Software for Agents

The entire internet was built for humans clicking buttons and reading screens. As AI agents become the primary 'users' of software — making API calls, processing data, executing workflows — they run into infrastructure designed for a different species. Authentication assumes a human logging in. Rate limiting assumes human typing speed. Error messages are written for human eyes. Payment systems require human approval. Every major software category needs to be rebuilt with agents as the primary user.

2025-2026 marks the transition from AI assistants (human-in-the-loop) to AI agents (autonomous execution). Companies like Devin, Cognition, and dozens of agent startups are deploying agents that operate independently for hours or days. These agents need infrastructure: identity management, payment processing, monitoring, debugging, rate limiting, and access control — all designed for non-human users operating at machine speed.

Developer infrastructure is a $50B+ market. Every category within it — observability (Datadog, $40B), identity (Okta, $15B), payments (Stripe), API management — needs an agent-native equivalent. The agent infrastructure market barely exists today but will grow to tens of billions as agent deployments scale. Early movers will define standards and protocols that become impossible to displace, similar to how Stripe defined online payments.

Infrastructure engineers who have built developer tools, APIs, or platform services at scale. Ideal backgrounds include working at Stripe, Datadog, Cloudflare, AWS, or similar infrastructure companies. You need to deeply understand both the developer experience and the technical challenges of building reliable, high-throughput systems. Founders who are actively building with AI agents and have felt the infrastructure gaps firsthand are uniquely positioned.

Aaron Epstein makes a bold prediction: the next trillion internet users won't be human. They'll be AI agents — and every major software category needs to be rebuilt for them. This isn't about adding an API to existing software. It's about rethinking fundamental primitives: How does an agent authenticate? How does it pay for services? How do you monitor millions of agents operating autonomously? The founders who build this infrastructure will power the next era of the internet.