Software finally eats services – Aaron Levie

Imagine a seasoned CEO casually revealing that 30% of his company’s code is now generated by AI, and that the most effective engineering teams have fundamentally stopped writing code to become full-time reviewers. This shift from creation to curation captures the staggering velocity of change as artificial intelligence rewires the very nature of work. In a wide-ranging discussion, Box CEO Aaron Levie joins a16z’s Steven Sinofsky and Martin Casado to dissect this transformation, while also diving into a heated policy debate around reforming the H-1B visa system to better serve startups and the tech ecosystem.

The conversation first tackles the contentious proposal to replace the current H-1B lottery with a price-based system, where companies would pay a significant fee (figures like $20k to $100k were debated) to sponsor a visa. Proponents argue this would dismantle the “gaming” of the system by large consultancies and body shops that currently dominate the lottery, freeing up visas for startups and high-merit talent while protecting mid-level IT jobs in the U.S. from wage suppression. Skeptics worry that a high price tag could still exclude cash-constrained startups, suggesting the system should more directly tie visas to graduating from U.S. universities or be paired with a higher salary floor to ensure it attracts top talent without hindering entrepreneurial companies.

The core of the discussion then pivots to AI’s tangible impact on productivity. Beyond the hype, the real story is a bifurcation in outcomes: while some studies show minimal gains, elite senior engineers and small, savvy startup teams are reporting 3x to 10x improvements. The key differentiator is a shift in workflow—from incremental, AI-assisted code completion to dispatching fully defined tasks to AI agents and then critically reviewing the output. This turns engineers into architects and editors. Furthermore, successful adoption is often bottom-up and personal, driven by individual engineers using tools like Cursor or ChatGPT, rather than top-down corporate mandates, which often fail due to overemphasis on control and deterministic outcomes.

This AI-driven velocity is creating a unique landscape for startups. For the first time, a small team can instantly achieve the scale of a much larger company through AI agents, neutralizing the traditional human-power advantage of incumbents. Meanwhile, distribution is less of a barrier since the technology is already in everyone’s pocket. This combination makes this AI wave a true platform shift, opening up vast new markets—particularly in verticals and professional services like agriculture, construction, or advertising—where software never existed before. The incumbents most at risk are not software companies, but service industries that can now be productized by AI-native insurgents.

Surprising Insights

• Early adopters are uniquely forgiving. Just as early internet users didn’t complain about slow dial-up speeds, today’s AI power users tolerate errors and non-determinism as part of the exhilarating process of exploring a new, magical tool.
• The biggest AI productivity gains go to experts, not novices. Contrary to the fear that AI democratizes all skills, its most powerful use as a “turbocharger” requires deep domain expertise to properly guide, judge, and integrate its output.
• AI is not lowering costs for professional creative work; it’s raising the ceiling. When professionals use AI for tasks like video or asset creation, they don’t spend less money or time—they instead produce more dazzling, iterative, and higher-quality outputs for the same budget.
• The most immediate job impact of H-1B reform may not be on software engineers. The discussion suggested the current system most acutely suppresses wages and availability for mid-tier IT admin and consulting jobs in the $80k-$120k range, more than entry-level coding roles.

Practical Takeaways

• Move from writing code to reviewing code. Engineer your workflow around giving detailed, high-level prompts to AI agents and focus your human effort on synthesis, architecture, and rigorous validation of the generated output.
• Favor bottom-up, individual AI tool adoption over top-down corporate pilots. Encourage and secure the use of powerful individual tools (like advanced code editors or ChatGPT) that employees choose themselves, as these grassroots movements often drive real productivity more effectively than centralized, controlled initiatives.
• For startups, aggressively leverage AI to compress development time and simulate scale. Use AI agents to rapidly prototype, research, and build, allowing a tiny team to operate with the output of a much larger one and compete directly with incumbents on speed.
• In policy advocacy, focus on freeing visas for startups and tying them to U.S. education. When discussing immigration reform, emphasize mechanisms like a price cap or direct visa grants for graduates of U.S. universities to ensure the system supports high-potential entrepreneurs and new companies, not just large corporations.