Summary
Alex Bauer, co-founder of Upside.tech (a go-to-market data platform for the agentic AI era), presents at the AI Engineer conference on practical design patterns for building trustworthy AI agents in business contexts. His central thesis: rather than chasing technical optimizations or magic prompt incantations, the most effective approach is to manage AI agents the way you would manage other humans — with clear intent, appropriate scaffolding, and structured verification mechanisms.
Three concrete examples anchor the talk. First, Bauer describes rebuilding Upside’s entire website with Claude: pure YOLO mode failed, but adding business context documentation (anchor assets, company definitions, workflow guides) unlocked reliable output — a task that at his previous 600-person company would have required months and a full team. Second, a “jury system” pattern where multiple agents independently evaluate the same output before surfacing answers to users, addressing the hallucination-as-confident-answer problem. Third, tiered agent permissions that match trust level to task stakes, drawing on principles from military commander’s intent doctrine: tell agents why, not just what, and they perform significantly better — while also resisting their trained tendency toward self-micromanagement.
The talk is aimed at go-to-market and business teams who lack deep ML expertise but are increasingly expected to build and deploy AI workflows in production environments, framing trust design as a familiar human-management problem rather than a novel technical challenge.
📺 Source: AI Engineer · Published July 11, 2026
🏷️ Format: Keynote Launch
