The Bootstrapped Founder

420: AI for the Code-Writing Purist: How to Use AI Without Surrendering Your Keyboard

October 24, 2025

Key Takeaways Copied to clipboard!

  • AI systems can be leveraged as a coding companion for investigation, research, and code review without writing production code, appealing even to code purists. 
  • AI excels at focused tasks like debugging by analyzing code snippets, tracing logic, and suggesting optimizations (e.g., SQL query tuning) when provided with precise context like execution plans and schemas. 
  • Thorough documentation and high test coverage significantly enhance the reliability and value of a codebase when using AI agents for review or development, making the business more attractive for acquisition. 

Segments

AI as Non-Code Companion
Copied to clipboard!
(00:01:24)
  • Key Takeaway: AI can be leveraged as a coding companion for investigation without writing production code.
  • Summary: AI systems can function as investigators, tasked with research to report back insights that developers can use to tackle projects on their own timeline. By explicitly forbidding code generation, the AI acts as a pair programming partner or code reviewer providing opinions and guidance. This method was effective even when AI code generation was unreliable, focusing instead on problem-solving insights.
Debugging with AI Investigation
Copied to clipboard!
(00:04:08)
  • Key Takeaway: Effective AI debugging requires providing specific input/output data and instructing the AI to maintain a detached, non-sycophantic perspective.
  • Summary: Developers should task AI with investigating specific bugs by providing the function, input data, erroneous output, and suspected code modules. LLMs suffer from ‘sycophantic exuberance,’ often claiming code is perfect, necessitating system prompts that enforce a professional, detached, or even antagonistic review stance. For complex bugs, providing precise input/output data is crucial, as static analysis tools may be better for highly typed, compiled languages.
Optimizing Cross-Language Queries
Copied to clipboard!
(00:08:08)
  • Key Takeaway: AI is highly capable of optimizing embedded SQL queries by analyzing execution plans and schema details.
  • Summary: AI can optimize slow or inefficient ORM queries or embedded SQL by taking the existing query, its execution time, and the output of an EXPLAIN ANALYZE command. By feeding this data along with the relevant database schema to the AI, it frequently suggests immediate optimizations like correcting index order or suggesting new indices. This optimization process suggests structural improvements rather than writing application code, maintaining developer control.
AI Security Review Capabilities
Copied to clipboard!
(00:10:25)
  • Key Takeaway: Agentic AI systems provide valuable security reviews, catching integration and configuration vulnerabilities missed by traditional testing.
  • Summary: Tools like Claude Code offer specific security review features that check recent changes for security implications based on established community best practices. By embedding security preferences into a system prompt, developers can automate checks for adherence to security standards before merging commits or deploying. This capability is especially valuable for solopreneurs who lack a dedicated security team, preventing issues like forgotten test values that could lead to system saturation.
Enhancing Codebase Value via AI Review
Copied to clipboard!
(00:14:42)
  • Key Takeaway: Utilizing AI for comprehensive code review (security, documentation, testing) increases business value for potential acquisition.
  • Summary: For solo founders considering selling their business, an acquisition-ready codebase requires high test coverage and excellent documentation to cover comprehension debt. AI agents can be tasked with suggesting tests for highest coverage potential or generating documentation based on existing high-quality examples within the codebase. This process, which involves the developer reviewing the AI’s output (e.g., documentation comments), makes the code base demonstrably more valuable to acquirers.
Configuring AI for Non-Code Assistance
Copied to clipboard!
(00:19:48)
  • Key Takeaway: System prompts in tools like Claude Code can strictly limit AI output to analysis, documentation, and testing, preventing production code generation.
  • Summary: Developers should use configuration files, like Claude’s .claud.md file, to set system prompts that forbid writing production code, restricting output to analysis, suggestions in a separate folder, or writing tests. This allows the AI to deeply understand the internal logic of the code base without altering the core application files. Augmenting capabilities through investigation and optimization maintains the developer’s craft while making the code better and the business more valuable.