In Chris Loyd’s recent post, he mentioned using a “ladder of ambiguity” to frame software engineer career progression in the context of comfort with ambiguity.

  • Junior engineer - Clear problem, clear solution, clear implementation
  • Mid-level engineer - Clear problem, clear solution, ambiguous implementation
  • Senior engineer - Clear problem, ambiguous solution, ambiguous implementation
  • Staff engineer - Ambiguous problem, ambiguous solution, ambiguous implementation

Your capability to efficiently deliver value in increasingly ambiguous situations - up to the point where even identifying the problem to be solved is unclear - is a primary measure of your career progression, and an important axis of growth.

I find this framing quite useful and I’d like to expand on it by discussing the mental models that can help engineers navigate this ambiguity using AI. Let me walk through each level of the ladder and explore how AI can serve different roles depending on the type of ambiguity you’re facing.

At the staff level, where problem ambiguity dominates, I find AI most useful as a research assistant and writing collaborator. I’m a big believer in the power of writing to clarify thinking. By writing down my understanding of the problem and asking AI to critique it, I can identify gaps in my understanding and refine my mental model of the problem. I also use AI to research the problem space—asking it to find relevant articles, papers, or code examples that provide additional context and insights.

Moving down to the senior level, where solution ambiguity becomes the main challenge, I find AI most helpful as a brainstorming partner and feasibility study expert. I use models like GPT-5 Pro to generate multiple solution ideas and evaluate their pros and cons. I also ask AI to act as a critic to challenge particular solutions and identifying potential pitfalls early on. When I have access to the codebase, I ask AI to provide feasibility-specific feedback on proposed solutions. This helps me quickly iterate on solution ideas and refine them based on feedback.

Finally, at the junior to mid-level, where implementation ambiguity is the primary concern, I treat AI tools as rapid prototyping partners to materialise different implementation ideas. Instead of waiting for one carefully crafted answer, I can quickly generate multiple implementations, compare them, and choose the one that best fits the context. I can also pick one implementation and ask AI to modify it to better suit my needs.

The key idea here is that AI’s role should evolve with the level of ambiguity. By deliberately choosing and matching the right mental model to your current challenge, AI can help you navigate each rung of the ambiguity ladder more effectively.