Stefan Poole

Welcome to the “Agent”-cy

AI Leaders emphasized the importance of community in growth, and the WordPress Foundation shows that in spades through its Makers contribution community, where creators from all over the world come together to bring their own insights to the WordPress ecosystem. The Leaders program also helped me focus on how I existed in the loop of these agentic workflows, a topic prevalent in WordPress’s insistence on creations that make the user better as well as the agent. Leaders emphasized anticipating failure modes. WordPress built that into agent skills by testing against common agentic errors before they reach users.

WordPress and AI Leaders together have helped me become a better planner and a better engineer. Here’s a little of what I’ve been getting into.

I’ve implemented testing harnesses and trust-and-tag policies to encompass my policy of taking ownership of the products that I make using AI. I aim to optimize my working time, and to that end, I prioritize my goals and separate my available time into manageable chunks to suit evolution in those particular spaces. I use a grill-and-plan framework (interrogate via /grill-me while inside of Claude’s Planning mode) when I’m creating a new product or workflow because I believe in starting from a stable and scalable foundation when working with any assistant, agentic or otherwise.

Through my work building agentic plugins and working with agents simulating complicated legal situations, I’ve become more certain than ever that a developer needs to stand in the loop, to grow with agentic assistance rather than letting it take over while our cognition withers away. It’s paramount that, as engineers, we know what we’re putting out, and that we know why. In all makings, we should keep this in mind: we will be the owner of the ending product. I believe that in doing these, we can become our own version of “software cyborgs”, working in tandem with these agentic advances, and growing exponentially with them towards a new future.

Navigating the Future: My AI Ethics in Development

In this new, quickly-developing digital reality, I’ve found it is very important for developers to meet their responsibility as humans in the loop; to verify, certify, and validify the findings our agentic assistants bring to us. One major facet of this responsibility is in terms of the guardrails we set for ourselves and our agents in terms of hallucination prevention in our workflows. For this particular issue, I’ve set up the following, “trust-and-tag” procedure:

When Claude gives me a answer I can’t completely verify, I will check it through a search for partial verification, if I have the space or time for that, which isn’t always the case. To maintain fidelity in these instances, I’ll either get an expert or colleague to verify or partially verify some of the information I get, or I’ll believe it in the moment, but add a tag to that facet of information in order to check it at another time, and to be aware at the moment that I may be working with false or stale information in that particular space.

This is how that procedure works as my ethical hallucination-prevention SOP throughout my projects:

Personal AI Ethics Criteria

  • [ ] Hallucination risk identified: Can I verify this independently, or does it need partial verification or flagging?
  • [ ] Verification method applied: Google search + trusted source, expert review, or trust-and-tag decision made
  • [ ] Uncertainty flagged in artifact: If unverified, is it marked so others know it came from AI and may need review?
  • [ ] Audit backlog maintained: Is this logged for post-hoc review if the project later reveals a problem?

WordPress Says:

Your core approach centers on integrating agentic systems within workflow engineering to enhance planning and development processes. By implementing testing harnesses and utilizing a grill-and-plan framework, you prioritize ownership and scalability in your projects. This ensures that you remain actively engaged in the development loop, fostering collaboration with AI while maintaining cognitive authority over the products you create.