The “SaaS-pocalypse” narrative is right. It’s just pointing at the wrong companies.
The standard argument — from a16z to Forrester — is that coding agents can replicate any feature on demand, so all application software eventually gets commoditized. I think that’s right in direction and wrong in scope.
AI isn’t killing SaaS. It’s killing replaceable SaaS. And most of that category never should have existed as a business.
The Utility SaaS Obituary
Extract tables from a PDF? Reformat data, clean CSVs, convert files? There were thousands of tools for all of it. They worked because building your own solution was too slow or too expensive. That’s gone.
Last month I wanted a quick read on Char’s run rate — dropped a bank CSV into a coding agent and had a full financial breakdown in minutes. Before, that’s a product category. Now it’s a throwaway script.
If your product exists solely to solve a narrow, one-off task, it’s not a business anymore. It’s a prompt.
AI Doesn’t Kill Products. It Kills Features.
AI doesn’t commoditize software — it commoditizes features.
A product that’s just a wrapper around a single capability can now be replicated by anyone with a laptop and a good prompt. There’s no moat in a feature. There never was — AI just made that obvious.
The average organization cut its SaaS app count by 18% between 2022 and 2024, according to BetterCloud. The cuts are coming from single-function tools that never grew beyond one capability.
If your product can’t evolve or embed itself into a broader workflow, it’s already dead. AI is just the autopsy.
What Actually Survives
The ones that make it share one trait: they own workflows, not features.
-
Deep workflow integration — Slack and Salesforce don’t win on any single capability. They win because replacing them means ripping out entire systems and retraining teams. The switching cost is the moat. AI can’t dissolve that.
-
Continuous systems of action — Software that requires ongoing coordination compounds in value over time. Notion and Monday.com aren’t static tools — they’re living systems that embedded AI into their core, not bolted it on as a feature. Monday.com hit 112% net revenue retention in Q1 2025 under the same AI pressure that’s killing point solutions. The more you use them, the harder they are to leave.
-
Distribution at scale — Incumbents can ship AI features to millions of users in a single product cycle. HubSpot went from four AI agents to over twenty in less than a year, deployed across their entire customer base. The technical gap is closing — the distribution gap is not.
More Products. More Casualties.
AI will produce more SaaS companies, and destroy more of them faster.
Building has never been cheaper. Distribution, conviction, and taste haven’t. Software startup funding jumped 29% year-over-year to $125 billion in 2024 — more products than ever, competing for the same shrinking pool of workflows AI hasn’t absorbed yet.
Most will fail not because they’re bad, but because they were built before the founders understood the problem.
Why Startups Still Win Anyway
Incumbents have distribution. They also have a structural problem: they can’t move against themselves.
They’re slower to cannibalize existing revenue, hesitant to collapse pricing or kill features that currently generate income. Every dollar of existing ARR is an anchor.
Startups can move into the gaps incumbents are structurally unable to address — and those gaps are widening faster than incumbents can patch them. That’s still the opening. It hasn’t closed.
How I Build Differently Now
Code quality is no longer the bottleneck. Speed is. And speed without feedback is just expensive guessing.
Every piece of software I’ve shipped breaks in the real world — not because the logic was wrong, but because the environment is messier than any spec anticipates. We built auto start/stop recording into Char — genuinely useful, one of those features users love until it breaks on them. Jabra headphones trigger it unexpectedly. Stopping to share your screen in Google Meet triggers it. Switching your mic to voice isolation mode in macOS triggers it. Each one is its own rabbit hole that you only find when a real user hits it in a real meeting.
Because code generation is cheap, the temptation is to over-engineer before launching. Don’t. Ship the smallest version that solves the core problem. Track where it breaks, not just usage metrics. Make one targeted change, then re-observe. Rebuilding from scratch resets your learning. Patching and watching compounds it.
SpaceX doesn’t simulate rocket launches forever. They launch, watch what breaks, and fix it. The only real advantage left is learning faster than everyone else.
Good Software Always Prevails
If the SaaS model feels shaky for your business right now, you’re asking the wrong question.
The business model for software has always changed — from mainframe time-sharing to packaged software to subscriptions. The model is just the container. SaaS isn’t sacred. Something will replace it.
But the question underneath has never changed: will your software keep getting better?
Static software dies — not because of AI, not because of pricing pressure, but because software that stops evolving stops earning its place.
So if you’re worried about whether SaaS is the right model, flip the question. Ask whether your product will ever stop needing to improve. If the answer is yes — if there’s a version you could ship and walk away from — you’re building a tool, not a business, and the model doesn’t matter anyway.
If the answer is no, you’re fine. The model will find you.
Good software always prevails. That part hasn’t changed.
March 24, 2026