Jon Udell rejects the phrase “human in the loop” because it suggests machines hold authority over the workflow. Instead, he proposes a model where human engineers remain the primary decision-makers while recruiting autonomous agents to join their existing teams. This approach treats software development as a collaborative process rather than a black box where prompts generate features without oversight. The goal is to integrate these tools into standard engineering practices, ensuring that every step remains transparent and reviewable by people who understand the codebase. Udell argues that agentic workflows should not exclude humans from the critical path of building software. By inviting agents into the fold, developers maintain control over quality and logic while gaining the efficiency of automated assistance. This shift moves away from treating AI as a replacement for oversight and positions it as a junior team member.
The distinction matters because it changes how teams manage risk and accountability in automated environments. Keeping humans at the centre of the process prevents the erosion of professional judgement that occurs when systems operate as isolated loops.
- Agents act as invited team members rather than autonomous authorities.
- Code generation must remain fully reviewable by human engineers.
- Workflows avoid the pitfalls of unreviewable pull requests.