When using AI coding tools, define the user's desired outcome and problem to solve, rather than prescribing a specific app. This prevents the AI from making incorrect assumptions about the 'why' behind the build, leading to a better, more focused product.
Beyond ideating on the "happy path," instruct your AI assistant to imagine and diagram the "worst path"—where everything breaks between users. This technique surfaces edge cases, user frustrations, and system vulnerabilities that are easily missed in early planning.
To compensate for the lack of a human engineering team, create a multi-step review process with different LLMs. Use one model (e.g., Claude) to build, a second (e.g., Codex) to challenge and find flaws, and a third (e.g., GLM) for a final pull request review.
As a project grows, a large context file (e.g., `claude.md`) can cause hallucinations and token waste. To solve this, create smaller, feature-specific context files and link to them from a master file. This keeps the active context lean, focused, and more accurate.
Traditional MVPs prioritize viability. Since AI drastically reduces coding cost, founders can afford to build past viability to "lovability." Amer built Clearlist until it fully solved his own emotional need—"not losing my mind"—creating a more complete and compelling initial product.
AI models have knowledge cutoffs and may recommend outdated libraries (e.g., Next.js v14 instead of v17). Proactively prompt the AI to check the latest API documentation and confirm it's using the most up-to-date versions to prevent significant, time-consuming refactoring later.
Instead of slow, manual research, use an LLM to analyze discussions on public forums like Reddit. Amer used Gemini to find the most common complaints about selling used goods, allowing him to prioritize product features based on the frequency and intensity of user-expressed pain points.
Go beyond the base AI model by incorporating specialized, pre-built skill "stacks." Amer uses Gary Tan's "G-Stack" for security and planning reviews and Matt Pocock's skills to refactor and improve codebase architecture, effectively leveraging expert knowledge packaged as reusable AI commands.
A non-engineer built a 6-agent system for his app Clearlist.me. Different agents handle distinct tasks like identifying items in photos, grouping them, researching local prices, and writing human-like listings. This demonstrates how complex, automated workflows can be orchestrated without deep engineering.