Last week, Anthropic launched Claude Code Security, and cybersecurity stocks such as CrowdStrike, Palo Alto Networks, Check Point, and others fell. Shares of traditional software companies also dropped sharply this week. IBM fell 13% in a single day after Anthropic said its AI tools can help with modernizing COBOL. A viral Substack post outlined a hypothetical scenario where AI rewrites the economics of knowledge work, and food delivery, payments, and Indian IT services stocks consequently fell.
Investors are clearly worried. But where others panic, I see opportunity.
New developments with the Claude platform have major implications:
The rise of AI as digital labor. We are moving from general-purpose chat to function-specific digital workers. Claude’s new “plug-ins” or customizable agents cover security, sales, marketing, finance, legal, and more. This is a structural shift from AI as productivity software to AI as digital labor.
Massive pressure on SaaS companies. If an agent can read your CRM, draft outreach, update records, generate reports, and coordinate across tools – many traditional products risk becoming databases with thin UI layers. The defensibility shifts from interface to proprietary data, deep workflow integration, and distribution. The market is pricing in that risk. Whether it’s overreacting is a separate question.
Verticalization accelerates, but not necessarily for the platform players. Generic models are powerful, but winning in verticals requires logic, auditability, industry nuance, enterprise sales, and proprietary data moats. Horizontal platforms like Anthropic may not be best positioned to capture that. What they will do is keep making their underlying models dramatically better, and those models will power the domain-specific, AI-native systems that actually win in regulated, complex, high-stakes industries. The foundation gets stronger. The opportunity for specialized builders grows with it.
Enterprise buying behavior changes. Functional leaders – CISOs, CFOs, CROs, VP Engineering – will increasingly buy AI agents tied to measurable outcomes. The conversation moves from “innovation budget” to “operating leverage.” In our 2026 Mayfield CXO report on enterprise AI, 46% of AI buying decisions are made by business leaders.
A new infrastructure layer becomes critical. When enterprises deploy dozens of Cowork agents, they’ll need governance, observability, permissions, cost controls, and audit trails. The control plane for AI labor is still early – and likely to be a significant category.
The advice I’m giving founders right now
Claude for Everything is both a threat and a gift. A threat if you’re building something shallow. A gift if you’re building something with a deep MOAT.
Here’s the advice I’m giving founders right now:
Build systems of work, not systems of record. The companies that endure will move from storing what happened to deciding and acting on what happens next.
Own a vertical, deeply. Depth beats breadth when the platform layer keeps getting stronger. Go where compliance, nuance, and trust create real moats.
Accumulate proprietary data loops fast. If your product doesn’t get smarter with every interaction, it will be replaceable. Generic data commoditizes. Learning loops compound.
Own painful, multi-step workflows, distribution, and the customer relationship. Direct distribution and mission-critical workflow ownership are the most durable competitive advantages in this era. Be careful of developer workflows such as code verification, code security, and DevOps. Claude Code and other coding agents will continue to expand there.
Price to outcomes. Seat-based pricing made sense when software was passive. As AI makes software active, your pricing model will reveal whether AI is core or cosmetic. Offer outcome-based pricing models tied to headcount reduction or measurable cost savings.
Design trust and safety from the start, with human-in-the-loop controls. And don’t build on a single model provider.
The market selloff is a signal, not a verdict
The market selloff this week is an overreaction to disruption anxiety. That anxiety is legitimate. But the winners in this next era won’t be determined by which incumbents survive the news cycle — they’ll be the companies, new and old, that figure out how to own execution inside the workflows that matter most.
The next decade won’t be defined by general chat interfaces. It will be defined by AI embedded invisibly inside of how work actually gets done by domain and vertical-specific agents.
The winners won’t be the companies with the smartest prompts. They’ll be the ones who own outcomes and price their products accordingly.
