Summary
Justin Schroeder, co-founder of stealth startup StandardAgents, delivers a conference talk at AI Engineer arguing that domain-specific agents — not general-purpose AI assistants — will define the next wave of intelligent software. He frames the current moment as an accelerated Industrial Revolution where agents are the machines that harness intelligence, just as earlier machines harnessed energy.
The talk’s central technical argument borrows a classical software engineering principle: composition over inheritance. The inheritance model — adding skills or MCP servers to a single general agent — works up to a point, but degrades with scale. Past some threshold of added context, the agent experiences diminishing returns. Schroeder’s alternative is composing systems from small, focused domain-specific agents, each with a dedicated system prompt, minimal and precise toolset, and isolated message history, coordinated by a higher-level orchestrator that communicates in plain language. He draws on his own open-source projects — Dmux, a multiplexer for coding agents, and ArrowJS, a UI framework for the agentic era — to ground the architectural discussion.
Schroeder also addresses why businesses are building custom agents despite abundant off-the-shelf AI: the answer is integration. Organizations believe that properly integrating their own data and workflows into AI will produce dramatic gains, and domain-specific agents are the natural mechanism for that integration. He predicts this layer will eventually commoditize into infrastructure, analogous to how databases and operating systems became foundational. Engineers and founders thinking about multi-agent architecture and long-term agent ecosystem structure will find this a useful framework.
📺 Source: AI Engineer · Published June 29, 2026
🏷️ Format: Deep Dive
