A six-principle manifesto for operating agentic systems, drawing on promise theory (Burgess) and Deming's quality management ideas. The principles emphasize: observation must precede action; distinguish durable principles from removable scaffolding; governance should address observed rather than imagined failures; human judgment belongs in system design not output inspection; precise specification is the real bottleneck; and four substrate properties (trust infrastructure, observable state, decision traces, explicit trust boundaries) must exist before removing process scaffolding. The author frames this as a starting point for a theory of agentic operations that doesn't yet exist, inviting practitioners to stress-test and break the principles.
Table of contents
1. Observation is the primary operation #2. Scaffolding is removable. Principles are not #3. Governance flows from observation, not from control #4. Human judgment belongs in system design, not in inspection #5. Precise thinking is the scarce resource #6. The substrate before the removal #What this is for #Questions this post answers
What substrate properties need to be in place before removing human oversight from an agentic system?
Four properties must exist before scaffolding is safely removable: trust infrastructure (failure is safe to report, authority is bounded and explicit), observable state (typed, versioned, queryable), decision traces (every judgment leaves a record), and explicit trust boundaries (who authorizes what, under what conditions). Removing ceremonies without these produces chaos with faster iteration. Teams shipping agentic workflows track emerging operational patterns like these on daily.dev.
Where does human judgment add the most value in an agentic or automated system?
Human judgment is most valuable in system design, not in reviewing outputs. The highest-value work is defining what to observe, what counts as drift, where to draw trust boundaries, and what evidence to demand. Reviewing outputs in a queue is maintenance; designing the system that produces trustworthy outputs is architecture. Architects deciding how to structure human-in-the-loop agentic pipelines find relevant debates on daily.dev.
7.4K Impressions2 Comments