A Common Enemy: The Threat Facing SaaS and Agentic AI Platforms
They might be competing against each other, but both meta categories also face the threat of losing leverage over their customers.
This is the last and final part in a series about how AI is driving category collapse. Head here to read part I.
SaaS and AI might compete against each other, but they also share the same problem: they are at risk of losing leverage over their customers.
Yes, businesses already have less reason to pay a premium for SaaS, since AI agent platforms can let them build their own software. But similarly, why pay a premium for an AI agent platform if you can just use Claude Code directly?
Both categories face a common enemy: the threat that customers capture value for themselves and reduce their pricing power.
There are four things these businesses can do to regain leverage over their customers: control points, network effects, domain expertise, and architecture. And that’s what we’re going to unpack today: how software companies of either sort can use these levers to avoid getting squeezed out of the middle.
First, some definitions:
Legacy SaaS: any SaaS business that predates AI and relies primarily on subscription-based pricing.
AI agent platforms: tools that allow businesses to create and manage their own AI agents, often on top of SaaS products the business already uses.
LLMs: the models that AI agent platforms run on top of (Claude, Grok, ChatGPT, etc.).
Software companies: For this article, I’ll use this to refer collectively to both legacy SaaS and AI agent platforms.
To get started, let’s take a closer look at why software companies are at risk of losing leverage over their customers. Then we’ll unpack the four things software companies can do to regain this leverage.
When Customers Can Build Your Product, You Lose Leverage
In the past, SaaS vendors could do something their customers could not: create software at an attractive price point. For most businesses, the “make vs buy” case was pretty clear: buy. But now, customers have regained some of that ability. How much they have regained depends largely on the type of software at hand. But the trend is there.
What does this mean for SaaS vendors? If your customers can build parts of your solution themselves, your pricing power declines.
For example, let’s say a manufacturing business uses Acumatica, a common ERP. There are a few things the manufacturer wishes Acumatica could do, so it starts experimenting with agentic AI tools like Dust or Claude Code. As it continues to explore what’s possible, it reaches an “aha” moment. “These AI agents can replace features we were already using in Acumatica, but better,” they realize. They’re paying for features they no longer need. At a certain point, the manufacturer really only needs Acumatica to serve as a behind-the-scenes database, while AI agents handle all the functionality. Then there’s the uncomfortable phone call. “Why are we paying so much,” they ask, “if we only need your back end?” Acumatica is forced to make pricing concessions.
And so it goes. This is how you lose power over your customers.
I’m not trying to claim that SaaS will go away completely. But if SaaS vendors have less value to offer, their pricing power will erode.
AI Agent Platforms Aren’t Immune From This Pressure Either
Does this mean that AI agent platforms are in the clear? Not so fast. They provide the tooling that enables customers to create, manage, and orchestrate their own AI agents. Many of them are wrappers around existing LLMs. What they’re selling (at the moment) is essentially middleware that bridges the gap between SaaS and LLMs.
Both SaaS and AI agent platforms face the same problem: they must retain power over customers and avoid commoditization.
Like SaaS, I haven’t heard a good case for why AI agent platforms will have much power over their customers either (if you do, let me know in the comments). What’s to keep a business from using one AI agent platform over another? Is one offering that much better than the other? Are there switching costs? Or worse, what will keep a customer from cutting an AI agent platform out of the picture completely and going directly to Claude Code? It’s too early to know for sure, but if I were running an AI agent platform, I would have this on my radar.
The point I want to make is that both SaaS and AI agent platforms face the same problem: they must keep their customers from reducing their pricing power.
4 Ways Software Can Regain Leverage
Let’s see what a software business could do about this. So far, I’ve found four ways they could regain leverage over customers: own control points, create network effects, leverage domain expertise, and build superior architecture.
Own Control Points
This concept comes from Dave Yuan at Tidemark. The idea behind a control point is simple: they are the core functions that businesses rely on to operate. As Dave explains:
Control points, as a function of their workflow and data gravity, inherently enjoy an unfair right to sell multiple products to their merchant customers, even potentially becoming an “operating system” within their vertical. This remains true not only for traditional software products but also for new AI offerings. (source)
Here’s an example from my own business. I use Gusto for payroll processing (yes, I pay myself as a W-2 employee, but that’s a topic for another day). Gusto “owns” the flow of money and data between my payroll system, my bank accounts, and my state tax accounts. This gives Gusto a degree of control, because ripping out and replacing all those connections is something I’d be loath to do.
