In late 2023, investor David Friedberg declared on the “All In” podcast that AI was going to kill vertical SaaS. He argued that AI will enable companies to build their own internal software solutions, with no-to-low code, at a fraction of the cost of buying comparable software from third parties — predicting this would result in increased churn and pricing pressure on the SaaS sector.
AI has made it temptingly easy to originate your own tools. You can build a chatbot in an afternoon. A spreadsheet summariser before your next meeting. A lightweight CRM in a weekend. Lovable, Bolt, Replit, Cursor, Windsurf — and whatever comes next — have siren calls that can be hard to resist, especially for folks who are builders at heart.
But just because you can doesn’t mean you should.
The allure of building
Many years ago, I worked at a hugely successful company that had built everything themselves — because they were engineers; because they wanted to; because they could. They clearly didn’t have the “build vs buy” conversation because “build” had prevailed every time.
Fast forward 15 years to when I joined… the company was an iconic market leader and the embodiment of innovation — on the outside. But on the inside? Teams were tearing their hair out over antiquated, no longer fit for purpose systems. The finance team struggled to access the information they needed, and the HR team was constantly undermined in their ability to support a modern day workforce, let alone enable it.
The problem was these systems had been perfectly fit for purpose when they were initially built. But over the years, despite requirements continuing to accrue, they were increasingly orphaned, as investment went into the commercial products that brought in the revenues. Plus the team members who had originally built the systems had often moved on, leaving little to no documentation, and new hires didn’t want to work on systems that used antiquated code.
To build or not to build?
At Olio, before we dive in to building any tools we ask ourselves three litmus-test questions:
1/ Is this a standalone piece of functionality?
If it’s small, self-contained, and low-risk — great! Build away. Worst case, you bin it later. No harm done. But beware: standalone rarely stays that way. Today’s “quick script” becomes tomorrow’s legacy system, and a spaghetti junction of ‘standalone’ tools bolted together becomes a painful headache.
2/ Is it core to your USP?
If this functionality is central to your unique value proposition — something that will truly make or break your differentiation in-market — then yes, investing in bespoke AI tools might be wise. But be prepared to keep investing. We’ve recently been looking at AI-powered customer support tools for example, and whilst the temptation to build our own has been strong, it’s clear that the market is evolving so rapidly that we’re going to want to take the regular upgrades from an agile market leader, rather than have to build those upgrades ourselves.
3/ Are your requirements truly unique?
This is the trickiest one. Because so often the team will insist that our requirements really, truly are unique. However what feels unique is often just a workflow that hasn’t been optimised around the tools that already exist. Or it’s a ‘bell’ or ‘whistle’ that if we’re being honest, we can do without, or work around. If you genuinely have a use case that no existing software can handle, then maybe, just maybe, building makes sense.
Join the dots before you code
Sometimes the real magic isn’t in building a tool — it’s in joining up the ones you already have. Enter tools like Zapier and Make. They can stitch together your SaaS stack in sophisticated ways, letting you automate, customise and streamline without writing a single line of code. It’s less sexy than spinning up a custom LLM agent… but it’s faster, cheaper and infinitely easier to maintain.
There’s more to tools than just code
Finally, for when you do decide to code, remember that even the simplest tool needs a lot more than code to function well:
- User journeys need mapping
- UX and UI need designing
- Edge cases need catching
- Maintenance and scaling need resourcing
Skip those steps, and you’ll end up with a “feature Frankenstein” — a tool everybody thinks they want, but nobody actually uses.
So yes, AI is unlocking incredible new possibilities. But take a beat before you build. Ask the hard questions, and don’t forget: sometimes, the smartest thing you can do… is not build at all.
