1. Embrace Ambitious Projects with AI
Actively seek out and tackle more ambitious projects, as AI tools significantly reduce the time and effort required for development, making previously daunting tasks feasible.
Leverage AI to enhance your existing expertise and learn new technologies quickly, turning potential skill atrophy into an opportunity for growth and broader capability.
3. Prioritize Rapid Prototyping
Utilize AI to quickly prototype multiple solutions for any design challenge, enabling faster experimentation and more informed decision-making on the best approach.
4. Implement Test-Driven Development (TDD)
Always ensure AI coding agents run automated tests on the code they generate, using “red/green TDD” to verify functionality, catch errors, and prevent regressions.
5. Hoard Knowledge and Experience
Systematically document past solutions, techniques, and AI-driven research outputs in reliable repositories like GitHub, creating a valuable backlog to solve new problems by combining previous learnings.
6. Start Projects with Good Templates
Begin all new projects with a minimal, well-structured code template that includes a single test and preferred formatting, as AI agents are highly effective at adopting and adhering to existing code patterns.
7. Manage Cognitive Load with Agents
Be mindful of mental exhaustion when running multiple AI agents in parallel, as the intensity of overseeing their work can quickly lead to burnout.
8. Aim for Higher Quality Software
Focus on using AI to produce software with fewer bugs, more features, and overall higher quality, rather than just prioritizing speed of development.
9. Differentiate Vibe Coding from Agentic Engineering
Use “vibe coding” for personal prototypes or exploration where bugs only affect you, but apply “agentic engineering” with professional practices and human review for production-ready code.
10. Explore “Dark Factory” Development
Investigate advanced “dark factory” patterns where AI agents handle code generation and continuous QA without direct human review, focusing human effort on high-level instruction and robust testing frameworks.
11. Use AI for Creative Brainstorming
Employ AI to generate a wide range of ideas, pushing beyond the obvious by asking for more iterations or combining disparate fields, to uncover novel and interesting directions.
12. Double-Check AI Research for Accuracy
Utilize AI for research and information gathering, but always manually verify facts, especially for published content, to prevent the spread of hallucinations.
13. Mitigate “Lethal Trifecta” Security Risks
Be aware of the “lethal trifecta” (private data access, malicious instructions, data exfiltration) in AI agent security, and prioritize cutting off data exfiltration to prevent critical breaches.
14. Limit AI Agent Capabilities
Assume AI agents exposed to external input are vulnerable to malicious instructions and limit their capabilities to minimize the potential damage or “blast radius” of any successful attack.
15. Incorporate Human-in-the-Loop for High-Risk Actions
Design AI systems to involve human approval only for high-risk activities, avoiding excessive prompts that lead to “click fatigue” and ineffective oversight.
16. Leverage Hosted AI Services for Risky Tasks
Perform risky tasks, such as browsing untrusted web pages, using hosted AI services on provider servers, and avoid exposing private data in these environments to minimize security risks.