While many are fixated on the raw power of AI models, the true transformation lies in the applications being built on top of them—software that fundamentally changes the economics of work by making people richer and lazier. This is the core thesis explored by a16z’s Alex Rampell, David Haber, and Anish Acharya, who argue that enduring AI companies won’t just be better versions of existing software; they will create entirely new categories by replacing or augmenting human labor directly. The conversation maps out a landscape where the speed of adoption is unprecedented, moving from novelty to enterprise essential in months, not years, and where defensibility comes not from the AI alone, but from building deep, product-based moats.

The discussion identifies three primary investment themes shaping the current AI wave. First is the “AI-native” transformation of traditional software categories, where new entrants like Rilli (for ERP) or CREA (for design) attack incumbent strongholds by targeting greenfield customers—new businesses or those at an inflection point who are open to a fundamentally better, AI-first solution. Second, and potentially largest, is software that directly substitutes for labor, tackling jobs where there are shortages, high turnover, or inefficiencies. Examples include Salient, which automates debt collection calls and increases recovery rates, and Eve, which handles the entire workflow for plaintiff-side attorneys, allowing them to take on more cases. The third theme is the “walled garden” business built on proprietary data that becomes immensely more valuable when combined with AI to deliver a finished product, such as Open Evidence for medical research or Ask Leo for procurement contracts.

Success in this new era hinges on building defensibility beyond just having a clever AI feature. The most compelling companies create “hostages, not customers” by becoming mission-critical systems of record. They achieve this by owning an end-to-end workflow, integrating deeply into a vertical, and leveraging unique, proprietary data that compounds over time. This moat is crucial because the underlying AI capabilities themselves are rapidly commoditizing. The speed of innovation means that while incumbents like Adobe or Intuit will also enhance their products with AI, massive opportunities exist for startups in greenfield markets and in creating software that performs jobs no traditional software was hired to do before.

Surprising Insights

• The “Brownfield” vs. “Greenfield” Distinction: For AI-native replacements of existing software (like a new CRM), it’s exceptionally difficult to displace an incumbent’s existing customers (“brownfield”). The real opportunity lies in capturing new companies or those at a key growth inflection point (“greenfield”).
• Labor Substitution Focuses on Value Generation, Not Just Cost Savings: The most successful “software eating labor” companies, like Salient in debt collection, are winning because they demonstrably generate more revenue (e.g., 50% higher collections) for their clients, not merely because they are cheaper than a human team.
• Incumbents Are Strongly Positioned to Benefit from AI: Unlike prior shifts like cloud or mobile, current software incumbents broadly recognize AI’s value and are actively embedding it into their products. This means they will likely become stronger and capture significant new revenue streams from their existing, sticky customer bases.
• Proprietary Data’s Value Multiplier: Data aggregation businesses (like PitchBook or CoStar) existed before AI, but AI acts as a force multiplier. By using AI to deliver a finished analysis or action—not just the raw data—these “walled gardens” can increase the value and price of their offerings by an order of magnitude.

Practical Takeaways

• Target Unmet Demand in Labor Markets: Look for repetitive, high-turnover, or unfulfilled job roles (e.g., certain types of customer support, administrative work, specialized research). Building software that does 80-90% of that job reliably can open a massive new market.
• Build Workflows, Not Just Features: To achieve defensibility, design your product to be a system of record that owns an entire workflow for a specific profession or business function. This creates switching costs far beyond a single AI capability.
• Seek Proprietary Data Feedback Loops: When evaluating an idea, ask: can usage of this product generate unique, non-public data that makes the product itself smarter and more effective over time, creating a compounding advantage competitors cannot access?
• Prioritize Greenfield Customer Acquisition: If you’re building a better version of an existing software category, focus your early sales and marketing efforts on brand-new companies or those experiencing a clear “rip-and-replace” moment due to growth or pain, rather than trying to convince satisfied users of an old system to switch.