Working with AI feels more like leadership than coding
Working with large language models as coding assistants is prompting engineers to compare the work less to traditional programming and more to management: defining goals, delegating tasks, and reviewing output from a swarm of fast but unreliable “junior contractors.” Some argue this rewards people with project or people-management experience, while others insist that deep technical understanding and conventional engineering discipline still matter just as much, with AI best treated as a powerful but error‑prone tool. Underneath the debate are concerns about rising technical debt, over‑anthropomorphizing models, and how reliance on AI may reshape both software quality and career paths in engineering.
What AI-Coding “Feels Like”
- Many see AI-assisted coding as closer to management than hands-on coding: assigning tasks, monitoring progress, correcting course.
- Others argue it still feels like coding: same mental model, architecture, and low‑level concerns, just less typing.
- Some propose alternative metaphors: directing a film, orchestration, managing contractors, supervising interns, or dealing with a bureaucracy.
Leadership vs Management Debate
- Several commenters insist leadership is fundamentally about people (vision, motivation, mutual influence), so LLM work cannot be leadership.
- Others use “leadership” loosely to mean higher‑level direction: setting goals, breaking down tasks, coordinating multiple agents.
- A common compromise view: working with LLMs is like product/tech lead work with the social/emotional components stripped away.
Skills that Matter
- Prompting is repeatedly likened to:
- Requirements gathering and acceptance criteria.
- Delegation, quality gateways, and exception handling.
- Designing “the bowl” that constrains “water” (LLM output).
- Some claim people with management or process-design experience get better results; others report the opposite.
- Effective use requires maintaining a strong mental model, reviewing outputs, and treating LLMs as bounded-task tools, not fully autonomous coders.
Reliability, Slop, and Technical Debt
- Strong concerns about “vibecoding”: blindly accepting AI output, producing huge volumes of low-quality code and technical bankruptcy.
- Anecdotes describe leaders with no coding background trusting AI uncritically, generating tens of thousands of lines of broken code and derailing projects.
- Counterexamples show success when AI is treated like a fast but error-prone junior: small tasks, strict review, tests, and guardrails.
Impact on Work and Careers
- Some feel empowered: faster iteration, less drudge work, more focus on architecture and problem definition.
- Others mourn loss of hands-on coding joy, describing a shift toward permanent management‑like work they never wanted.
- Several note reduced hiring or frozen expansion of dev teams because LLMs cover more output.
- There is disagreement on whether LLMs “equalize” by lifting weaker devs and neutralizing top talent, or simply amplify existing skill differences.
Anthropomorphism and Limits
- Many warn against treating LLMs as people: no agency, no morale, no career stakes.
- Managing LLMs lacks the hard parts of real leadership: emotions, conflict, retention, and accountability for human outcomes.