Building for trillions of agents (X article)

TL;DR

Levie argues software is entering an agent-first era: organizations may run 100x–1000x more agents than people, so future products must be designed for agent usability (APIs, automation interfaces, machine-operable workflows), not only human UX.

Key points

  • Agents are rapidly evolving from chatbot assistants into persistent workers with tools, code execution, memory, and sandboxed compute.
  • Adoption is spreading beyond coding into broad knowledge work (contracts, support, audits, research, sales materials, web transactions).
  • Net effect is not only substitution of existing tasks, but expansion of total work done (more simulations, prototypes, analyses).
  • This could produce enterprises with orders-of-magnitude more agents than employees.
  • Strategic shift: “make something agents want” (analogous to “make something people want”).
  • Product implication: become strongly API-first and machine-operable:
    • complete API coverage,
    • CLI/MCP support,
    • clean and deterministic interfaces,
    • programmatic account and lifecycle operations.
  • Business model implication: seat-based pricing may be insufficient; agent-native usage likely pushes toward consumption/volume-based models.
  • New infrastructure opportunities: agent compute/sandboxes, agent data/memory systems, identity, communication, payments, security/governance/compliance.

Why this matters

The essay frames a practical roadmap for builders: if agents become primary software users, competitive advantage shifts from human-centric UX alone to agent interoperability, reliability, and governance at scale.