Where AI Will Win—and Where It Won't
February 12, 2026
The "SaaSpocalypse" is
here. Software stocks are down 25% year-to-date. Traders are in
"get me out" mode. The narrative: "AI will kill
SaaS".
Here's my take on what AI can and
can't do—and what it means for software companies (and You
as investor).
AI Is a Mirror (Mostly)
Modern LLMs are trained largely on internet data,
plus synthetic data and human feedback. What goes in shapes what
comes out. AI reflects humanity's collective knowledge,
patterns, and limitations.
This means AI excels at
problems humanity has already solved and documented. Coding?
Decades of open-source repositories. Legal boilerplate? Millions
of contracts online. Customer support scripts? Endless training
data.
Software development *costs* are genuinely
plummeting. That part of the narrative is real.
Where AI Struggles
AI struggles with two things:
1) novel
reasoning beyond its training distribution, and
2) problems
requiring real-world causal intervention.
Consider
macroeconomics. A policy change takes years to play out.
Millions of variables interact.
But the deeper issue
isn't just slow feedback—it's that you can't run
experiments. You can't try a policy, observe the outcome,
and retrain. AI can pattern-match on historical data, but it
can't reliably predict the effects of interventions never
tried.
The hardest problems for AI are those
requiring counterfactual reasoning about novel situations in
complex systems. If you can't experiment, you can't
generate reliable training signal.
The Network Effect Moat
Here's what the SaaSpocalypse narrative
misses: network effects.
Facebook isn't valuable
because its software is hard to build. It's valuable
because everyone you know is already on it. Salesforce
isn't sticky because CRM is technically complex. It's
sticky because your entire sales organization's workflows,
integrations, and data live there.
AI can lower the cost of
building software. It can't easily replicate an installed
base of millions of users who would have to coordinate a
simultaneous switch.
Companies with strong network
effects can use AI to reduce operating costs while retaining
users. They become more profitable, not obsolete.
My Prediction
SaaS companies without network effects—point
solutions, undifferentiated tools, anything easily
replicated—are genuinely at risk. AI makes building commodity
software trivially cheap.
SaaS companies with network
effects—platforms where value increases with each additional
user—will likely survive and flourish. They'll use AI to
cut costs and improve products while competitors struggle to
dislodge entrenched user bases.
The sorting mechanism
isn't "SaaS vs. AI." It's "network
effects vs. no network effects."
Why This Matters for Remake.ai
We're building a software ecosystem for
consumer robots - apps, 3rd-party apps, apps marketplace,
platform, SDK for app developers, SDK for robot developers.
Today,
robot vacuum owners have zero third-party apps. Any ecosystem is
infinitely more than the status quo.
Similar to
Google Android and Apple iOS app store, once customers purchase
Remake hardware, apps and once 3rd party developers invest into
learning Remake SDKs and develop apps for sale - switching to
another platform is not easy.
Network effects kick in
- Remake having more users attracts more Remake developers, who
make more apps and Remake-compatible hardware. Having more
Remake-compatible apps and hardware attracts more Remake users,
making it a self-reinforcing cycle.
The real moat is
ecosystem + hardware + number of users. If we get there first
and developers build for us, competitors have to convince users
to replace physical hardware to switch.
I could be
wrong. Predicting the future is hard—especially when feedback
loops are long and variables are many. But if I had to bet on
which software companies survive the next decade, I'd bet
on the ones where users are locked in by each other, by their
data, or by their hardware—not just by code.