I suspect that “Agentic replaces SaaS” will turn out to be wrong, at least for now.

There is a mainstream idea that newer coding tools like Claude and Codex will dramatically lower the cost of software development. Therefore, it will become much harder to make money selling software. Makes sense. Agentic coding is powerful.

However, it's a mistake to believe that SaaS customers are buying software.

Many SaaS customers are buying other types of business value, and software is simply the delivery mechanism.

Who is likely to be disrupted?

Some web businesses are ripe for replacement due to agentic coding. The software functionality these companies offer can be replicated in by a single developer with decent domain knowledge and AI tools. No kidding. These are fairly simple web platforms and they operate on data that a business itself generates or owns.

  • Customer support, ticketing/helpdesk, chatbots
  • CRM, customer management
  • Marketing automation
  • Analytics and tracking
  • Knowledge bases, document management
  • Source control
  • Monitoring and logging
  • Expense tracking
  • Website builders, CMS
  • Workflow automation
  • Project management

What these have in common is they follow the same formula: website + database

Who owns the data is actually very important! If a company has all the necessary data, it can now build these software products for itself, quite cheaply.

Some of these products might SEEM complex. For example, if you've ever configured "project management" software for an organization with multiple teams, you know it can be pretty gnarly.

But the REASON that it's complex is not some inherent quality. It's complex because the SaaS vendor requires deep configurability at every level to handle thousands of different customers with all kinds of weird requirements. Handling any one specific requirement is easy. Handling all those requirements simultaneously is what leads to complex software.

A customer ticketing system shouldn't need a million lines of code. But it does when that same code needs to support thousands of global companies, ranging in size from 10-10,000 employees, across hundreds of different industries, each with their own requirements.

Bespoke software avoids this problem. It can be simpler because it only needs to work for a single company.

With a competent developer, a decent ticketing system could be built in an afternoon, using agentic coding tools (or a week, at most). It would likely cost between $20-$100 in API tokens and run between 1,000-5,000 lines of code, depending on the specific feature set, and your definition of "lines of code". It would be specialized for one specific business, and would ignore every feature that particular business doesn't need. It's a one-time cost and eliminates monthly subscription fees.

Cheap, custom software sounds pretty disruptive.

Who won't get disrupted?

There is a more resilient category of SaaS business. Here, software is simply the delivery channel for the actual value these companies provide.

  • Accounting, taxes
  • Payroll, benefits management
  • Security, identity
  • Compliance tracking
  • E-signatures
  • SEO, ranking tools
  • Video conferencing
  • Advertising platforms
  • Job boards

What they're actually selling is trust, legal compliance, or network effects.

Trust comes from experience, reputation and perhaps branding. It's impossible to replace a business like Docusign with a prompt. To be sure, agentic coding could easily replicate the fundamental aspects of e-signature technology. But the code itself isn't what makes a company like Docusign valuable. There are lots of SaaS companies in the legal and compliance arena that provide something like "trust as a service". Coding faster doesn't affect that.

Regulated areas like hiring, taxes and payroll are also difficult to replace with custom software. It's not exactly that the software is difficult. But it needs to be exactly correct in handling local, regional and national regulations that change frequently. Many companies outsource these functions to a SaaS vendor, partly to save time, but largely to ensure legal compliance. There's little reason to bring big risks in-house by creating homemade software.

Network effects are another SaaS benefit that agentic coding tools cannot offer. Advertising platforms provide access to thousands (millions) of websites. Job boards draw applicants through popularity. Sites offering SEO keyword research work by providing access to propriety user data, gleaned from millions of people and purchased from ISPs.

You could argue whether these are truly SaaS, but they're certainly software-delivered and they rely on monthly or usage-based fees. What's being delivered is not "just software", but additional value in the form of trust, compliance or network effects.

Agentic coding cannot simply wizard those things into existence.

Is SaaS cooked?

I've just argued that there are two types of SaaS businesses, and that "just software" vendors may be in trouble. Agentic coding lowers the cost of development and lots of software can be brought back in-house to save on subscription fees.

But actually I don't think that's what's gonna happen.

Two reasons.

First, just because something is cheaper doesn't mean that it's a good idea. Agentic coding only provides the code. It doesn't offer 24/7 uptime. It doesn't offer resilient infrastructure or security guarantees. If there is a data breach, there's no one to sue. Businesses gain a lot from SaaS, and cost savings are not the only consideration. There's even value in being able to pick up the phone and yell at a vendor when something goes wrong! I'm only partly joking.

Second, look at open source SaaS. In nearly every software category, there are dozens of 100% free, open-source options available. Many of them are popular and quite good! But that hasn't put SaaS vendors out of business. On the contrary, many SaaS vendors offer a "community edition" of their software as a way to bring on new customers (or to create lock-in, ugh) My point is that "free" is nothing new here, and that SaaS companies have been competing against "free" for decades.

I'm arguing with myself, because I do think some reckoning will occur. But it will be mild compared to the devastation that many predict. There will be LOTS of new competitors because it's easier than ever to get started. Legacy vendors will struggle to adapt, and the categories of SaaS identified above as "just software" will struggle the most, but it won't be the end of the industry.