Why it matters
Barr Yaron presents Amplify Partners’ survey of 1,048 AI practitioners: 97% described AI’s organizational impact as net positive, while more than 90% also reported downsides such as review burden and eroding technical depth. Agent write access tripled year over year, evaluation remained the most commonly cited stack problem, and teams tended to buy inference while keeping prompts, retrieval, and evals in house.
My takeaway: Survey enthusiasm and operational risk are rising together. Budget review and maintenance capacity alongside generation volume, instrument every write-enabled agent, retain internal ownership of evaluation criteria, and treat self-reported adoption data as directional evidence rather than proof that a particular architecture is mature.