sure. IBM is right that decades of integration can’t be replicated by converting syntax. The market is probably overreacting in the short term. But the direction is irreversible.
In the last two weeks, I’ve built more software than in the previous ten years. That’s not a metaphor. No matter what legacy system your organisation is sitting on, I’m confident that someone with the right AI tools could walk in, extract, and migrate your core logic within weeks. Not years. Not quarters. Weeks. I am confident that even I could do it.
So if the old competitive advantage, hoarding technical complexity, just evaporated, what’s the new one?
Here’s my bet: when building and migrating are nearly free, the scarce resource shifts from technical capability to curiosity and imagination. What’s something that you could offer in your ERP system that no one has even considered? You’ll have a short window of time to capitalise on it before everyone else catches on. And when they do, you’ll have to wow your customers again, to keep them with you.
That’s what a dry moat looks like from the other side. The water’s gone. Now you actually have to compete. Call me crazy, but I think the most exciting times in software development are ahead of us.
Alright, short post today. I need to get back to building, building, bulding.