The Economics of AI Usage and What's Next For SaaS | Benedict Evans on podcast insight

Episode At a Glance

• Podcast: a16z show
• Episode: The Economics of AI Usage and What's Next For SaaS | Benedict Evans on a16z
• Guests: Benedict Evans
• Hosts: Erik Torenberg
• Published: June 8, 2026
• Duration: 1 hr 33 sec

Episode Overview

• Summary: Benedict Evans argues that AI has clearly broken through in coding, but many of the bigger questions around consumer use, SaaS, model differentiation and value capture remain unresolved. The episode frames AI economics through earlier platform shifts: mobile data, cloud, browsers, spreadsheets and enterprise software adoption.
• Central question: Where will AI create durable value once the current scarcity of tokens, compute and model performance settles into a more normal market?
• Core argument: Coding has product-market fit now, but the broader AI economy depends on pricing, workflow design, infrastructure costs and whether models stay differentiated.
• Why it matters: The answer shapes SaaS valuations, enterprise adoption, foundation model strategy and whether AI value accrues to infrastructure or applications.

I used PodFaro to organize the podcast transcript into structured notes.

👉 Core Insights

1. Coding is AI first breakout market

Evans says the clearest change since his earlier AI thesis is that agentic coding moved from being useful to changing how developers work. He compares the moment to early PCs and the early internet: exciting, powerful, but still unstable and not fully understood. The first proven AI market is software development, not because it is the only market, but because developers are pulling the product into use.

2. Most AI questions remain open

Despite growing usage, bigger models and rising capex, Evans argues that many core questions from two or three years ago still lack answers. He points to uncertainty around model winners, value capture, daily consumer behavior and whether chatbots are enough as products. AI has momentum, but the market structure is still unsettled.

3. Token pricing echoes mobile data

Evans compares AI pricing today to mobile data in 2009 and 2010, when usage surged faster than capacity and pricing models broke. Flat-rate bundles, surprise bills, throttling and capacity investment eventually forced telecom pricing back toward cost and perceived value. AI pricing is in disequilibrium because demand, supply, marginal cost and perceived value have not yet lined up.

4. Foundation models may become infrastructure

Evans does not claim model commoditization is certain, but he lays out why it is plausible. If models lack durable network effects and applications need workflow, data, interfaces and configuration, value may move above the model layer. The open question is whether foundation models become the product or the infrastructure beneath better products.

5. SaaS faces more competition and ambiguity

AI should make software cheaper and faster to build, which implies more competition and new margin structures. But Evans cautions that enterprise software is tangled across core systems, vertical apps, spreadsheets, email and internal tools. The SaaS impact is real, but it will not be evenly distributed or easy to price.

6. ROI may become consumer surplus

Evans says early AI benefits often show up as better analytics, customer support and productivity, which are hard to attach to precise revenue. Like Excel, AI may let people do much more work without letting companies charge proportionally more for that work. Some AI value will become a competitive necessity rather than a visible profit line.

👉 Stories from the Conversation

1. Anthropic Focuses on Coding

Evans contrasts OpenAI experimentation with Anthropic focusing on coding after raising less capital. Whether intentional or stumbled into, that focus worked because software development became the place where customers visibly pulled the product forward.

Speaker: Benedict Evans

Why it matters: It explains why coding became the reference use case for current AI product-market fit.

2. A Commodities Cash Flow Use Case

Evans describes a commodities company using LLMs to improve cash flow forecasting because it deals with many small producers and uncertain invoice timing. That use case differs from casual chatbot usage because it is a focused back-office problem with operational consequences.

Speaker: Benedict Evans

Why it matters: It grounds enterprise AI in measurable workflow problems outside Silicon Valley.

3. Mobile Data Pricing Shock

Evans recalls the mobile data era when users could receive huge bills or overload flat-rate networks as iPhone usage took off. Carriers had to align pricing, capacity, marginal cost and perceived value through bundles, throttling and network investment.

Speaker: Benedict Evans

Why it matters: It gives a concrete historical model for AI infrastructure economics.

4. The Salesforce Workflow Question

Evans uses enterprise workflow examples to ask whether a process belongs in Workday, Salesforce, a vertical app, Excel, email or a new AI-built internal tool. He argues that AI enters a messy landscape where companies already move work between packaged software and improvised systems.

Speaker: Benedict Evans

Why it matters: It clarifies why SaaS change will be fragmented across workflows and organizations.

👉 Memorable Quotes

👉 Data Highlights

👉 Points of Debate

1. Will coding generalize?

Host view: Erik asks whether coding could have been predicted as the first breakout use case and what comes next.

Guest view: Benedict says coding makes sense because developers were the first users, but no one knows which non-coding category will match it.

Where they agree: They agree coding has unusually clear product-market fit today.

2. Where does value accrue?

Host view: Erik compares AI to internet, cloud and hardware layers where value settled differently.

Guest view: Benedict argues comparisons are useful but not predictive; models may become infrastructure while value moves up the stack.

Where they agree: They agree the current cycle still has multiple possible paths.

3. Is SaaS facing apocalypse?

Host view: Erik asks whether AI implies a less consolidated SaaS industry and what investors should do with software stocks.

Guest view: Benedict expects more competition and some companies to be hurt, but says it is unclear which ones and by how much.

Where they agree: They agree AI will change software margins and competition.

4. Will AI spend face reckoning?

Host view: Erik asks whether companies are overshooting AI usage and may pull back after proper ROI studies.

Guest view: Benedict says pricing must realign with cost and ROI, but early benefits are often real even when difficult to measure.

Where they agree: They agree the current token and infrastructure market is not yet in equilibrium.