Employ AI agents like hires: 170 questions before the job

The article is written by an author who talks regularly with data and AI leaders at global companies about agents in production, not pilots. The pattern those leaders describe is a shift in posture: prompts describe tasks, while job descriptions describe roles, including who is accountable for output. Many companies skipped that step, which the author links to concern about human oversight and rogue actions.

The examples span a music technology company that audits workflows for repetitive, high-volume work, a digital marketing agency that placed a solution architect between AI teams and business units, and an energy solutions company that found building its repair-engineering agent took about 20 minutes while gathering data from scattered sources took a multiweek effort. Snowflake's own research is cited on data readiness: 65% of companies say breaking down AI data silos is challenging or very challenging, and 62% say the same about prepping data to be AI ready.

Managing agents is where programs can fall apart, the article argues, and every leader interviewed gave the same answer on ultimate responsibility — humans are. Trust, one leader estimated, takes eight or nine months of iteration to earn. The open question the piece leaves is whether management structures can keep pace with the technology, since the author frames the distance between having agents and employing them well as a management gap rather than a technical one.