Agent Is New App

Brett Taylor, former CTO of Meta, co-CEO of Salesforce, and current chairman of OpenAI, shares his strategic vision for how the AI market will evolve, with agents becoming the dominant product form factor.

The Three Segments of the AI Market

• Frontier/Foundation Model Market

  • Will consolidate to a small handful of hyperscalers and large labs
  • Not viable for startups due to massive capex requirements
  • Models deteriorate in value quickly, requiring scale to generate returns
  • Similar to cloud infrastructure market consolidation

• Tooling Market

  • "Selling pickaxes in the gold rush" - data labeling, eval tools, specialized models
  • Risky because it's "close to the sun" - adjacent to infrastructure
  • Vulnerable to being obviated by foundation model providers moving up the stack
  • Can succeed but requires differentiation that infrastructure providers can't easily replicate

• Applied AI Market (Agents)

  • "Agent is the new app" - the primary product form factor
  • Will evolve like the SaaS market with higher margins
  • Will pay "taxes" to model providers (using their APIs)
  • Focus will shift from technology to product and business outcomes

Why Agents Will Transform Software

• Autonomous Completion vs. Productivity Enhancement

  • Traditional software: makes humans slightly more productive
  • Agents: accomplish jobs autonomously without human intervention
  • Similar to PC-era productivity gains that eliminated entire job categories (e.g., drafting)

• Measurable Outcomes vs. Nebulous Productivity Claims

  • Traditional software: "If every salesperson sells 5% more..."
  • Agents: Clear, attributable outcomes (e.g., "this call was resolved without human intervention")
  • Enables outcomes-based pricing models that align vendor and customer incentives

• Evolution of Agent Companies

  • Today: Technical differentiation in orchestrating agentic processes
  • Future: Technical aspects become commoditized (like databases in SaaS)
  • Long-term: Differentiation through domain expertise and workflow optimization
  • Will spawn many vertical-specific agent companies for niche use cases

Outcomes-Based Pricing for AI

• Aligns incentives between vendor and customer

  • Only pay when the agent achieves the desired business outcome
  • Example: Call center pays only when AI successfully resolves customer issues
  • Creates true partnership rather than vendor relationship

• Requires measurable, attributable outcomes

  • Must be able to clearly measure when the agent succeeds
  • Different from consumption-based pricing (tokens/usage)
  • Tokens used ≠ value delivered (similar to measuring productivity by lines of code)

• Forces companies to be customer-centric

  • Must deeply understand customer problems to get paid
  • Can't just "throw software at the wall"
  • Requires ongoing optimization to improve success rates

Realizing Productivity Gains with AI

• Current tools are immature but improving

  • AI-generated code often has errors that require human review
  • Reviewing others' code (including AI code) is cognitively demanding

• Strategies for improving AI productivity today:

  • Self-reflection: AI supervising AI (code review)
  • Root cause analysis: Identify why AI makes mistakes
  • Context engineering: Improve the context provided to models
  • Create systems for continuous improvement rather than waiting for models to improve

• Building virtuous cycles of improvement

  • Use AI to identify opportunities for improvement
  • Systematically address failure cases
  • "Let AI put the needles at the top of the haystack"

Go-to-Market Strategies for AI Products

• Developer-led motion

  • Appeals to individual engineers within CTO departments
  • Works for platform products used by technical teams
  • Particularly effective for selling to startups

• Product-led growth

  • Self-service signup, trials, credit card purchases
  • Works when user and buyer are the same person
  • Effective for SMB software

• Direct sales

  • Necessary when buyer and user are different people
  • Making a comeback for AI products
  • Required for complex enterprise solutions

• Choose based on purchasing process, not preference

  • Analyze how decisions are made for your product category
  • Be first-principles about your go-to-market approach
  • Don't avoid direct sales just because it seems less modern