Listen to this article
The market looked at AI agents, looked at per-seat SaaS pricing, and did the math. The businesses that adapt will own the next decade, while the rest keep paying rent on legacy software.
”Black Tuesday for Software”
In five trading days, $800 billion evaporated. 1 Over the next twelve months, the sector lost $2 trillion. It was the worst non-recessionary drop the software sector has seen in thirty years.
The media called it the SaaSpocalypse, and analysts compared it to the dot-com bubble. 2 They were right.
The market was asking one specific question: if an AI agent can do the work of the people paying for SaaS seats, what happens to companies that monetize headcount?
| What happened | The number |
|---|---|
| Software value erased in 5 trading days | $800 billion Bloomberg |
| Total software sector losses (12 months) | ~$2 trillion Fortune |
| S&P 500 Software Index drop (5 days) | -13% Bloomberg |
| Software underperformance vs. S&P 500 | -24 pts Reuters |
| Atlassian single-week drop | -35% MarketMinute |
| Software price-to-sales compression | 9x → 6x Bain & Company |
| Forward earnings multiples collapse | 39x → 21x MarketMinute |
| US tech company loans in distressed territory | $46.9B Bloomberg |
An AI agent that does the work of the people paying for SaaS seats
Anthropic shipped Claude Cowork 3 — an agent that navigates enterprise apps autonomously instead of just acting as a copilot. It manages inboxes, reviews contracts, and executes workflows across Excel and Salesforce. It operates as a worker.
Wall Street reacted to the immediate implication: seat compression. 4 If an agent handles work that used to require three employees, you need fewer licenses. The entire SaaS business model relies on per-seat recurring revenue, assuming human headcount drives software consumption. That assumption is dead.
“As AI automates work previously requiring multiple employees, vendors are moving away from per-user charges toward models based on tokens consumed, workflows executed, or transactions processed.” 5
— PYMNTS
The old SaaS playbook of adding employees, buying more seats, and expanding the contract has stalled out. Bain pegs net revenue retention at a plateaued 90% 6 — the average SaaS customer is spending less on renewal. AI agents don’t need seats. When one agent handles tasks across your CRM and project management tools, vendors lose recurring revenue. They’re already frantically pivoting toward models based on “tokens consumed” or “workflows executed.” The seat-based model is on its way out.
It’s not just Wall Street. Companies are already making the switch.
Companies are already gutting their stacks. You can see it in the data — Retool found 35% of enterprises replaced at least one SaaS tool this year 7 — and you can feel it in the market.
of enterprises have replaced at least one SaaS tool
plan to build more internal tools
have shipped production software using AI
created software outside formal IT oversight
What’s being replaced
Enterprises are targeting the worst-performing pieces of their stacks. Workflow tools lead at 35%, followed by internal admin at 33% and BI dashboards at 29%. Support, PM, and CRMs are also targeted.
The money is shifting. Startups are building AI-native platforms aimed directly at incumbents. Monaco raised $35M to go after Salesforce, while Revian replaces 21 SaaS tools with one platform. AI has effectively destroyed the moat that protected legacy enterprise software for the last decade.
SaaStr’s updated 90/10 framework nails the new logic: buy 90% off the shelf when adequate solutions exist, but build the 10% where existing tools lack AI. 8 Their own team replaced a paid portal tool in a single day using Claude and vibe coding. Their deciding factor is ruthless: they build a replacement the second a paid SaaS product lacks AI functionality.
IT budgets are growing. The money is moving.
Global IT spending will hit $6.2 trillion this year, up 10.8%. 9 But the destination for that spending has completely changed.
$2.52 trillion in global AI spending — up 44% in a single year. Faster adoption than cloud, mobile, or SaaS itself. Hyperscalers like Amazon, Google, Microsoft, and Meta are each writing checks north of $100 billion for AI data centers alone, 10 with combined capex of $470–690 billion. Companies now spend 3–4x more on AI-optimized servers than traditional ones.
