The SaaS Companies That Win Will Be Agent-Ready
Codex, Claude, and other AI agents are going to put pressure on traditional SaaS. The companies that survive will expose their workflows, reduce friction, and become tools agents can actually use.
I wrote recently that the next wave of SaaS is trust.
I still believe that.
But I think there is another layer underneath it now.
The next wave of SaaS will also be about whether AI agents can actually use the product.
That sounds like a small technical detail. I do not think it is. I think it is going to decide which products become more valuable and which ones get quietly routed around.
Codex, Claude, ChatGPT, and the agent systems built inside companies are going to change what users expect from software. A user is not only going to ask, “Can I log into this app and do the work?”
They are going to ask:
Can my AI do the work here for me?
That is a different test.
SaaS is getting squeezed from both sides
AI is lowering the cost to build software.
That means more people can create apps, prototypes, internal tools, workflow automations, and polished product surfaces. A smaller team can get further. A solo builder can ship something that used to take a small company.
That is one side of the squeeze.
The other side is customers.
Customers are also getting better tools. They can use AI to write code, generate reports, move information between systems, analyze data, draft emails, create dashboards, and operate across tools.
So the question becomes uncomfortable for a lot of SaaS products:
If an AI agent can do most of this workflow directly, why does this product need to exist in its current form?
Not every SaaS app is at risk in the same way.
But a lot of software is basically a wrapper around forms, tables, dashboards, notifications, permissions, and workflow steps. That can still be valuable. But if the value is mostly “click through this interface to move data from one place to another,” AI is going to put real pressure on it.
The user does not want more tabs.
The user wants the work done.
The old SaaS model assumed humans in seats
A lot of SaaS pricing and product design still assumes a human logs in, clicks around, fills out forms, exports reports, routes tasks, and updates records.
That model is not going away tomorrow.
But it is going to get challenged.
If a small team can use agents to do work that previously required five people clicking through five tools, the old per-seat model starts to feel strange. If an agent can summarize a pipeline, reconcile a spreadsheet, draft a customer update, and update the CRM, the customer is going to look harder at every expensive seat and every product that only makes sense when a human is trapped inside the UI.
This is where cost pressure shows up.
Companies are going to need to reduce friction to win customers. They may need cheaper entry points, better usage-based pricing, better automation support, and less “talk to sales before you can try the thing” energy.
That does not mean all SaaS should become cheap.
It means the value has to be clearer.
If the product saves serious time, owns important workflow, protects critical data, or becomes the trusted system of record, customers will pay.
But if the product is just a thin interface over work an agent can now perform, it is going to get squeezed.
The builder-first warning sign
Pieter Levels made this point concrete in a recent post on X. He said he had replaced almost all of his SaaS subscriptions with his own vibe-coded replacements. In a follow-up post, he listed some of the tools he had already rebuilt for himself: screenshots for social images, image resizing, content moderation, uptime monitoring with status pages, backups, and similar internal services.
That is the builder-first version of the pressure I am talking about.
Why pay for another service if you can build the exact version you need?
That is not always the right answer. There are still categories where buying is the responsible move: email deliverability, payments, business phones, infrastructure, security, compliance, and anything where reliability or trust is the product.
But for the $50/month or $100/month tools that mostly wrap a narrow workflow, the math is changing. A builder can now look at the invoice and ask a sharper question:
Is this a product I should keep buying, or a workflow I can now own?
That does not kill SaaS.
It raises the bar.
A product has to be harder to replace than a weekend project with an agent. It has to own distribution, data, trust, reliability, collaboration, compliance, network effects, or deep workflow knowledge. Otherwise the buyer may not churn to a competitor. They may churn to their own agent-built replacement.
The winners will expose the work
The SaaS companies that win will not treat AI agents like weird edge-case users.
They will assume agents are part of the workflow.
That means clean APIs. Real documentation. Stable schemas. Webhooks. Audit logs. Permission boundaries. Test environments. Clear object models. Good auth flows. Tool surfaces that Codex, Claude, internal company agents, and future AI systems can call safely.
It also means the product itself needs to be easier to reason about.
If the only way to understand the workflow is to click around a complicated UI and infer what the app is doing, that is a problem. Humans tolerate that because we are used to bad software. Agents will expose how messy that really is.
An agent-ready product should make the important actions obvious:
- Create this record.
- Update this status.
- Fetch this context.
- Summarize this activity.
- Approve this step.
- Escalate this exception.
- Generate this artifact.
That is not just an integration detail.
That is product design.
The companies that make their workflows legible to AI will become easier to adopt. The companies that keep everything locked behind a slow UI and brittle exports will force customers to route around them.
The losers will be closed boxes
I think the most fragile SaaS products are the ones that act like closed boxes.
They expect users to live inside the product. They make data hard to move. They have weak APIs. They rely on manual clicking. They charge for access instead of outcomes. They add complexity because the interface is the product.
That was already annoying.
AI makes it more obvious.
When a user has an agent that can work across tools, the closed box starts to look like a tax. It slows the work down. It hides context. It makes automation harder. It forces the user back into manual mode.
Some companies can get away with that because they own a system of record, have deep compliance requirements, or sit inside a large enterprise procurement structure.
But even there, pressure builds.
People will ask why the agent cannot use the tool directly. They will ask why the export is bad. They will ask why the API is limited. They will ask why the product is priced like every user needs to spend all day inside it.
The answer cannot just be “because that is our business model.”
Trust still matters
This is where the trust point comes back.
Agent-ready software is not only about giving AI access to everything.
That would be reckless.
The valuable products will be the ones that expose the right work with the right boundaries. They will make it clear what the agent can do, what it cannot do, what changed, who approved it, and how to recover when something goes wrong.
That is trust at the product layer.
Users will not trust a SaaS product just because it says it has AI features. They will trust it if the product gives them control, visibility, and confidence.
Can I see what the agent did?
Can I limit what it can change?
Can I approve high-risk steps?
Can I undo mistakes?
Can I bring my own agent instead of being forced into one narrow AI assistant?
Those questions are going to matter.
The winning products will not be the ones with the loudest AI badge. They will be the ones that let humans and agents work together without making the whole system feel unsafe.
This will create new opportunities too
The concerning version is real.
Some software companies will disappear. Some tools will get compressed into features. Some workflows will move from apps into agents. Some teams will realize they do not need as many products or as many people clicking through them.
I do not think it helps to pretend otherwise.
But there is an opportunity here too.
There will be room for companies that help businesses make this transition. There will be room for products that become the trusted layer between messy business data and AI agents. There will be room for vertical tools that understand a specific workflow deeply and expose it cleanly. There will be room for smaller teams that use AI to build better software with lower overhead.
The winners may not look like old SaaS companies.
Some will be software companies.
Some will be AI deployment companies.
Some will be service businesses with reusable software inside them.
Some will be tiny teams running products that used to require a much larger staff.
The common thread is that they will embrace the agent layer instead of pretending it is a temporary feature wave.
The point
I do not think SaaS is dead.
That is too simple.
But I do think a lot of SaaS is going to be repriced, rebuilt, or routed around.
The products that win will be the ones that earn trust, reduce friction, and become useful to both humans and agents.
If you are building software now, I think the default assumption should be:
My user is going to bring an AI agent to this workflow.
If that agent cannot use your product, your product may become invisible.
If that agent can use your product safely, your product may become more valuable than it was before.
That is the line I keep coming back to.
AI is not only changing how software gets built.
It is changing what software needs to be.
