Skip to content
Lenny's 🏦 Vault

Three AI Market Segments: Avoid Foundation Models

by Brett Taylor on August 2, 2025

Brett Taylor outlines a strategic framework for understanding how the AI market will evolve, where opportunities exist for founders, and how business models are fundamentally changing with agent-based software.

AI Market Structure: Three Distinct Segments

  • Frontier/Foundation Model Market

    • Will consolidate to a small handful of hyperscalers and large labs
    • Requires enormous capital expenditure (capex) - not viable for startups
    • Models deteriorate in value quickly, requiring constant reinvestment
    • Similar to cloud infrastructure market consolidation pattern
    • Advice: "No entrepreneurs should build a frontier model" (unless you're Elon Musk)

  • AI Tooling Market

    • "Selling pickaxes in the gold rush" - data labeling, platforms, eval tools
    • Examples: specialized models like Eleven Labs' voice models
    • Warning: "Close to the sun" - at risk of being obviated by foundation model providers
    • Similar to how cloud providers eventually built competing products to Confluent/Databricks
    • Viable but requires differentiation that can withstand competition from infrastructure providers

  • Applied AI Market

    • "Agent is the new app" - building autonomous solutions for specific business problems
    • Examples: Sierra (customer service), Harvey (legal), content marketing, supply chain analysis
    • Higher margin businesses selling business outcomes rather than technology
    • Will "pay taxes" to model providers but maintain better margins
    • Most promising area for startups and entrepreneurs

Why Agents Will Transform Software

  • Autonomous vs. Assistive

    • Traditional software helps humans be more productive
    • Agents actually accomplish jobs autonomously
    • "Software is going from helping an individual be slightly more productive to actually accomplishing a job autonomously"

  • Measurable Productivity Gains

    • Unlike traditional productivity software where gains are hard to attribute
    • Agent performance is directly measurable (e.g., call containment rates)
    • "Not only is it actually truly driving productivity in a very real way, but it's measurable as well"

  • Outcomes-Based Pricing

    • Traditional software: seat-based or consumption-based pricing
    • Agent software: priced based on business outcomes achieved
    • Example: Sierra charges per customer service issue resolved, not per token or seat
    • "The whole market is going to go towards outcomes-based pricing... it's just so obviously the correct way to build and sell software"

Building Effective AI Systems

  • Self-Reflection and Layered Intelligence

    • Use AI to supervise AI - "having AI supervise the AI is actually very effective"
    • Layer multiple AI systems to achieve higher accuracy
    • "If you produce an AI agent that's right 90% of the time... make another AI agent to find the errors... wire those things together and have something that's right 99% of the time"

  • Root Cause Analysis

    • Don't just fix errors - understand why they occurred
    • Improve context engineering to prevent future errors
    • "Rather than just fixing it, try to root cause it... what context did [the AI] not have that would have been necessary to produce the right outcome?"

  • Systems Thinking

    • Don't wait for models to "magically get better"
    • Create virtuous cycles of improvement
    • "You really have to think of this as a system... if you want the gains now, you gotta put in the work"

Go-to-Market Strategy for AI Products

  • Developer-Led

    • Appeals to individual engineers within CTO departments
    • Works for platform products used by technical teams
    • Examples: Stripe, Twilio
    • Best for products where engineers have latitude to choose solutions

  • Product-Led Growth

    • Self-service signup, trials, credit card purchases
    • Works when user and buyer are the same person
    • Ideal for small business software
    • Fails when buyer and user are different stakeholders

  • Direct Sales

    • Traditional enterprise sales motion
    • Necessary when buyer and user are different people
    • "I've seen more recently a lot of AI companies direct sales come a little bit more back into fashion"
    • Many AI opportunities require this approach despite its reputation

  • Match Go-to-Market to Purchase Process

    • "Where I see entrepreneurs stumble is they'll choose a go-to-market motion without thinking through the process of purchasing the software"
    • Be first-principles about who evaluates, who buys, and how decisions are made
View full episode