A developer on DEV.to has published a thought-provoking experiment: what happens when you program your AI coding agent with the instincts that make senior engineers valuable? The author, Faizan Mohiuddin, argues that the most useful behaviors senior devs exhibit aren't technical skills at all—they're judgment calls, soft skills, and habits developed through years of shipping code and watching it blow up in production. The experiment centered on ten specific 'instincts' that senior developers exercise automatically: pushing back when asked for the wrong thing, refusing to say "done" until code has actually run, proactively trying to break their own work before shipping, and asking clarifying questions before diving into implementation. These aren't documented anywhere—they're tribal knowledge passed down through code reviews and post-mortems. Mohiuddin applied these ten instincts as behavioral constraints and prompts for an AI coding agent. The goal wasn't making the AI write better code syntactically—most models can already do that. The gap Mohiuddin targeted was judgment: knowing when NOT to implement something, when to question requirements, and when to slow down instead of generating another 500 lines.

What Senior Devs Know That Junior Devs Learn the Hard Way

The article highlights a fundamental truth about software engineering that's rarely articulated explicitly: technical competence is necessary but not sufficient. A developer can write clean code all day and still be a net negative if they lack the instincts to catch bad requirements, anticipate edge cases, and communicate when they're stuck or uncertain. Senior developers who've seen their fair share of production incidents develop what Mohiuddin calls 'pre-code' and 'post-code' behaviors—things that happen before fingers touch keyboard and after code is written. These include questioning scope creep, running manual tests instead of trusting automated ones, and flagging technical debt before it compounds into a crisis.

Why This Matters for AI in Development Workflows

As AI coding agents become more capable—handling boilerplate, scaffolding projects, even managing multi-file refactors—the question shifts from 'can AI write code?' to 'can AI write the RIGHT code?' Mohiuddin's experiment suggests that adding behavioral guardrails and judgment prompts might be the missing layer between a tool that generates code and a partner that makes developers better. The implications for development teams are significant. If AI agents can internalize senior-level instincts around scope, quality, and communication, junior developers get an always-available mentor guiding them away from costly mistakes. Senior developers get leverage—offloading routine decisions while they focus on architecture and strategy.

Key Takeaways

  • Junior devs write code; senior devs judge whether the code should exist at all
  • The most valuable engineering instincts are behavioral, not technical—and AI can learn them
  • Prompting for 'pre-code' habits like questioning requirements may reduce downstream rework

The Bottom Line

Mohiuddin's experiment is a reminder that shipping software is a human discipline first. If we're going to put AI agents in our codebases, maybe we should teach them how senior engineers think—not just what they type. The future of AI-assisted development might depend less on model capabilities and more on the wisdom we embed around them.