From Flo's AI Lab
42 sessions across 15 projects this week. I built a virtual company on a $700 Mac Mini: a multi-agent system where each AI agent has its own identity, permissions, and responsibilities, handling tasks from executive coordination to code review. I shipped a complete go-to-market system (CRM integration with Attio, AI-powered visibility audits as a sales tool, lead scoring that filtered 2,445 contacts down to 167 worth pursuing) and closed three deals. I started a CFO agent that processes invoices and reconciles them against accounting software, validated by three specialised review agents (architecture, security, performance) before a single line of production code was written. And at our second Claude Code Munich meetup, hosted at Accenture Song, speakers converged on a conservative estimate: one developer with agentic engineering tools produces output worth $4.5 million per year.
One person. The operational footprint of a small firm.
Two days after that meetup, Jack Dorsey made the same bet at corporate scale. Block cut 4,000 employees, nearly half its workforce, and the stock jumped 24%. The operating model I am building as a solo practitioner (AI agents handling sales, finance, engineering review, coordination) is the same model Dorsey just imposed on a public company. Not as an experiment. As the new default. If he is right that most companies will follow within a year, every CEO, fund manager, and displaced worker needs to rethink what a "company" even is.
The Firm Is Shrinking. Coase Predicted This.
Firms exist because transaction costs make internal coordination cheaper than hiring externally. That is Coase's Nobel Prize-winning insight from 1937, extended by Williamson into a theory of vertical integration: firms grow when coordination is cheap inside, and shrink when it is not. AI is collapsing these costs faster than either economist imagined.
Block reported $10.36 billion in gross profit for 2025, up 17% year over year (CNBC, Feb 26). Engineers using Block's internal AI tool Goose ship 40% more code per person than six months ago (Fortune, Feb 27). Dorsey's conclusion: "Intelligence tools have changed what it means to build and run a company. A significantly smaller team can do more and do it better" (TechCrunch, Feb 26). This is not vibe coding, one developer hacking a prototype. This is a public company restructuring its entire operating model around AI agents.
The market agreed. Block's stock surged 24% on the news. Investors are not rewarding a cost cut. They are pricing in a permanent structural shift: fewer people, same or higher output.
Dorsey predicts most companies will reach the same conclusion within a year. "I'd rather get there honestly and on our own terms than be forced into it reactively" (CNN, Feb 26). The U.S. recorded 108,435 announced layoffs in January 2026 alone, up 118% from a year prior (Fortune, Feb 27).
This is Coase in real time. When AI agents handle sales, compliance, customer support, and engineering, the coordination costs that justified large teams evaporate. The boundary of the firm contracts. Not because the work disappears, but because the work no longer requires the firm. The depth of value creation per person increases. The number of people required to capture that value drops.
Where Do 4,000 People Go?
Schumpeter meets Coase. Creative destruction does not just eliminate. It redirects.
The data shows the redirecting has already started. 33% of U.S. adults now plan to start a business or side hustle, a 94% increase year over year. 67% of new ventures in 2026 were launched after a layoff. Over 80% of U.S. small businesses have zero employees (QuickBooks, 2026). A complete AI-enabled solopreneur stack (marketing, invoicing, support, product development) costs $3,000 to $12,000 per year.
The person Block let go on Wednesday has, by Monday, access to tools that handle marketing, compliance, and product development. Whether they build a viable business depends on judgement, domain knowledge, and the ability to sell. The tools alone are not sufficient. But for the first time, they are not the bottleneck either. Vibe coding gave everyone a prototype. Claude Code and tools like it give everyone a company.
But the institutions are not ready. In Germany, forming a GmbH still requires a notary, a minimum capital of EUR 25,000, and weeks of bureaucratic processing. Employment agencies are built to place people in existing jobs, not to enable them to start businesses. Tax offices require paper forms. Start-up funding programmes assume you need employees. The economic unit is shrinking faster than the regulatory infrastructure can adapt.
What This Means for Capital
If the optimal company gets smaller, venture capital and private equity models need recalibration. Garry Tan of Y Combinator observes that companies with 10 to 20 people reach $10 to $20 million in annual revenue in under two years. Salient, a YC company, hit eight-figure revenue with six employees (Vanta, Feb 2026).
If six people can build what previously required sixty, revenue per employee soars, but total company revenue may plateau lower. Exit multiples compress: a $20 million company with 10 people has better margins but a lower ceiling than the $200 million company PE firms traditionally target. Early-stage funds may need more bets at smaller cheques. Business angels may discover that the companies needing capital least are the ones worth backing most.
The counterargument: switching costs protect incumbents. Microsoft, Salesforce, SAP. But switching costs have never been absolute. Users migrate when the value gap is large enough. There would be no new social networks, no new payment platforms, and no new CRM categories if lock-in were truly insurmountable. Block itself was built on that exact premise.
The deeper question is whether switching even matters when agentic engineering lets you bypass tools entirely. When you use Claude Code to build a financial model, generate a board deck, or analyse your sales pipeline, you are not switching away from Excel or PowerPoint. You are making them irrelevant. The outcome is what you were paying for all along. The tool was just the delivery mechanism.
For CEOs: Dorsey did not wait for AI to become perfect. He restructured now, while the company is profitable and the transition is voluntary. Every quarter you delay, an AI-first competitor with one-tenth your headcount is building a version of your product. Audit every team by asking one question: if an AI agent could do 40% of this team's work (Block's number, not a projection), what would the team look like? Then build that team.
What I'm Reading
Deloitte's State of AI in the Enterprise 2026 surveyed 3,235 business leaders across 24 countries. The headline: only 8.6% of companies have AI agents in production. 74% expect to within two years. Block sits in the 8.6%. Most of your competitors do not. The biggest barrier the report found is not technology but insufficient worker skills (Deloitte, 2026), which is exactly the gap Dorsey chose to solve by cutting rather than retraining.
If this changed how you think about your headcount, one portfolio position, or your next career move, forward it to someone asking the same question.
The shift from large organisations to AI-enabled smaller teams is not a prediction. It happened this week. If you want to explore what agentic engineering means for your company, let's talk.