Is SaaS Dying? How AI Agents Are Reshaping Software Buying
DoorDash's agent ordering launch signals a shift where AI agents replace app interfaces, forcing SaaS, Salesforce, Microsoft and Meta to rethink value.
What’s actually changing in how software gets used?
People are quietly stopping to open the apps they used to live in. Instead of logging into a recruiting tool, a CRM, or a food delivery app, they’re asking an agent to do the task and reviewing the result. DoorDash made this visible on September 30th when it announced a way to order through an agent instead of the app itself. That’s a consumer example, but the same pattern is showing up inside companies: agents reading contracts, sales notes, and email threads, then drafting the plan a person used to build by hand inside a piece of software.
The software behind the scenes still runs. The company that holds the customer records, the candidate database, or the sales history still gets paid, because someone still needs to store and serve that data. What’s eroding is the habit of opening the tool directly. That shift, repeated across millions of workers, is why there’s real debate about whether traditional SaaS pricing and stickiness can survive the agent era.
TL;DR
- Agents are absorbing the “information processing” layer of jobs, the part where someone finds, arranges, or acts on data inside a tool, which used to be the core reason people opened that tool every day.
- The underlying data and distribution still have value, which is why companies that hold customer records, candidate pools, or transaction history keep getting paid even as direct app usage declines.
- The agent itself becomes the sticky asset, because it accumulates months of trained-in context (approval rules, exceptions, personal judgment calls) that’s hard to migrate if a company wants to switch AI vendors.
- Major vendors are repositioning around three roles: owning the data layer, building the distribution or trust surface agents plug into, and offering their own agent to sit on top of that data.
- Salesforce, Microsoft, and Meta are all wrestling with the same dilemma despite being wildly different businesses, because all three combine proprietary data with interfaces people are starting to bypass.
- Switching AI vendors is becoming an organizational fight, not a simple procurement decision, because different teams count the costs and benefits of an agent differently and often can’t see how colleagues actually use it.
- Point-solution SaaS tools with a single, narrow job are the most exposed, especially when their value proposition was always a blend of storing, finding, arranging, and acting on information rather than one clear function.
Remy doesn't build the plumbing. It inherits it.
Other agents wire up auth, databases, models, and integrations from scratch every time you ask them to build something.
Remy ships with all of it from MindStudio — so every cycle goes into the app you actually want.
Why does this threaten traditional SaaS?
Most SaaS products never had one clean value proposition. A recruiting tool, for example, stores candidate data, helps you find candidates, helps you arrange that information into a shortlist, and helps you act on it by sending it to a hiring manager. That bundling was fine when a human had to do all four steps inside the same interface. It becomes a liability once an agent can do the finding and arranging part on its own, pulling data from a provider, drafting the shortlist, and handing it to a person who just reviews and approves.
Once that happens, the tool’s role narrows to being a data source or a workflow backend, not a daily destination. The subscription might still get paid because the data has to come from somewhere, but the relationship changes completely. The buyer is no longer paying for a seat used eight hours a day, they’re paying for access to a layer the agent calls on. That’s a much smaller, more commoditized slice of value than the SaaS vendor originally sold. This is why so many software companies are rushing to describe themselves as “AI native” at every turn: they’re trying to get ahead of being reduced to a data pipe.
What happens to the employee’s relationship with the tool?
The practical version of this shift looks the same across recruiting, sales, engineering, and product: someone used to open a dedicated application, do manual work inside it, and produce an output. Now they supervise an agent that does most of that work, and they open the original application less and less.
Over weeks and months, the agent accumulates things nobody ever wrote down. Which customers need security approval before kickoff. Which hiring managers want a certain kind of candidate profile. What a “good fit” actually means for a specific role, learned by feeding back who actually got hired. None of that institutional memory lives in the original SaaS tool. It lives in the agent’s context, often tied to one person’s setup or one team’s configured workflow.
That’s the twist: the standalone app gets easier to leave, but the trained agent gets harder to replace. If leadership later wants to switch AI vendors to cut costs or gain flexibility, they’re not just migrating a license, they’re trying to rebuild months of accumulated judgment that may not be written down anywhere else. That’s a harder, messier negotiation than a typical software renewal, and it’s one most organizations aren’t yet prepared to have.
