The Agent vs Application Decision

You're a startup founder in 2025 trying to figure out where to build in the AI market. You're seeing three distinct segments emerging:

1. Foundation Models (OpenAI, Anthropic, Google)

• Requires billions in CapEx
• Models deteriorate in value quickly as new ones release
• Only hyperscalers seem viable

2. AI Tooling (data labeling, eval tools, specialized models)

• Clear immediate demand
• But very close to the infrastructure layer
• Risk: What if OpenAI/Anthropic release exactly what you built on their next developer day?

3. AI Applications/Agents (Harvey for legal, Sierra for customer service)

• Selling outcomes, not tools
• Further from the infrastructure layer
• But requires deep domain expertise

You have a great engineering team and some capital. You're technical enough to build in any of these segments. Your investors are pushing you toward tooling (faster revenue, clearer market) but you're intrigued by building agents.

Here's what you're weighing:

Tooling pros: Immediate market, lots of companies need this, technically interesting, faster to monetize Tooling cons: Feels risky being so close to foundation model providers, many tools are commoditizing quickly

Agents pros: Larger eventual market, harder to commoditize, outcome-based pricing has better unit economics Agents cons: Requires domain expertise you don't have, technology is still immature, longer sales cycles