A lot has changed in just the last month, its felt like a year! Everyone got hooked on Claude Code over the Christmas break and OpenClaw has seem massive adoption (I used it to pay a parking ticket for me and it worked pretty well).

It's pretty easy to extrapolate this trend over the next year: AI agents are becoming the primary users of software.

They call APIs, write to databases, browse the web, and coordinate with other agents. The DOM is way too slow for them. They're going to need API access to your system to get things done.

Whether thats an API or an MCP sever (Linear does this really well): companies that expose their systems to agents will win.

What dies

Seat-based SaaS pricing is the first casualty. If one agent does the work of five employees, the customer won't pay for five seats. They'll demand usage-based or outcome-based pricing instead.

The evaluation criteria for enterprise software also change. Companies used to pick tools based on UI polish and feature breadth. When agents do the clicking, those criteria become irrelevant. What matters is API depth, data extractability, integration flexibility, and latency. A product that an agent can't talk to is a product that gets replaced.

This kills a specific category of software: the UI wrapper. Any product whose value was "a nicer interface on top of underlying data" is vulnerable. An agent can replicate that value for almost nothing. The survivors will be systems of record that own critical data and workflow orchestrators that companies can't operate without.

What gets built

Startups for the agentic era will look different from traditional SaaS companies. They will have minimal UIs, extensive APIs, and maybe a few hundred power-user customers who deploy the product across their entire organization via agents. The human interface becomes a place to do spot checks, not the main way work gets done.

The core design principles are straightforward. Make the AI the primary user. Provide comprehensive, well-documented APIs. Treat your product as infrastructure, not a casual web app. The agents move faster than humans by an order of magnitude and your infra will need to be able to handle it.

A big opportunity might be replacing any process where multiple software tools exist and people glue them together, an agent can replace the glue.

In each case, the pattern is the same. A vertical workflow that required multiple tools and people shuttling data between them gets replaced by a unified AI-driven solution.

Most incumbents wont adapt

The interesting question is why existing SaaS companies can't just add a couple API endpoints and keep going. They could, but they will also need to change their billing model to not sell seats.

The public comapnies will be hit the hardest, becuase at least if youre private, you dont need to hit guidance numbers.

The real constraint

The bottleneck for these new startups wont be technology, software is cheap now! The bottleneck will be trust. No company wants a rogue agent accidentally deleting production databases or approving $50,000 refunds without oversight.

Startups that solve for trust will win. Building guardrails, human-in-the-loop approvals, audit trails, and slow progressive rollouts. You can start by recommending actions. Then auto-execute the low-risk ones. Then expand scope as confidence builds.

The companies that get this right will be incredibly sticky. Once an agent is embedded in your critical workflows and proving itself reliable, switching costs are enormous.

"software is eating the world" turned into "AI is eating the software."