Markdown plans from AI agents are becoming too long and unreadable. HTML allows for richer, more engaging artifacts with visuals and better formatting. This improves human oversight and collaboration with the AI, as the plans are more likely to be read and understood by the engineer.
The use of rich HTML artifacts extends beyond code plans to internal communications. By having an AI read Slack messages and generate a weekly status update in HTML, communication becomes more engaging and consumable for managers. This is a practical application of AI to improve the effectiveness of routine internal reporting.
As AI agents run for longer periods, the primary decision is no longer just about engineering time but about allocating expensive compute resources. The product manager's role shifts to deciding which tasks are valuable enough to spend significant AI compute budget on, a decision made during the spec and planning phase.
Instead of editing a complex AI-generated plan via text prompts, ask the AI to build a custom, throwaway HTML interface for a specific part of the plan (e.g., a table of rules). This "micro software" provides a more intuitive way to interact with and modify the plan, improving the quality of human feedback.
Instead of relying on scattered design docs or linking a repo, generate a "living design system" as a single HTML file. This artifact visually represents colors, typography, and components. It's easily passed to an AI agent in any new project, providing a compressed, comprehensive, and visual understanding of design constraints.
As AI agents become more autonomous and capable of executing complex, long-running tasks, the upfront planning phase (specs, PRDs) becomes even more critical. These documents are no longer just for human alignment but are essential for directing the AI and ensuring its expensive compute time is spent on the right objectives.
The best prompts strike a balance between providing enough specific information (e.g., "include code excerpts") and not over-constraining the model. Adding a phrase like "whatever is needed to give me maximum context" gives the AI an "out" to use its own judgment and provide additional, helpful information you didn't ask for.