This is Part II in a series on how agentic AI will drive category consoliation. Head here for Part I. To make sure you get every issue, subscribe here.
You’ve probably heard the phrase, “Nobody got fired for buying IBM.” It’s a story that dates back decades, when IBM sold mainframe computers. Yet the dynamics that gave rise to such a statement foreshadow what may happen with agentic AI platforms. Today, we’ll find what that connection is, so you have a better sense of what’s to come in the age of AI.
The Lesson from IBM Mainframes
By the mid 1970s, IBM was in an enviable position. Not only was it generating north of $14B in revenue each year (about $86B in today’s dollars), it reportedly had up to 70% market share in mainframes. IBM’s position in the category was so dominant, in fact, that its competitors were called “The 7 Dwarves.“
There are many reasons for IBM’s success, but one of the biggest drivers came from IBM’s decision to create a computing platform that could evolve with a business, while competitors offered fragmented solutions that were onerous to upgrade. Here’s what I mean.
Back in the 1960s and 1970s, computer operating systems, hardware, and software were not interoperable the way they are today. Most mainframes were vertically integrated and essentially proprietary. Buying from IBM meant buying IBM’s operating system, IBM’s hardware, and IBM’s software. Same story if you bought from a competitor, like Burroughs or Honeywell.
There was a big problem with this approach, though. Upgrades could mean putting your entire computing investment at risk. Imagine wanting to move from macOS Sequoia to Tahoe, but finding out you’d need to replace your laptop and every piece of software you’ve purchased. But that’s how computers worked then.
One of the biggest drivers of IBM’s success came from its decision to create a computing platform that could evolve with a business, while competitors offered fragmented solutions that were onerous to upgrade.
At least, until IBM made a $5B bet.
In 1964, they introduced System/360. It was a family of computers that had a compatible architecture across different components. For the first time, a business could make an upgrade without having rewrite its software, retrain its staff, or go through an expensive integration process. And this wasn’t a one-time release; System/360 was a new architecture that would provide continuity for years.
It was such a game-changer that, within a few years, IBM mainframes became the industry standard. Jim Collins, the author of Good to Great, even ranked it as one of the top three business accomplishments of all time.
With a proven architecture that could grow with a customer, a strong brand and reputation, and a growing ecosystem of available software, there was little reason not to choose IBM. What an enviable place to be, huh?
I’d love to share more of the story, but we’re here to talk about agentic AI. What’s the connection?
Preferences vs. Incentives
In his work, Micromotives and Macrobehavior, economist Thomas Schelling famously showed how choices are often driven less by what people prefer and more by the incentives and constraints that shape those choices. Or as Charlie Munger is famous for saying, “Show me the incentive and I’ll show you the outcome.” This same dynamic was at play with IBM’s mainframe business, and now, it’s starting to play out with agentic AI.
Here’s an everyday example of how this works: buying your next car.
If you’re shopping on preference, you might choose a car based on styling, brand, or features that make the driving experience more comfortable. You might buy a Jeep Wrangler because it looks cool, a Range Rover to show your neighbors how successful you are, or a Subaru WRX because you love how fun it is to drive. In other words, you’re shopping based on your tastes.
But buying based on incentives means you have hard constraints.
For example, if your long commute means fuel costs are eating up your budget, you might buy a Tesla Model 3, even though it doesn’t suit your tastes. My wife’s car had even tighter constraints. With four kids, a minivan was the only option that could hold everyone and leave enough space in the garage. And living in Colorado, so was all-wheel drive. Her incentives drove her choices down to one: a Toyota Sienna.
In the mainframe era, a buyer might have preferred to buy from one of IBM’s competitors. Perhaps its software was better suited to its situation, or they offered an option that better aligned with its initial budget. But as strong as those preferences might have been, there was an even stronger incentive to buy from IBM (and stay with them). Since IBM was trusted, proven, and was less likely to cause headaches when it came time to upgrade, buying from IBM meant you were making a safe choice that wouldn’t put your career at risk.
Why This Matters for Agentic AI
That’s the same dynamic that will play out with agentic AI.
Today, some buyers might have preferences for “best of breed” solutions. But as we discussed in the last issue, AI agents require a unified data layer, the ability to coordinate, and context about what’s happening in the business to work well.
In the past, a “best of breed” approach meant an acceptable tradeoff: better functionality and no vendor lock-in in exchange for fragmented data. That’s a preference-driven choice. But with agentic AI, that equation changes. Fragmented data doesn’t mean a tradeoff; it means that agentic AI can’t even work reliably. That’s an incentive that buyers simply have to accept.
In the next issue, we’ll take a closer look at this and explore why agentic AI is only going to accelerate the consolidation movement that’s already in play.
See you then.
As the founder of Flag & Frontier, John Rougeux partners with executive teams to align on their strategic narrative, build belief in the market, and win the next chapter of their business. You can chat with John here or connect with him on LinkedIn.