How are Salesforce, Microsoft, and Meta each responding?
Three very different companies are converging on the same strategic puzzle: own the data, own the distribution, and figure out where the agent fits so you’re not just reduced to a backend.
One coffee. One working app.
You bring the idea. Remy manages the project.
Salesforce, the company that effectively invented SaaS, is leaning on its two long-standing advantages: a deep data layer from two decades of CRM records, and major distribution through events like Dreamforce. At Dreamforce on September 15th, Salesforce announced Koa, a reasoning model built on its Agentforce stack and tuned specifically for CRM work. Koa is currently in pilot and trained on synthetic data rather than live customer records, reflecting Salesforce’s long experience with how customer data actually flows. Rather than betting only on Koa, Salesforce is also positioning itself as an open gateway, reportedly supporting partnerships with other agent providers like Claude, so customers can bring outside agents into its data environment. The strategy stacks three layers: hold the data, provide the trusted surface agents plug into, and offer a native agent (Koa) as one option among others.
Microsoft faces a parallel version of the same problem. It holds a deep data layer through productivity tools like Outlook and Excel, and it’s pushing to be a control plane for agents (an effort reflected in tools like Autopilot) while those same productivity interfaces risk being bypassed as people ask agents to manipulate the underlying data directly instead of opening the app.
Meta, not traditionally thought of as a SaaS company at all, is moving in from the consumer side. Its Muse personal agent product expanded quickly into “Muse for small business,” adding connections to business accounts and tools like Shopify and QuickBooks. The logic is simple: people already use a personal agent for life admin, and the jump from personal to small-business use is short. Meta’s apparent ambition is to make Muse a kind of business HQ, where a small business owner’s agent handles finding customers and running day-to-day operations by tapping into existing SaaS data layers rather than replacing them outright.
Is software disappearing, or just changing shape?
The underlying message isn’t that software goes away. It’s that the value shifts to whoever holds the data layer, whoever provides the trusted surface agents operate through, and whoever builds the agent that customers trust enough to let run on sensitive information. Vendors that only sold a direct interface for humans to click through are the most exposed, especially smaller point-solution tools built around a single workflow.
For buyers, the practical complication is that evaluating software now means reasoning about where a vendor sits across those layers, not just what features it demos today. A vendor can look “AI native” in a sales pitch while still being structurally exposed if its core offering is a clickable interface that an agent can route around. That’s a harder judgment call than traditional procurement, and it’s part of why the agent-versus-software competitive landscape has gotten so much broader and harder to map.
Frequently Asked Questions
Will SaaS companies stop making money because of AI agents?
Not necessarily. Companies that own valuable data or provide distribution and trust tend to keep getting paid even as direct app usage drops, because agents still need somewhere to pull data from and some trusted surface to operate through. The risk falls hardest on tools whose main value was the interface itself.
What does “data layer” mean in this context?
It refers to the proprietary information a company holds, like customer records, candidate pools, or transaction history, that agents need to do useful work. Companies with a strong data layer (Salesforce’s CRM data, for example) are in a stronger position because better AI models make that stored data more valuable, not less.
Why is switching AI vendors harder than switching software?
Plans first. Then code.
Remy writes the spec, manages the build, and ships the app.
Because an agent accumulates unwritten, trained-in context over time, like approval exceptions or personal judgment calls, that often isn’t documented anywhere else. Migrating to a new vendor risks losing that accumulated knowledge, and different teams inside a company may not even agree on how much value it holds.
What is Salesforce’s Koa, and is it widely available?
Koa is a reasoning model built on Salesforce’s Agentforce (Neotron-based) stack, tuned specifically for CRM tasks and announced at Dreamforce on September 15th. It’s currently in pilot, trained on synthetic rather than live customer data, and is one of several agent options Salesforce supports on its platform.
Should businesses stop buying point-solution SaaS tools?
Not automatically, but they should look closely at whether the tool’s value is mainly the interface (which an agent can bypass) or the data and workflow logic behind it (which tends to hold up better). Tools bundling storage, search, and action into one vague value proposition are the most exposed to agent disruption.