a16zThe Economics of AI Usage and What's Next For SaaS | Benedict Evans on a16z
At a glance
WHAT IT’S REALLY ABOUT
Benedict Evans on AI economics, commoditization, and SaaS’avenir ahead
- Agentic coding has become the first clear “pull” use case for LLMs, concentrating product focus while other daily-use consumer workflows remain uncertain.
- The market is in a temporary disequilibrium where token demand, infrastructure capacity, and pricing are misaligned—similar to early mobile data shocks—so today’s economics likely won’t persist.
- Evans argues foundation models and chatbots are unlikely to be durable end-products because differentiation and network effects are weak, pushing value creation up the stack into software, workflows, and domain solutions.
- Historical platform analogies (PCs, web, mobile, telecom) are useful for asking questions but not for predicting winners; at this stage multiple paths remain plausible.
- In enterprise and services industries, the biggest impacts may come from reorganizing work (tasks vs jobs, pyramid structures) and from AI enabling entirely new analyses and decisions, not just faster versions of old workflows.
IDEAS WORTH REMEMBERING
5 ideasCoding is the clearest near-term AI wedge, but it doesn’t answer “then what?”
Evans sees agentic coding as the first indisputable PMF that customers “pull,” yet the larger question is which other domains will cross that threshold and become daily habits.
Today’s token economics resemble early mobile data chaos—expect pricing systems to normalize.
He compares surprise token bills and underpriced subscriptions to 2009–2010 mobile data, where operators had to realign pricing, throttling, and bundles to match marginal costs and capacity.
Foundation models may behave like commodity infrastructure more than platforms like iOS.
Evans doubts sustainable differentiation and notes the lack of obvious network effects; if multiple labs offer similar capability on similar chips, long-run pricing power is hard to justify.
“Chatbot” is a V1 interface, not the product end-state.
Most valuable applications require tooling, data access, guardrails, and workflow design—akin to purpose-built software (TurboTax, InDesign) rather than generic prompting.
AI likely increases software proliferation, not consolidation into a single model UI.
Enterprise work already spans big horizontal systems (SAP/Workday), hundreds of SaaS apps, and “Excel/email” improvisation; AI becomes another option that spawns more apps, features, and internal tools.
WORDS WORTH SAVING
5 quotesAgentic coding went from being kind of useful to really changing everything. It's going to be magic, and in 20 years' time we'll just say, "Well, of course, that's how it is. Computer's always done that."
— Benedict Evans
I don't think foundation models are a product. I don't think a chatbot is a product. I think the value will be further up.
— Benedict Evans
History teaches us nothing except that something will happen.
— Benedict Evans
One of the characteristics of tech is that the moment that you understand something and you know how it works and what's gonna happen is the moment you should move on to something else.
— Benedict Evans
We are in this extreme scarcity. Like, we can't spend $10 trillion a year on our AI infrastructure 'cause there isn't $10 trillion a year there to spend on it.
— Benedict Evans
High quality AI-generated summary created from speaker-labeled transcript.
