Teams rewrite workflows to stop supervising AI

Rethinking human oversight. Melina’s team at the Design Systems Collective mandates that any change to agent memory passes a QA agent check, followed by a person approving the pull request (Design Systems Collective). When agents fail to agree after three rounds, the report moves to a person for resolution instead of looping further. Justyna Tarkowska requires that every change to agent skills and checks goes through a pull request, treating AI updates like standard code.

Measuring workflow outcomes. Lenovo assesses AI readiness using a five-point index across architecture, operations, trust, workforce, and value, finding that businesses embedding AI across core functions score 4.1 compared to 2.7 for those still evaluating potential applications (Workplace Insight). Rakshit Ghura at Lenovo tracks workflow-level indicators such as process completion times and repeat incidents to confirm AI value, as conventional productivity metrics fail to capture meaningful improvements. Only 12 percent of organizations measure adoption and effectiveness at the team level to refine training and technology decisions.

Updating roles and responsibilities. Kyndryl surveyed 3,700 senior business leaders and found 48 percent changed their AI strategy this year, yet only 27 percent formally updated roles and responsibilities (Kamil Banc from AI Adopters Club). Phil’s team at IBM identified existing systems blocking transformation by asking participants which processes hindered their work and then removing those specific obstacles (Tiffani Bova). IREM and AppFolio report that operators in their highest maturity band achieve 94 percent productivity gains and 67 percent net operating income gains, while those in the lowest band report 21 percent and 15 percent gains respectively (The CRE Report).