Software finally eats services - Aaron Levie

The H-1B Visa Debate: Pricing vs. Lottery

The conversation kicks off with the contentious issue of H-1B visa reform, specifically the proposal to replace the lottery system with a pricing mechanism.

• The Argument for Pricing: A price-based system is presented as a logical market solution to allocate a scarce resource. Proponents argue the current lottery system is gamed by large consulting firms and tech giants, locking out startups. Setting a price—for instance, Keith Rabois’s suggestion of $20,000—could effectively squeeze out the “body shops” that saturate the market for lower-level IT administration jobs in the $80k-$120k salary range. This would target the arbitrage exploited by large consultancies and potentially reopen these jobs in the U.S. market.

• Goals and Counterarguments: A key question raised is, “What are we optimizing for?” Different goals—preventing wage depression, protecting American jobs, or attracting the world’s highest-merit individuals—lead to vastly different policy frameworks. A high price, such as the initially floated $100,000, might simply allow the wealthiest companies like Google and Amazon to capture all the talent, continuing to disadvantage startups. The real problem is the immense cost, complexity, and uncertainty of the current system, which disproportionately benefits large companies with dedicated resources to navigate it. A better approach might be a more nuanced system that ensures immigrants are net-positive to local wages and that startups can viably participate.

AI’s Real-World Impact on Developer Productivity

Contrasting with academic papers that question AI’s benefits, the speakers provide strong anecdotal evidence of massive productivity gains, particularly in software development.

• From the Trenches: At a company like Box, approximately 30% of new code is now generated by AI tools like Cursor. While self-reported developer productivity gains vary widely from 20% to 75%, the most significant variable isn’t seniority but a willingness to push the AI’s capabilities.

• The Startup Superpower: The most dramatic improvements are seen in small, senior-led startups, which report 3x, 5x, or even 10x productivity gains. This is driven by a fundamental shift in workflow. Instead of incremental, type-ahead assistance, these teams use AI agents to generate large blocks of code from detailed prompts. Their primary role evolves from code writing to code reviewing. This allows tiny, agile teams to achieve the output of a company with 50-100 engineers.

• Challenges in Measurement: Measuring these gains is difficult.

  • The Dazzle Effect: AI models are so impressive that users can be “dazzled” into feeling productive, even if the output isn’t precisely what they needed.
  • Shadow Productivity: The most significant gains come from bottom-up adoption of personal tools (e.g., ChatGPT, Cursor) by individual employees. Top-down corporate “AI pilot projects,” which are easier to measure, almost always fail because they can’t replicate this personal, secular movement.
  • Redefining Work: Productivity isn’t just about speed. AI is automating tedious tasks like writing documentation and tests. While this may not accelerate feature delivery, it results in higher-quality, more robust, and more maintainable code. The true impact is the velocity gained by compressing multi-day workflows into minutes.

The Platform Shift: Startups vs. Incumbents

The speakers argue that AI represents a true platform shift, creating the most favorable environment for startups in over a decade.

• A Reset of the Landscape: For much of the 2010s, many core software categories (e.g., chat, video conferencing) were mature, leaving fewer opportunities for new founders. AI completely resets this landscape, creating greenfield opportunities.

• Neutralizing Incumbent Advantages:

  • Scale: The traditional advantage of a large engineering workforce is neutralized when a small startup can achieve similar output using AI agents.
  • Distribution: While incumbents retain distribution advantages, the pervasive nature of AI (accessible on billions of phones) and the potential for viral, bottom-up adoption levels the playing field.

• The Innovator’s Dilemma: Incumbents are ill-equipped to handle the new, non-deterministic nature of AI and the rapid evolution of user behavior. Large companies struggle to deploy solutions that aren’t perfectly consistent, whereas startups can embrace this flexibility. Historically, during platform shifts, incumbents’ advantages are overestimated. While they don’t disappear, they rarely dominate the new paradigm.

AI Expanding the Market Beyond Software

A critical insight is that AI’s greatest impact may be in creating new markets by converting services into software, targeting a TAM that was previously inaccessible.

• New Competitors, New Customers: For the first time, software companies are competing not with other software companies, but with professional services firms, agencies, and manual workflows. A startup building AI for construction is competing with traditional construction companies, not an existing software incumbent.
• A Symbiotic Disruption: In a unique twist, the entities being “disrupted”—such as creative agencies or consulting firms—are often the primary customers for the new AI tools. They adopt the technology to make their own services faster and better, creating a dynamic where the disruptor sells to the disrupted. This opens the door for entirely new, AI-native service businesses (e.g., ad agencies, systems integrators) to be built from the ground up with an inherent structural advantage.

• Tags:
• Future of Work
• Startups
• AI
• Developer Productivity
• Platform Shift
• H-1B Visa
• Incumbents