← back
Bounded Autonomy: Between Free Will and Determinism — Angus J. McLean, Oliver
Takeaway
Bound agent autonomy with deliberate context and workflow constraints tied to the actual business task.
Summary
- Oliver uses agents for creative production, audience research, competitor analysis, and campaign localization, reporting around 4,000 assets daily.
- Advertising spend provides real-world performance feedback for iterating on generated material.
- The speaker recommends slowing tool adoption and designing around model limitations and finite context.
- Curated context acts as a soft constraint; unrestricted web research can overrepresent promotional and SEO-optimized information.
bounded-autonomycontext-engineeringadvertising
Original description
Angus McLean spent time building a complex agent application to generate his CV. Four letters beat it: HTML. He puts the improvement at 100x. The talk is from Oliver's AI Director, where agents generate around 4,000 creative assets a day for 200 plus brands, assets you have probably seen and had no idea were AI. The core argument: models are naturally verbose and tend toward complexity, and so are the developers working with them. His counter is to strip back. Replace internet access with curated documentation, ask how little context you can use and still complete the task, and never automate a job you cannot do yourself. Speaker info: / angusjmclean Timestamps 0:14 Introduction and talk theme: Bounded Autonomy 1:13 About Oliver and GenAI in advertising 2:24 The structure of an ad agency 3:35 Why agents are used for speed and scale 4:27 Slow down and look at model limitations 5:52 The problem with current "band-aid" solutions 6:51 The role of context windows in agent capability 8:11 How to effectively constrain models with documentation 9:29 Constraints as a driver for creativity 10:32 Fundamentals: Building your own harness and memory 11:25 The power of simplicity (The CV/HTML example) 12:34 AI as a translation process 14:14 Using multiple representation structures 15:22 Agent workflows and the "don't automate what you can't do" rule