Two sector-wide selloffs from a single company in three weeks. What happens to competitive strategy when building software costs almost nothing?
From Flo's AI Lab
35 sessions across 11 projects this week. The highlights: I deployed a second AI agent as an executive assistant on a dedicated Mac Mini (inbox triage, message routing, inter-agent coordination via shared memory). Built a GTM system from scratch, including CRM cleanup (899 ghost records deleted, 3 business pipelines rebuilt). Shipped an invoice automation skill that generates compliant invoices directly from the terminal. Prepared a community event website for production (CSS from 300 KB to 12 KB, GDPR-compliant font hosting, structured data). And closed three consulting deals.
The pattern across all of it: the bottleneck is never building. It is deciding what to build.
On Friday, Anthropic launched Claude Code Security, and cybersecurity stocks dropped within hours. CrowdStrike fell 8%. Okta lost 9.2%. The BUG cybersecurity ETF hit its lowest since November 2023 (Benzinga, Feb 20). Three weeks ago, Anthropic's Cowork plugins erased $285 billion from SaaS valuations (CNBC, Feb 6). Two sector-wide selloffs from a single company in three weeks. The pattern is clear. But the real business question is not which sector gets disrupted next. It is what happens to competitive strategy when building software costs almost nothing.
The Zero-Cost Building Trap
The tool that crashed cybersecurity stocks finds vulnerabilities the way a senior security researcher would, not through pattern-matching but through reasoning about how components interact. During testing, it found over 500 high-severity flaws in production open-source code, some dormant for decades (Fortune, Feb 20).
Claude Code alone now generates $2.5 billion in annualised revenue, more than doubling since January (Constellation Research, Feb 2026). At a $380 billion valuation after a $30 billion Series G (CNBC, Feb 12), Anthropic does not need to build specialised security tools, or legal tools, or marketing tools. The general-purpose model simply makes them redundant.
That is the difference between this wave and every previous software disruption: the disruptor is not a competing product. It is a capability layer that renders entire product categories optional. The era of vibe coding as a novelty is over.
Garry Tan, CEO of Y Combinator, puts numbers on this shift. YC companies with 10 to 20 people are reaching $10 to $20 million in annual revenue in 10 to 20 months (Mixergy, Feb 2026). Salient, a YC company, hit eight-figure revenue with six employees. Boris Cherny and members of Anthropic's technical staff have confirmed on multiple podcasts that 95% of production code is now written by Claude Code. Each engineer performs at the capacity of 20. Tan calls this the "20X company." But the two things AI cannot replace are agency (identifying the right problem) and taste (the judgement that separates a product people love from one they tolerate) (Vanta, Feb 2026).
Peter Steinberger proved this. One developer built OpenClaw, a proactive AI agent with persistent memory: 175,000 GitHub stars in two weeks (CNBC, Feb 2). OpenAI hired him, with Altman calling it "core to our product offerings" (TechCrunch, Feb 15). The code is open source, anyone can fork it. What made it win was vision and taste.
What to do Monday morning
If you run a company: Look at the tools your teams use every day. Excel models for financial planning. Access databases for internal tracking. Word templates for proposals. Publisher for marketing materials. Each of these is a process that an AI agent can now replace with a purpose-built internal tool in days, not months.
The make-or-buy equation has flipped. Building a custom solution tailored to your exact workflow is becoming cheaper than licensing generic software and paying consultants to configure it. The same logic applies upward: capabilities you currently outsource (compliance checks, market analysis, first-line customer support) can be brought in-house with AI agents at a fraction of the cost. Vertical integration is no longer a luxury reserved for corporations with large IT departments. A team of five can now do what a department of twenty did last year.
Start with one process. Pick the one where your team spends the most time fighting the tool instead of doing the work. Build a replacement this quarter.
If you manage a portfolio: The Anthropic pattern (SaaS crash, then cyber crash) will repeat. But the defensive play is not just scoring moats. It is actively integrating agentic engineering into your portfolio companies now, so they capture the value instead of losing it. The companies that use AI agents to vertically integrate will pull ahead. The ones that wait for their SaaS vendors to "add AI features" will fall behind. Push your management teams to run a proof of concept this quarter, not next year.
If you haven't started: A single person built what OpenAI considers core to its future. A six-person team reached eight-figure revenue. The barrier to building is gone. Vibe coding gave everyone a prototype. Agentic engineering gives everyone a product. The question is whether you define your strategy before your competitors build their version of your product this quarter.
What I'm reading
Berlin VC Revent sent a warning letter to its portfolio companies: the old playbook of tripling revenue is no longer enough. AI tools make software faster to copy and cheaper to build. Revent now demands 4x to 10x growth to outrun competitors who can replicate your product in weeks (Handelsblatt, Feb 2026).
If this changed how you think about one vendor contract or one portfolio position, forward it to a colleague facing the same question.
The analysis above identifies the shift. Making it work for your specific business is a different challenge. If you want to explore how agentic engineering applies to your competitive position, let's talk.