The SaaS Model Was Built for Humans. Agents Don't Care.
For years, I was the Salesforce guy at McKinsey. Any client question that touched CRM came to me, and I spent most of that time on calls with executives telling me a version of the same thing: I need to get more value from this, I hate this thing, and my reps hate this. Funny enough, some of those companies were up on the posters at Dreamforce.
The software was working as designed, though. That was the problem; it was designed around a person sitting down and typing information into a form so that someone else could read it later, a model that goes all the way back to Siebel. Now every vendor in enterprise software is racing to put artificial intelligence on top of that same design. Copilots, assistants, and agents are being bolted onto systems built from the ground up for human users, and the premises of software-as-a-service as they were written 10 years ago are being treated as settled; they are worth upending.
Agents aren't people; they don't log in or read dashboards. They need to pull context from across many systems at once, synthesize it, and then act on it, which is a fundamentally different access model than anything enterprise software was designed to support.
What's happening now is vendors trying to sneak AI in through a side door, letting agents peer through the curtain at data that was never structured for them, and that's just not going to work.
Here is what it looks like when it fails: most CRMs show a red, yellow, or green health score on every deal, and a rep can read it in half a second. The score is stored as a field, so an agent can pull it without any trouble, but can't pull the deeper context on why the color is what it is. Sales management doesn't fully trust the score, which is why they end up asking the reps questions anyway: What happened on the last call? Who has gone quiet? What does the color actually reflect? An agent has no one to ask, so it takes the color as truth. Asked which deals are at risk, it hands the chief revenue officer a confident answer resting on a number sales management never trusted on its own. The company never notices until a deal everyone thought was green falls out of the forecast.
We've had workflow automation for years, but AI adds the ability to reason; it can think through a problem, find a pattern, and come up with a solution rather than execute the next step of a sequence. To do that well, it needs the right context and the right information, served in a way that makes sense for how agents actually work. The questions themselves have not changed, and they are the ones revenue leaders have always asked: How do I get to my number? Which deals are at risk? Where should my team focus? What's new is that an agent might be the one working toward those answers.
Most vendors haven't rethought what it means for their products. They're still building for maximum human visibility, capturing everything, and presenting it to a person. The agent just becomes another consumer of a system that wasn't built for it.
And a lot of vendors are making their customers do work they shouldn't have to do, deciding what stays with humans and what goes to agents, and figuring out which information lives where and which systems agents can actually reach. That work belongs to the vendor, and most vendors haven't done it yet. The underlying capability should work regardless of who or what is consuming it, human or agent; the process should be fluid.
Give it 12 to 18 months. Vendors bolting AI onto human-built systems are going to hit a wall, and it will not be subtle; their agents will fetch data just fine, but they will not reason well over it, because nobody rebuilt the foundation underneath. Customers will start noticing the difference between a product with an AI feature and a product actually built for AI, and that gap will show up in renewal conversations. Then, some of those renewals will not happen. We're at a break point. The SaaS model is changing. Agents need to get information themselves, synthesize it, and have the agency to act. Software that can't support that isn't ready for what's coming.
The companies thinking clearly about this are redesigning their product experiences for both humans and agents from the ground up. They are starting from whether their architecture actually supports agents operating inside it, which is a harder question and a better place to begin.
Jason Ambrose is CEO of Backstory.