The Mental Models I Use to Work with AI
9 hours ago
- #Productivity
- #Creative Workflow
- #AI Mental Models
- Invest in upfront alignment by making implicit assumptions explicit in the initial prompt to avoid unexpected AI outputs.
- Restart with better context rather than steering AI when initial outputs are far off; avoid compounding errors through path dependence.
- Equip AI agents with the same tools you use (CLI, MCP, browser control) to handle setup, testing, and instrumentation tasks.
- View early bad AI outputs as taste signals that help define creative direction, not just failures.
- Use visual references instead of lengthy prose prompts to guide visual outputs more effectively.
- Build a digital reference library to map your taste and provide raw material for AI to avoid generic results.
- Apply design principles to steer models away from slop and maintain unique, opinionated outputs.
- Use adversarial reviews with different LLMs to pressure-test outputs and reduce human bottleneck in review processes.
- Think ambitiously beyond MVP instincts as cheap execution allows for bigger, more innovative projects.
- Provide AI with tools to verify its own outputs (e.g., browsers, testing frameworks) to create internal feedback loops.
- Codify lessons from AI mistakes into memory systems (e.g., AGENTS.md) to prevent recurring errors.
- Recognize that LLMs reflect consensus, not judgment; your own taste and context are crucial for differentiation.
- Focus on clear direction and context over exact wording; systems matter more than prompt syntax.
- Be skeptical of impressive AI outputs in domains you lack expertise; verify to avoid misinformation.
- Understand the AI security lethal trifecta: combining private data, untrusted content, and external communication risks data exfiltration.
- Prioritize direction over execution as execution becomes commodified; mistakes in direction are more costly.
- Reframe SaaS AI tools as coworkers rather than just tools, explaining the shift towards chat-based interfaces.
- View loud AI debates as belief systems (like religion) rather than evidence-based arguments.
- See frustration with AI capability gaps as opportunities for new products and workflows.
- Evaluate SaaS ideas by whether a smart person with an LLM could replicate them easily; differentiate on brand, distribution, etc.