When building multi-agent systems, tailor the output format to the recipient. While Markdown is best for human readability, agents communicating with each other should use JSON. LLMs can parse structured JSON data more reliably and efficiently, reducing errors in complex, automated workflows.
Related Insights
Implement human-in-the-loop checkpoints using a simple, fast LLM as a 'generative filter.' This agent's sole job is to interpret natural language feedback from a human reviewer (e.g., in Slack) and translate it into a structured command ('ship it' or 'revise') to trigger the correct automated pathway.
When building Spiral, a single large language model trying to both interview the user and write content failed due to "context rot." The solution was a multi-agent system where an "interviewer" agent hands off the full context to a separate "writer" agent, improving performance and reliability.
When building for AI-powered environments, design tools to be equally usable by humans and the AI model. An elegant, simple design for humans often translates directly into an effective tool for AI agents, simplifying development and promoting shared logic.
Long, continuous AI chat threads degrade output quality as the context window fills up, making it harder for the model to recall early details. To maintain high-quality results, treat each discrete feature or task as a new chat, ensuring the agent has a clean, focused context for each job.
Anthropic's Claude models are specifically trained on XML. By structuring system instructions using XML tags (e.g., <role>, <instructions>), you align with the model's training data. This provides better organization and can unlock additional functionality and more reliable outputs compared to using plain text prompts.
To get consistent, high-quality results from AI coding assistants, define reusable instructions in dedicated files (e.g., `prd.md`) within your repository. This "agent briefing" file can be referenced in prompts, ensuring all generated assets adhere to a predefined structure and style.
Separating AI agents into distinct roles (e.g., a technical expert and a customer-facing communicator) mirrors real-world team specializations. This allows for tailored configurations, like different 'temperature' settings for creativity versus accuracy, improving overall performance and preventing role confusion.
Instead of providing a vague functional description, feed prototyping AIs a detailed JSON data model first. This separates data from UI generation, forcing the AI to build a more realistic and higher-quality experience around concrete data, avoiding ambiguity and poor assumptions.
Documentation is shifting from a passive reference for humans to an active, queryable context for AI agents. Well-structured docs on internal APIs and class hierarchies become crucial for agent performance, reducing inefficient and slow context window stuffing for faster code generation.
While chat works for human-AI interaction, the infinite canvas is a superior paradigm for multi-agent and human-AI collaboration. It allows for simultaneous, non-distracting parallel work, asynchronous handoffs, and persistent spatial context—all of which are difficult to achieve in a linear, turn-based chat interface.
