Researchers tracked 112 seasoned developers using AI agents in real work and found they never hand off vague prompts and trust outputs blindfolded. Instead they plan architecture, review every diff, limit tasks to small scopes, and supervise the AI like a junior dev. Letting go led to a 92% failure rate in production and a 19% drop in productivity.

Engineers face difficulties in transitioning from deterministic programming to probabilistic agent engineering, as they often struggle to trust the adaptive capabilities of AI agents. Traditional practices, such as strict typing and error handling, clash with the need for flexibility and context-aware interactions in agent systems. Emphasizing the importance of semantic understanding and behavior evaluation, engineers are encouraged to embrace a new approach that balances trust and oversight.