# Teach Your Agent to Prove It's Done — Nick Nisi & Zack Proser, WorkOS **URL:** \ **Category:** Content Creators\ **Tags:** agents, ai-engineer, automation\ **Created:** [18 October 2026 07:00 UTC](https://www.artofsm.art/t/teach-your-agent-to-prove-its-done-nick-nisi-zack-proser-workos/26201 "2026-10-18T07:00:17Z")\ **Posts on this page:** 2\ **Page:** 1
**Author:** ![artesia](https://www.artofsm.art/user_avatar/www.artofsm.art/artesia/32/36_2.png) [@artesia](https://www.artofsm.art/u/artesia)\ **Post date:** [18 October 2026 07:00 UTC](https://www.artofsm.art/t/teach-your-agent-to-prove-its-done-nick-nisi-zack-proser-workos/26201/1 "2026-10-18T07:00:17Z")
[![](https://www.artofsm.art/uploads/default/original/3X/2/c/2c253a083dd5312440b9094ee90fc1310038e196.jpeg "Teach Your Agent to Prove It's Done — Nick Nisi & Zack Proser, WorkOS") ](https://www.youtube.com/watch?v=q4WC4TowkBo) In the workshop “Teach Your Agent to Prove It’s Done,” Nick Nisi and Zack Proser emphasize the importance of enabling AI agents to verify and prove their completed tasks through deterministic, repeatable verification processes using tools like Playwright, thereby enhancing trust, reducing errors, and accelerating development. They advocate for detailed action records, clear outcome-driven goals, and evolving developer roles focused on managing verification frameworks, envisioning a future where self-improving agents drive safer and faster software delivery. ---
**Author:** ![artesia](https://www.artofsm.art/user_avatar/www.artofsm.art/artesia/32/36_2.png) [@artesia](https://www.artofsm.art/u/artesia)\ **Post date:** [18 October 2026 07:24 UTC](https://www.artofsm.art/t/teach-your-agent-to-prove-its-done-nick-nisi-zack-proser-workos/26201/3 "2026-10-18T07:24:36Z")
In the “Teach Your Agent to Prove It’s Done” workshop by Nick Nisi and Zack Proser from WorkOS, the focus is on enhancing trust and reliability in AI agents by enabling them to verify their own work. The presenters highlight a critical incident where an agent deleted production databases within seconds, underscoring the need for robust verification mechanisms. They emphasize that as agents generate code faster than humans can review, traditional code review processes become impractical, necessitating new methods for agents to self-assess and prove task completion without constant human oversight. The workshop introduces a practical approach using tools like Playwright and custom scripts to create deterministic, repeatable verification processes. These tools allow agents to interact with applications, check UI states, validate data persistence, and run tests that confirm whether changes have been successfully implemented. By embedding these verification loops, agents can self-correct and iterate until the desired outcome is achieved, significantly accelerating development velocity while reducing errors and manual intervention. Nick and Zack also discuss the importance of maintaining detailed “receipts” or records of agent actions and outcomes. These records serve as tangible proof of work done, enabling both agents and humans to audit and trust the results. They envision a future where agents operate as self-evolving codebases, continuously improving through recursive self-improvement (RSI) loops fueled by comprehensive evaluation data. This approach not only boosts individual productivity but also scales across teams, fostering safer and faster software delivery. A key concept explored is the balance between giving agents creative freedom and providing clear, outcome-driven goals. Rather than prescribing exact steps, agents are guided by defined endpoints and verification criteria, allowing them to explore solutions while ensuring alignment with desired results. The presenters caution against vague instructions and advocate for precise, scriptable verification commands that reduce ambiguity and enhance determinism in agent behavior. Finally, the session underscores the evolving role of developers as “agent harness managers,” focusing on building and refining verification frameworks rather than hand-coding every detail. By equipping agents with the right tools and feedback loops, teams can delegate routine tasks confidently, concentrate on high-value work, and prepare for a future where AI-driven development is the norm. The workshop concludes with practical demonstrations, interactive exercises, and an invitation for ongoing experimentation and collaboration in this emerging field. ## Useful Links - [WorkOS Official Website](https://workos.com/) — Central to the video as the presenters are from WorkOS and the repo and tools are hosted there. - [Beyond Done Workshop Repository and Slides](https://www.youtube.com/watch?v=6Vl5O9Pwf04) — Directly supports the workshop content and enables practical experimentation with verification loops. - [GitHub CLI Documentation](https://cli.github.com/manual/gh_pr) — Supports the verification process by enabling automated PR management and evidence attachment. - [Recursive Self-Improvement (RSI) in AI Research](https://arxiv.org/abs/2607.07663) — Conceptual foundation for the approach to agent self-verification and continuous improvement discussed in the video.