How AI Is Making Software Vendors Easier to Replace

I was talking to a founder who builds software for banks, and we drifted onto a theme I have not been able to let go of since: AI is aiding disintermediation and quietly removing vendor lock-in.

If you think of old-school tech lock-in, perhaps the best example you can think of is banks and mainframes. The code is decades old, the vendor knows everything about it, and nobody on the inside wants to touch a migration. This is the founder’s business. Between the contract lengths and the termination fees, he told me, “It’s very, very sticky, and it’s very, very hard to move away.”

That stickiness has been the business model for a big chunk of enterprise software. I think it is about to stop working, and the way I would frame it is as an opportunity for small businesses, in two different directions.

Understand what made lock-in work

Software vendors in particular assumed that if they served a piece of your technical infrastructure, they would be tricky to get rid of. And for most of the last few years, this has been true. If you provided someone with their core banking infrastructure, or the CRM system they use to manage their sales and customers and so on, well, it was risky to migrate away from you. Companies like Salesforce have managed to latch onto a huge part of the digital economy simply by being the vendor of choice and making it very difficult, risky and cumbersome to leave.

Banking shows how extreme this gets. Reuters counted 220 billion lines of COBOL still in production, a 1959 language propping up 43% of banking systems. Leaving one of the big core providers, the founder told me, has meant “dedicated teams to work through migrations over the course of 18 to 24+ months in order to move from one provider to the other.”

But AI risks changing all of this, for a few reasons. The technical barriers are lower, and the incremental grunt work needed to migrate is less of an issue. AI can now write COBOL, and IBM sells a code assistant whose main pitch is translating mainframe COBOL into Java. Meanwhile, as a consumer, I can export all of my data from Google, from Meta, from pretty much any provider, and either migrate, share or just move it from one place to the other.

Perhaps ironically, this is also happening between AI agents. Tech connoisseurs find it easy to move from one model to another by simply migrating their context from one place to the other. The models themselves are not exempt from this.

Renegotiate everything at your next renewal

There was this belief for a while that everyone would just roll their own software and the cost of software would go to zero. I think that is very unlikely to happen. Public software won’t go to zero; the bar is just going to be much higher. Companies are unlikely to want to roll their own CRM or their own core banking system, simply because they want to focus on actually dealing with their customers, no matter how much AI capability they have. They want to use those AI capabilities to solve their customers’ issues, not necessarily to build a piece of infrastructure.

What is true is that migrating away from these vendors has become a lot easier, and there should be less fear of doing it.

The founder told me what he found when he went inside one client’s operation. “The complexity that exists as a part of their tech stack was kinda mind-blowing to me when I looked under the hood,” he said. “The fact that they still perform manual tasks to me is also quite surprising.” I would bet your own stack looks like that too, just smaller.

So before the next renewal lands, price the exit. Ask a developer, or honestly just point a coding agent at it, and get a rough scope for moving your data and your integrations somewhere else.

Then ask your vendor for a full export of your data in an open format, and watch what happens.

And then negotiate a lot harder. You can review your vendors with a lot more certainty now, and you basically do not have to worry about your expenses so much, because none of them are locked in the way they used to be.

Go after the incumbents’ customers

The second direction is bigger, and it is the one I find more interesting: taking business away from big legacy companies.

Every industry has its own version of the core providers — vendors that got comfortable because their clients could not realistically leave. “You’ve only got two or three providers that primarily power these institutions,” the founder said of his corner of the market, and “those core providers do not enable them to do what they want to do.”

Where some vendors have gotten lazy because they knew the client could not leave, you can now step in and give those clients what they actually need, and offer to handle the migration itself, the part that used to kill these deals. The pitch is the same exercise you just ran on your own vendors, pointed the other way: Show up with their exit already scoped.

I actually ran this whole theory past him on the call: the idea that the locked-in vendors got lazy and that leaving is just not that hard anymore. “I think you’re spot on,” he said. You are not stuck anymore, so act like it, starting with whichever renewal hits your inbox next.

I was talking to a founder who builds software for banks, and we drifted onto a theme I have not been able to let go of since: AI is aiding disintermediation and quietly removing vendor lock-in.

If you think of old-school tech lock-in, perhaps the best example you can think of is banks and mainframes. The code is decades old, the vendor knows everything about it, and nobody on the inside wants to touch a migration. This is the founder’s business. Between the contract lengths and the termination fees, he told me, “It’s very, very sticky, and it’s very, very hard to move away.”

That stickiness has been the business model for a big chunk of enterprise software. I think it is about to stop working, and the way I would frame it is as an opportunity for small businesses, in two different directions.