High-Velocity Development via Multi-Agent Orchestration

The Philosophy of High-Output Engineering

Theo argues that maintaining high-velocity code contributions while managing a business requires a shift from 'hands-on' coding to 'orchestration-based' coding. The core bottleneck for most developers is not typing speed or technical skill, but cognitive load and context switching. By treating AI agents as autonomous workers, he offloads the burden of research, auditing, and implementation, allowing him to manage 50+ PRs per week while traveling.

Infrastructure and Parallelization

Crucial to this workflow is offloading execution from the local machine. Theo emphasizes that running multiple agent threads on a local Mac is inefficient and prone to system instability. Instead, he utilizes a dedicated, over-spec Linux box to handle all agentic workloads. This separation ensures that his primary machine remains responsive for communication and management tasks, while the Linux server handles heavy parallel tasks like codebase audits, testing, and compilation without hitting CPU or memory bottlenecks.

The 'Burn Tokens, Not Motivation' Strategy

Theo’s primary heuristic for productivity is to avoid demotivating tasks at all costs. If a task requires manual setup (like configuring auth or testing environments) that risks failure and frustration, he instead 'burns tokens' by asking AI agents to audit the code, predict failure points, or prioritize his backlog. He treats tokens as a cheap resource compared to his own time and focus. If an agent’s output is difficult to parse, he doesn't force himself to read it; he prompts the agent to simplify or reformat the information, maintaining momentum by keeping the cognitive friction low.

Managing Agentic State

To prevent getting lost in a sea of open threads, Theo implemented a 'settle' workflow in his IDE (T3 Code). Threads are treated as active tasks; once a PR is merged or a task is completed, the thread is 'settled' (archived). If a task is not immediately actionable, he uses a 'snooze' feature to hide it from his active view. This keeps his workspace clean and ensures he only interacts with threads that require immediate human intervention.

Notable Quotes

• "A CEO that is a part-time coder is somehow probably outshipping you."
• "One of the things I optimize for is preventing loss of motivation."
• "If you spend all of your time sitting and watching threads waiting for them to have an answer you're not being productive."
• "I treat my Macs as one or two tasks at a time and I treat my Linux boxes as effectively unlimited tasks at a time."