And here’s the illusion: enterprise software’s 15.2% growth looks healthy, but roughly 9 points are just vendor price hikes — not new capabilities. The real new spending? About 6%. And most of that is going to AI.
The checks are being written. They’re just going to AI infrastructure and AI-native tools instead of SaaS renewals. CIOs are funding what comes next.
Your vendor is selling a bolt-on. It will fail.
Your vendor is going to push an AI feature at your next renewal. Salesforce will push Einstein. HubSpot will push Breeze. 12 They are scrambling to staple AI onto architectures designed a decade ago. Don’t fall for it.
The architecture won’t support it. You can see the failure rate in the data: 11
“Over 40% of agentic AI projects are expected to fail by 2027 because traditional enterprise systems lack the real-time capabilities, modern APIs, and modular architectures needed for true agent integration.”
“Many organizations attempt to simply automate existing human-centric processes rather than reimagining workflows for agent-native environments.”
Bolt-on AI
What your current vendor is doing
- × AI features added on top of legacy architecture
- × Data locked in proprietary silos that agents can't access natively
- × UI designed for humans that agents must work around
- × Per-seat pricing that doesn't account for agent workflows
- × Limited by what the platform's architecture allows
AI-native
What's replacing it
- ✓ AI built into the foundation of the architecture
- ✓ Open data layers that agents read and write natively
- ✓ APIs and interfaces designed for both human and agent interaction
- ✓ Pricing based on value delivered, not headcount
- ✓ The AI shapes the system, not the other way around
Architecture determines the ceiling. Bolt-on AI provides chatbots and copilots to incrementally improve existing workflows. AI-native architecture allows you to rethink workflows from the ground up.
Companies are gutting their stacks and pulling their data out
Companies aren’t just switching tools. They are ripping their data out of vendor clouds. 13 Call it repatriation or call it sovereignty. It means one thing: you need your data on infrastructure you actually control.
in sovereign-cloud infrastructure spending, growing 35% YoY
Gartner
projected sovereign cloud market by 2028
Vultr
of existing cloud workloads could shift to local/sovereign providers
Gartner
Ownership is becoming a competitive advantage. Nearly half of the $235 billion software loan market is rated B- or lower. 15 Cash flows are drying up, risking defaults. Companies that control their data, infrastructure, and AI stack can move faster and operate more efficiently than those renting from vendors.
Your competitors are already deploying agents
Agentic AI went from an experiment to an enterprise mandate. 14 These systems don’t just answer questions. They execute tasks, make decisions, and coordinate workflows autonomously.
report regular AI use in at least one business function
McKinsey
in production with or piloting agentic AI
Mayfield
at least experimenting with AI agents
McKinsey
mix internal builds with vendor solutions
Mayfield
The deployment gap
There’s a massive gap between experimenting with AI and actually putting it into production. Every enterprise is running a pilot right now, but look at the drop-off when it’s time to deploy: 88% are using AI, 72% are piloting, but only 14% have deployable systems and just 11% are in production. 58% cite data readiness as the top blocker. 60% report no formal AI governance.
The companies closing this gap succeed because they have the right architecture. They use systems designed natively for AI, rather than legacy platforms with added AI features.
Five shifts you can’t ignore
01 — Stop auto-renewing your stack. A third of enterprises replaced a core SaaS tool this year. AI crashed the cost of custom software. If your vendor isn’t delivering exponentially more value than they did last year, you are overpaying.
02 — Reject the bolt-on. Stapling an AI feature onto legacy architecture fails in production. A chatbot widget on a ten-year-old platform isn’t transformation. It’s a retention tactic.
03 — Per-seat pricing is dead. One agent does the work of three humans. Per-seat pricing can’t survive that math. If you’re still paying by the seat, you’re subsidizing a dying business model.
04 — Own your data. Sovereign AI spending is surging. The companies that own their infrastructure move faster and spend less. Renting your core stack is a liability.
05 — Find the weakest link. 16 Not every tool needs replacing today, but there’s almost always one that costs too much and delivers too little. For most companies, that’s the CMS — the easiest starting point with fast ROI. Start there and build momentum.