World's Fair 2026
What Does Done Even Mean? Agents and Paperclip's Liveness Model - Dotta, Paperclip
Overview
This talk addresses the ambiguity of "done" in the context of AI agents, arguing that a simple boolean check is insufficient. It proposes that "done" is a bundle of claims requiring verification across multiple dimensions, including artifact production, evidence of completion, adherence to a rubric, clear ownership for next steps, and survival of real-world conditions. The core thesis is that agent systems need a liveness model to manage the tension between keeping work moving and ensuring its quality.
Who should watch
- AI Engineers building agentic systems
- Product Managers defining task completion criteria
- Builders seeking to improve AI agent reliability and verification
- Those struggling with AI agents creating more work than can be verified
Key takeaways
- Agents can produce code and documentation faster than humans can verify, leading to a new failure mode of creating unmanageable workloads.
- "Done" is not a single state but a bundle of claims: artifact produced, evidence of completion, adherence to a rubric, defined next steps, and owner.
- There are different levels of "done," from producer claims to reviewer approval and survival of real-world conditions.
- Exhaustive human verification is impractical at high volumes, leading to "verification theater."
- A protocol is needed to define task progression, ensuring work moves without getting stuck in invalid states, tied to specific contracts and constraints.
- Balancing "liveliness" (work continuing without blockers) and "verification" (assurance of correctness) is crucial; too much liveliness leads to AI slop, while too much verification creates review queues.
- Agentic work requires a control plane that ensures productive work continues, only real blockers stop work, and infinite loops are bounded.
- Treating "done" as an object rather than a boolean allows agents to distinguish between artifact, scope, rubric, evidence, verifier, approver, and next action.
Notable quotes
*Programming is solved, and agents can now produce more code and documentation faster than any human can ever verify.*
*Saying that something is done is actually a bundle of claims.*
*You want to keep work moving, but also having it verified.*
Unofficial community note. Prefer the recording for nuance.