Interview, Fireside Chat, Conference Presentation
The Economics of AI Usage and What's Next For SaaS | Benedict Evans on a16z
- Agentic coding is projected to transition from a useful tool to the default method of operation within a timeframe ranging from two to twenty years, a shift where current capabilities will be viewed as inevitable and normalized.
- Software development is expected to remain the primary use case with immediate product-market fit, while other applications like cash flow forecasting, churn improvement, and advertising will mature more slowly as AI moves beyond simple correlations to understanding causal "why" factors.
- Infrastructure spending of $1–2 trillion is anticipated over the next couple of years, followed by a correction from disequilibrium to a pricing and capacity equilibrium similar to the mobile data industry, where traffic may increase 1,500 to 2,000 times.
- The market is forecast to settle with three to six companies producing frontier models, facing thin margins and commoditization similar to telecommunications, with model relevance lasting only three to six months at a time.
- A "trillion two trillion dollar" deployment of capital will likely not be sustainable indefinitely due to physical and financial limits, potentially leading to a pullback in usage as companies conduct proper ROI studies after realizing consumer surplus.
- Foundation models will likely evolve into commodities with marginal cost pricing, creating a "murderous price war" where pricing systems align with costs and developers freely swap between identical models.
- Value capture is expected to shift away from the layer raising the most money (foundation models) to the application layer, potentially requiring 300+ apps built on top of models since they lack network effects and leverage.
- New business models will emerge where AI agents interact directly with system-of-record software, bypassing traditional human interfaces, while the "SaaS apocalypse" may see a percentage of existing companies wiped out as software solves problems created by other software.
- The software engineering market structure faces uncertainty regarding the roles of junior and senior engineers, with no deterministic prediction available for the market configuration even three years into the future.
- Large technology corporations like Google, Meta, and Microsoft are expected to continue heavy capital expenditures to avoid an existential threat comparable to Microsoft in the 2000s or IBM in the 1990s.
- Productivity gains will be competed away through consumer surplus, where companies perform more work with fewer people for the same price, though specific winners in the emerging landscape remain unknown.
- Specific industries achieving product-market fit equivalent to agentic coding are not certain beyond the current focus on software, despite the infinite nature of demand for such technology.
- The industry will eventually shift from doing the "old thing but more" to enabling previously impossible or cost-prohibitive tasks, such as rebuilding Linux or YouTube from scratch.