The Post-Software AI world
When software becomes cheap, what's left to defend?
When software becomes cheap, what’s left to defend?
Every team is restructuring around AI. But what happens when AI is standard, not special? The real question is this: what survives the SaaSpocalypse?
This article is for founders, operators, and technical leaders deciding where to place their next dollar. The goal is simple: help you spot what stays defensible when software creation gets cheap.
Core Assumptions
We’re betting on four things:
AI becomes nearly autonomous, requiring only high-level human direction (”build me a dashboard for X”)
Orchestration systems emerge to monitor and manage AI output continuously
Knowledge becomes on-demand, AI provides reliable expertise across domains, instantly
Software dominates. We leave robots and physical automation out of this thought experiment.
The Hiring Model: Then and Now
The old playbook was simple. Hire specialists as needs emerged. Need marketing? Contract a consultant. Hit a growth threshold? Build a department. Turn external expertise into internal headcount.
This worked. It was the only model that worked.
SaaS Arrives
SaaS flipped the script. Instead of building internal teams, companies bought solutions off-the-shelf. Instant expertise. Instant deployment. Zero HR overhead.
The trade-off? You paid per seat. Per-seat pricing gets expensive fast. ChatOps tools can run into tens of thousands per month. Enterprise agreements compound the cost. Vendor lock-in limits your options. Small teams often skip high-value tools because org-wide seat math kills the business case.
SaaS solved speed. It created new problems at scale.
The Self-Hosting Rebellion
Cloud costs ballooned. Sovereignty mattered. Performance demands grew. Companies started looking inward again, self-hosting infrastructure (Kubernetes, Postgres, Kafka) that SaaS providers had locked them into.
The catch? You needed time and skilled engineers. Lots of both. You could avoid vendor lock-in and per-seat fees. But you also burned months hiring, onboarding, and operating the stack. Opportunity cost often erased the savings.
This became the constraint: capable engineers + time.
AI Collapses the Constraint
Both constraints just disappeared.
An engineer plus AI can scaffold in days what used to take weeks. Domain expertise is now on demand. Kubernetes architecture. Database optimization. Infrastructure automation. You can access all of it faster.
This flips the OPEX and CAPEX trade-off. SaaS won because it reduced upfront CAPEX. Now AI amplifies internal engineering output, so self-hosting is far more practical for more teams.
Small teams can now build what once required entire departments. Mid-market and enterprise companies can redirect renewal spend into AI-assisted internal tools. Lower TCO. More technical control in-house.
The middle ground is here: self-hosting economics with enterprise-grade knowledge, zero per-seat fees, zero vendor lock-in.
What Can’t Be Replicated? Moats in the AI Era
Moat is the 2025-2026 buzzword. It means defensibility. Why AI cannot replicate you and take your customers. There are only a few durable categories:
1. Deep Knowledge
TSMC. ASML. EUV lithography expertise that no LLM can summon. Niche, hard-won knowledge embedded in people, processes, and data. AI can’t teleport you there.
2. Network Effects & Brand
Even if you could instant-clone SAP tomorrow, they would still own the market. Decades of trust. Millions of users locked in. A brand that carries weight. Network effects competitors cannot fast-forward through.
3. Human Touch
Game development. Streaming personalities. Comedy. Content creation. AI can help design, optimize, produce, but the source is still human. That authenticity doesn’t copy.
4. Regulatory Burden
You cannot AI-into-existence a payments platform without PCI DSS compliance. You cannot build fintech overnight without navigating data protection, licensing, and audits. Compliance does not disappear at scale. It becomes a moat.
5. Physicality
Some work lives in the real world. Manufacturing. Construction. Last-mile logistics. Field service. Surgery. These require physical presence, specialized assets, and fast judgment under uncertainty. Robots exist, but they are still brittle in dynamic settings. A surgeon can improvise when anatomy changes. A construction crew can adapt when soil shifts. A logistics network can reroute when roads flood. AI still struggles with emergence. When work is novel, unpredictable, or deeply human, physical presence stays defensible.
What Actually Matters Now
Execution is money + AI competency. That’s it.
And this plays hard into the hands of companies that already have moats. Large fintech firms can absorb massive token costs. They have in-house engineering talent. They have capital reserves. They can accelerate.
Watch TSMC. Watch NVIDIA. Watch RAM manufacturers. They’ve become the infrastructure backbone, the real chokepoint. Foundational model providers will likely follow the same path: whoever controls the compute layer controls the game.
AI is a double-edged sword. Scrappy startups can move fast and break out. Large companies get the same tools plus existing moats plus deep reserves. They convert capital into execution at scale.
The race right now is artificial and expensive: token maxing. Companies burning capital to force-fit their way to AI maturity. It works, but it’s wasteful.
The smarter play: Get to a position of strength fast, within your constraints. Then pivot. Stop racing. Focus on what AI can’t do. Fill the gaps only you can fill. Generate value where it actually matters.
Your moats matter. But speed matters more. And speed + moats? That’s unbeatable.
If you want to pressure-test your AI strategy and moat, contact us.