✓ Humans still do this better
System architecture and technical design
Choosing the right architecture for a specific business context — factoring in team skills, existing systems, and long-term maintainability — requires engineering judgment AI cannot replicate.
Debugging complex, context-dependent failures
Diagnosing failures that span distributed systems, infrastructure state, and business logic requires reasoning across large, partially observable systems that AI handles poorly beyond short context windows.
Requirements gathering and stakeholder communication
Translating ambiguous business needs into technical specs — and managing expectations across product, design, and leadership — is inherently human work.
Security review and adversarial thinking
Identifying attack surfaces in novel code, thinking like a bad actor, and making risk trade-offs requires strategic reasoning that current AI applies inconsistently.
⚡ AI does this now
Already automated
Boilerplate and scaffolding code generation
Tools: GitHub Copilot, Cursor, Tabnine
Already automated
Unit test generation
Tools: CodiumAI, Copilot, Claude
Already automated
Code review and linting suggestions
Tools: Copilot PR review, CodeRabbit, Codeium
Already automated
Documentation and inline comment generation
Tools: Mintlify, Copilot, GPT-4
Already automated
SQL query generation and data transformation
Tools: Text-to-SQL (Claude, GPT-4), Seek AI
Rapidly automating
Simple feature implementation from clear specs
Tools: Devin, Cursor Agent, OpenHands