The Future of Software Engineering
AI for software development tasks has gone from research to daily practice. If we stop caring about code, the rigour needs to:
- Specification and precisely defining what should be built
- Verification to ensure that its correct, secure, and meets standards
- Governance and accountability
Note
AI handles the accidental complexity of writing code, but the essential complexity of specifying what needs to be built remains.
Supervisory engineering is emerging as a middle loop sitting between the inner loop of writing code (automated by AI) and the outer loop of CI/CD. We need to:
- Direct AI agent towards the right goals
- Evaluate and validate agent output
- Calibrate trust
- Encode standards and constraints
- Define safe operating boundaries
| People | Process | Technology |
|---|---|---|
| - AI tools increase cognitive load - Engineering roles shift towards management - Decision fatigue is a real concern | - Tier software by criticality and focus human review where it matters most - Strong CI/CD guardrails are even more important - Agent topologies define how agents fit into team structures and workflows | - Safe constrained technologies may be better suited for AI generated code - Shift toward verification over inspecting code - Self healing systems informed by incident history |