I was catching up recently with an old colleague who works in compliance at a financial firm. They’ve implemented vendor systems that work. The platform’s not perfect, but nobody loses sleep at night worrying that they’ve accidentally aided terrorists or money launderers.
But those vendors are not free. So, my colleague has been repeatedly asked whether the vendors could be “replaced by AI.”
They’re not alone. Vendors tell me that nearly anything they pitch now runs into “we can build it ourselves.” Business folks tell me they’re being pushed to ditch long-standing vendors so that IT (or vibe coders) can take a crack at building CRMs (Customer Relationship Management systems, like Salesforce), KYC (Know Your Customer) compliance platforms, accounting systems, and the like.
The appeal is obvious, but we’ve seen this movie before.
Why we love to build
Technologists have always had a bias toward building instead of buying. It’s more fun to build. The satisfaction and positive attention from a build success are much greater than from a smooth vendor rollout. Nobody throws a party for a well-negotiated contract renewal.
And it plays to our egos. Most people in all fields think other people’s jobs are easier than their own. In technology, that reads as everyone else’s software looks easy to build.
The cat by the tail
In practice, building a proof of concept is easy. But that’s the bait that lures us into poor decisions that mire us in months or years of delays and dissatisfied stakeholders. Building production software is always (always!) more difficult than it first appeared. If your team argues this point, increase your skepticism about their experience level.
Even the worst legacy codebases incorporate years of learned experience: edge cases, exceptions, seemingly simple but surprisingly complex logic. Much of that can only be relearned by making the same mistakes again. One of the most common utterances heard from developers rewriting old systems is, “Oh! That’s why they did that.” Mature vendors have already made the mistakes and learned the lessons. And they did it on someone else’s dime.
As the line usually credited to Mark Twain goes, “A man who carries a cat by the tail learns something he can learn in no other way.”
The conventional wisdom
Hence, the conventional wisdom has always been to focus on your competitive advantage and bias strongly toward buy everywhere else. Trust the efficiency of the markets. If it’s easy to build a cheaper version of a vendor’s system, that vendor will soon be disrupted and a cheaper version will become available.
Note how infrequently this actually happens. That’s not evidence that the markets are inefficient. It’s the core of the argument. If this stuff were easy, the market would be full of cheap replacements.
What AI changed
AI disruption has created a whole new wave of companies deciding to build systems they used to buy. Certainly, there is some merit in this. There is software out there that adds little to the core capabilities Claude and ChatGPT provide out of the box: generic copywriting tools, “chat with your PDF,” meeting summaries, translation tools, grammar checkers, etc. Those categories got hit hard for good reasons.
The same goes for a lot of small internally built tools: report formatters, data cleanup scripts, lightweight data extraction. If you’re paying teams to maintain these tools, by all means, stop.
What these all have in common is thin logic, minimal domain knowledge, and a low cost of being wrong. The scope of what’s being pulled in-house now goes way past that. It has become ridiculous.
What AI didn’t change
AI changes a lot, but there’s no reason to think it changes the core economics of vendor specialization. If what a vendor does can be replaced cheaply using AI, then new vendors will arise to take advantage of that arbitrage, and they will be cheaper and/or better.
Operating, as they do, on the same planet as us, vendors also have access to AI. They get the same productivity boost, and they are better incentivized to act on it because they can spread the investments across hundreds of clients.
If a vendor has managed so far to evade disruption without innovating, it’s probably because their business model is more difficult to replicate than it first appears. The core use case, the one you’d describe in a sentence, usually isn’t the hard part. The crux is knowing everything the code has to handle and maintaining it securely at scale in production.
AI makes it easier to grab the cat’s tail. But, it doesn’t make the cat any less likely to scratch.
The cost of ownership
Yes, switching vendors will be painful, as it always is. But switching to an in-house build carries the same pain, plus more potential for error and a long tail of maintenance, which, BTW, includes keeping pace with new feature requests.
Once it’s live, someone has to monitor it, respond when it goes down at 2 a.m., and fix the bugs that only surface months later. Periodically, outside events will force updates: a security hole turns up in a component the tool relies on, or the cloud platform it runs on retires an old version. Either way, you’re upgrading or migrating whether you want to or not. A vendor spreads that work across all of its clients. When you build it yourself, it’s all yours.
Then there are the feature requests, and they keep coming. Industry practice evolves, regulations change, and new data becomes available. Vendors are always working through long backlogs of requests from their users, and the improvements those bring are part of what you pay for. If you build it yourself, each one becomes an incremental cost.
AI will make these changes easier than ever, but they still take human initiative and oversight. That’s an ongoing cost. And again, the vendors get the same AI boost.
Owning the build does give you more control over prioritization on that one product. But it also means those features compete for prioritization with all the other competitively mispositioned software you built. Congratulations, you’re now a small software vendor with exactly one customer.
What we should do differently
The wisdom holds. If it’s your competitive edge, invest heavily. If it’s not, bat it away as quickly as you can.
The one thing I would change is your approach to vendor evaluations. Expect a phase of rapid disruption, with many new vendors challenging older, slower-moving ones. Look at the market more often, take the newcomers seriously, and be prepared to take advantage of that shift.
But, no! Do not build your own CRM. And for heaven’s sake, do not build your own KYC platform.