Why AI Still Needs Human Judgment

Dan Wald, cofounder and chief AI officer at Sciemo, joins The Tech Trek for a sharp conversation about what AI can and cannot do inside real business workflows. The big question: can AI move beyond quick answers and actually support the messy, context heavy work that still lives in Excel, data teams, and functional expertise? Dan breaks down why consumer style AI has trained people to expect instant answers, why that creates risk inside companies, and why the next wave of AI products needs more than a chat box. It needs context, transparency, guardrails, and humans who understand the work well enough to challenge the output. The conversation also gets into AI agents, coding, entry level talent, narrow workflow specific AI, and why replacing judgment is a much harder problem than replacing repetitive tasks. Key takeaways • AI tools are only useful when they understand the context behind the question, not just the wording of the prompt. • Excel remains powerful because users can see the data, change assumptions, and understand the logic. AI products need to earn that same level of trust. • The best AI workflows are not black boxes. They let users inspect assumptions, challenge outputs, and adjust the answer. • Agents can speed up work, but they still need human judgment, especially when the task requires strategy, constraints, or domain expertise. • AI may change entry level work, but companies still need people who can think critically, solve new problems, and understand why the output is right or wrong. Timestamped highlights 00:40 Dan explains how Sciemo helps consumer brands unify messy data and apply AI to inventory, pricing, assortment, and promotion decisions. 02:30 Why the single prompt experience has changed what people expect from AI, and why that expectation can break down inside the workplace. 04:19 How purpose built AI differs from general AI, especially when the workflow requires context, guardrails, and a clear goal. 07:41 Why Excel is still hard to replace, and what AI systems need to learn from the control and transparency users already expect. 12:57 Dan compares AI agents to unlimited interns, useful for many tasks, but still limited without expert direction. 21:57 The slap chop analogy, and why faster tools do not automatically make someone better at the underlying craft. 31:15 Why predictions about technology and work are so hard to get right, even when productivity clearly improves. A line that stuck “Used properly, they’re great. Used poorly, it’s a very new technology. There will be more mistakes than there are winners.” Practical points worth taking • Do not treat a confident AI answer as a complete answer. • Build AI around real workflows, not generic prompts. • Keep humans close to the assumptions, especially when the decision has business impact. • Use AI to move faster, but make sure someone still understands the logic behind the work. Listen next Follow The Tech Trek for more conversations with founders, operators, and technical leaders building through the next wave of AI, data, and product change.