But Tidemark also points out that control points “…have the unfair right to offer most products a merchant needs.” That’s exactly what Gusto did recently, when they purchased 401 (k) provider Guideline. Now, Gusto manages my payroll and my 401(k) plan (which is also thorny to set up, by the way). Now, Gusto owns two control points. They’ve gotten stickier, which means it’s harder for me to disintermediate them. Good business move. This is the kind of play every software vendor needs to explore.
Foster Network Effects
Network effects occur when more value accrues to customers when other customers use the same product. Facebook and LinkedIn are classic examples. One way network effects come into play with software is through the ability to improve outcomes for AI agents. As more customers join a platform, there’s more data for agentic AI to learn from, which means better results for everyone. The question is: how much value does this create relative to what a customer would get by building their own software? Your mileage may vary. But keep it on your radar, especially because network effects can edge competitors out, too.
Leverage Domain Expertise
I’ll use the Gusto example again. There is so much involved with reporting data on payroll taxes, getting these taxes paid, and doing everything in a compliant way. It’s not something you can get “mostly right.” It needs to be 100% right. Software vendors that know how to navigate compliance or regulatory territories have an advantage over their customers because they know how to solve something their customers don’t. Being able to vibe code doesn’t mean you have the domain expertise to tell those agents what to do in the first place. This isn’t to say that domain expertise can’t be obtained by LLMs themselves (see Anthropic’s legal offering), but at a minimum, it buys some software companies time to figure out the rest.
Build Superior Architecture
I may get in trouble with this one. I’m not a technologist, and I don’t claim to understand all the nuances of how technology platforms work. But here’s what I’ve learned from people who do. There are essentially two layers to software: the experience layer and the architecture layer. The experience layer is the interface where features and functionality live. The architecture layer is the behind-the-scenes foundation that allows the experience layer to work.
Historically, SaaS companies handled both. But with vibe coding, that changes.
The experience layer is now something customers can build directly. But if they want AI agents to perform well, the architecture layer becomes even more important to get right. (In part I of this series, we described this as a unified data layer). What engineers have told me is that building high-performance architecture is no trivial thing. So while customers can vibe code AI agents themselves, building the right architecture for those agents to work may be too formidable an undertaking. This provides an opportunity for software vendors, both legacy SaaS and AI agent platforms alike, to regain control.
We’ve Come Full Circle. AI Will Drive Consolidation
I noticed something else while writing this.
These four levers not only give software companies leverage over customers, they also give them leverage over each other. In other words, if a given software vendor were to gain ownership of control points and create more value through network effects, domain expertise, and superior architecture, they’d be more likely to win on all fronts. Competitors without such leverage wouldn’t just be reduced to a commodities, they might also be edged out of business altogether.
And so we have come full circle.
At the beginning of this series, I posited that the need for a unified data layer would push out point solutions and favor consolidated software offerings, driving “category collapse” from one vector. Now, we’ve seen that category collapse will come from a second vector: as software companies try to regain leverage over their customers, they will edge out each other in the process.
Interesting times ahead. Good luck.
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.
Great write up John! You taught me a couple new moats.
Founders need to be way more thoughtful about what and how they build. Moat building is existential now vs just a few years ago when all u needed was a big funding round.
Really exciting times and I think we are going to enter an era of really great software. Anything short of world class won't make it...exciting for the consumer!
John, great piece and you've nailed all four levers. As an electrical engineer with 30+ years in technology and digital transformation, the architecture layer point really resonates. The Base 44 ad that aired during the Super Bowl kinda pissed me off if I'm being honest. It reminded me of big pharma commercials that encourage patients to ask their doctor about prescribing a specific drug to them. Clearly the pharma company's priority is getting their drug adopted, not necessarily the patient's best health outcome. Similarly, Base 44 celebrated how easily anyone can build the experience layer while completely ignoring what's underneath. Obviously, their priority is gaining a foothold in your organization, not necessarily doing right by it. And sure, if you've ever been frustrated by your IT department's reluctance to give you what you need, that ad must have been music to your ears. But for any organization with shadow IT policies and data governance requirements, it should be a very real wake-up call. I wonder how many people jumped on Base 44 the Monday after the Super Bowl and started vibe coding apps? In my hands-on experience with tools like Base 44 and Lovable, they're genuinely impressive for prototyping, but the back-end data models they generate are rarely well-architected unless the person prompting them has enough software engineering expertise to guide the design intentionally. That's the catch: you still need domain expertise to know what to ask for in the first place. These platforms will get there eventually, but for now the architecture layer remains a meaningful moat, and to your point, smart software companies should be doubling down on it.