Build the systems agents need

  • Theo describes a period of feeling less like an engineer as language and framework details mattered less to his daily work. His response was to focus on improving the systems in which agents operate. 24:58 25:18
  • Concrete examples include exposing application actions through MCP, adding Web MCP to dashboards, and building cheaper execution environments using isolates. He says he built about five Rust applications that year while exploring these problems. 25:47
  • He estimates that at least half the T3 Code feature requests he sees could be fulfilled by prompting the existing product. This is his estimate, not a counted analysis of the issue tracker. 26:47

Example: display tools enable many outputs

  • T3 Code can produce an HTML visualizer and show it in an integrated browser. Theo wants an inline display capability so the agent can create visualizers directly within a conversation. 27:13
  • He frames that capability as infrastructure for users to generate many dynamic outputs, rather than building a separate fixed interface for every possible visualization. 27:41

His argument about engineering work

  • Theo calls the loss of confidence a "trough of despair." He argues that engineers can move through it by recognizing deeper systems problems that become valuable and feasible when agents help build software. This is his interpretation of the transition. 25:18 28:09 28:30
  • His examples of newly worthwhile work include his own operating system, compilers, and a JavaScript runtime for his cloud. He says both the incentive to solve such problems and the complexity one person can handle have increased. 28:59 29:31 30:03
  • He contrasts inspecting every SQL query with investigating the database-to-server connection method, including keeping connections warm without substantially raising costs. The engineering attention shifts toward system behavior. 29:31
  • His closing recommendation is to explore what agents can already do, then build what they still need to do more. The conclusion is motivational; the video does not measure a general increase in engineering productivity. 31:04 31:24