When and Why SaaS Still Matters in the Age of Personal AI

Status: Still a WIP but I think the ideas are still worth writing for now

We are now at the stage where AI can build a minimal working version of your SaaS product in an afternoon. Dashboards, forms, workflows, and integrations that took a startup previously years to ship are now a weekend prompt away. The era of personal software has arrived and a logical thought floating around is that people will increasingly build the tools they need instead of paying for them. So SaaS is finished, right?

I propose that is half right. AI will eat a large part of what software companies used to charge for. But it eats the parts that previously cost time and not the parts where the true value lived.

The useful question is no longer “can we build this software at all?” It is “which parts of our stack should we own as agents and internal tools, and which parts should we still rent as trusted, governed backbones?”

There is still a moat, you just have a raft now

AI can now clone the visible surface of most software. The UI components of list and detail views, standard workflows, lightweight CRUD schemas, etc. are all designed for non-technical humans to interact with software. AI has seen and learned from countless versions of the web UIs and can stand up something that looks and feels like a real product in days. Before this was the “moat” of creating software that was gated behind skilled developer labor.

This has created lots of discussion in the industry about the Software Development Lifecycle (SDLC). AI compresses the Build phase of the SDLC. Getting to a demo which clones the same UI and inspires ideas is now easy, but getting to a durable, governable, production-grade system is still hard. Many demos we see are from people saying “I made this in an afternoon, let’s ship it!”. In most cases they have codified a true problem they have and have created some form of software that solves that problem, but that is just the surface layer of the Product Lifecycle (PDLC).

To be clear, I don’t think that AI has removed any moat though, it still exists. Before we just had to spend lots of time and effort to swim across, where now AI has provided us a raft to get across with much less effort. So many of the vibe-coders get to the other side and ask “Now what?”. You can follow the SDLC and make sure that feature or idea you built is a little more sound, but if you are asking AI to review itself it will not ask the big questions. The hard substrate of software is the part you cannot screenshot. It includes rich entity graphs and messy real-world relationships, invariants that encode domain semantics, compliance and auditability, and the operational maturity to run all of it under load. None of that shows up in the UI that AI is imitating.

The question to ask is no longer whether AI can imitate your app, because it can, but instead ask “Why would customers still use your underlying system instead of building their own?”

Databases matter more

If AI can generate the workflow, the durable value has to move deeper than the workflow. It has to live in shared truth.

That sounds boring, but this is the part most people underestimate. The database is not valuable because it stores rows. It is valuable because the data is structured, labeled, searchable, permissioned, and stable enough that humans and agents can use it as context. A clone can copy the screen. It cannot magically inherit the years of customer records, history, relationships, permissions, audit logs, and edge cases that make the screen useful.

This is where SaaS can still charge. Not because customers could never export data into Postgres, but because most people do not want to own the full operating burden around that data. Identity, sync, backups, permissions, compliance, search, lineage, and recovery are not exciting features. They are the reason the product can be trusted as the backbone for other work.

AI makes that more important, not less. Agents are only useful when they have context they can trust. If your product owns clean context, exposes it clearly, and keeps the rules around it intact, then you are selling more than an app. You are selling the place where both people and agents can safely understand what is true.

Repeatable logic matters too

The second piece is logic. AI changes what a feature request means.

Before, if a customer wanted a weird workflow, they filed a ticket, waited for a roadmap, or built something awkward on the side. Now they can ask AI to write the script, build the agent, or wire together the one-off tool. That does not mean your SaaS is irrelevant. It means your SaaS needs to give that custom logic a better place to run than a random script on someone’s laptop.

This is the other business model that survives. Provide the official extension points. Let customers add lifecycle hooks, custom actions, small APIs, automations, and eventually personal agents that run near the data. Charge for the substrate around that: hosting, permissions, logs, retries, rate limits, secrets, audit trails, and the ability to turn something off when it goes wrong.

Email is a clean example. A user probably does not want to vibe-code an email client from scratch. They do not want to own Gmail auth, sync, threads, drafts, attachments, search, labels, spam behavior, and provider edge cases. They want the one custom workflow that makes email fit their life. Maybe that is a triage agent, a follow-up script, or a personal classifier. The SaaS product should own the boring backbone and provide the safe place for that custom logic to run.

That is the version of SaaS that still matters. AI may take over more of the visible product surface, and it will make building software cheaper. But the hard part moves to product judgment, durable data, governed extension points, and operating repeatable logic safely over time.

Closing thoughts

So I do not think SaaS is dead. I think the job description changes. The app becomes less of a fixed surface and more of a trusted substrate. The value is no longer “we built the interface you could not build.” The value is “we hold the truth, expose the right hooks, and keep the whole thing running when your agents start acting on it.”