When Coding Agents Became a Team
In 18 months, coding agents took Rod from needing someone else to build his ideas to supervising parallel teams that review their own work.
Developed from a conversation between Pete Winn, Rod Bishop and Andy David
In May last year, Rod began testing whether AI could remove a constraint that had defined his career as a non-technical founder. Early coding agents were unreliable and expensive. He was spending between $2,000 and $3,000 a month on API calls for results that would look poor today. Yet they crossed the threshold that mattered. Before them, he could not build a technology product without another person. With them, he could.
The workflow then changed quickly. In December, Rod was still watching a single agent work inside an IDE. By January, credible multi-agent tools let him run several agents from a task board. Within months, he could leave multiple models working autonomously, ask them to review one another’s code and instruct an agent to generate its own code-review sub-agent. He could even request ten reviews before examining the result himself.
Rod describes the progression as moving from zero to one, then towards 10x. His role has shifted accordingly. Instead of approving every tool call or closely supervising implementation, he can bring human judgement in after the agents have built and scrutinised the work. By the time he inspects the output, he says it is usually good. The capability that once required costly, constant assistance has become a parallel, self-reviewing development team.