The death of SaaS

With Claude Code, we hit an inflection point - it is now safer and safer to trust AI tools on longer and more complex tasks. The combination of model and planning loop has achieved in a lot of use cases a certain level of trustability - the time economics are now worth it. This is great both for individuals as well as businesses.

It like with a median filter - if most of the signal is noise, it won't do any good. But as soon as the median is more signal than noise, these can do wonders.

With the advent of "cheap but great code", many SaaS will be obsolete - it is now easier than ever to code your own in-house CRM app, your own photo-sharing service, your own blogging platform. Big, complex apps are still out of reach, but the gap will shrink with time

Consequences:

• Things that models CAN'T do will become valuable, the moats of SaaS
  • Communities
  • AI Models (because models depend on data as much as code)
  • Certifications/permits - think banking -
  • I don't think patents will help much, there's a lot of holes in the logic of patents when talking about software.
• Saas will have to become leaner, on the dev team anyways (for now)
  • As it becomes cheaper and cheaper to build something "good enough", SaaS will have to stay above the rest.
  • They're competing with the "ill build it myself" mentality

The "problem" is that whenever someone DARES to charge 10$ for a sock-sorting monthly subscription, someone will easily be able to come in and say "Claude code, make the sock-sorter" - their sock sorter will suck, but will work for 0$/month.

We've already finished the era of personal computing, it's going to be the era of personal, individualized apps.