Why Senior Engineers Get More Out of AI Than Junior Developers

A counterintuitive pattern is emerging in how developers use AI coding tools. One might expect junior developers to have the advantage given their comfort with generative AI. Instead, senior engineers consistently extract more value from these tools. The difference is not about prompt engineering or knowing the right tricks. It is something more fundamental about how software gets built.

The Process of Building Software

Whether using AI or not, building software follows the same fundamental process:

1. Define and understand the problem statement clearly
2. Figure out the solution: contextualize with the existing codebase, consider architecture, constraints, tradeoffs, and identify components
3. Break down components into units of work with clearly defined expected outputs from each unit
4. Execute on each unit of work

Coding agents can assist with each of these steps, but not autonomously. When working on existing large codebases, understanding context is critical. Agents can help navigate to specific parts of a codebase, but planning changes requires human judgment. Agents can generate multiple ideas and approaches, but selecting the best approach based on constraints and tradeoffs requires human judgment.

A key observation: most decisions made to arrive at solutions are not documented anywhere in the codebase. Engineers must understand those decisions on existing code and make new decisions for changes. Agents can generate options, but selecting the right option based on context requires human judgment.

Beyond Prompt Writing

Effective use of coding agents is not about writing a prompt and receiving code. It requires writing high-signal prompts that produce expected output, breaking down work into units that agents can assist with, and reviewing and verifying agent output.

These capabilities require experience and judgment that senior engineers have developed over years of working on complex codebases.

The Agency Problem

The level of agency is low with junior developers, and similarly with coding agents. Without agency, poor judgment is inevitable. When things go wrong, responsibility cannot be attributed to the agent. Developers are accountable for every PR that gets merged.

"You didn't take the time to write this code, but you expect me to take the time to make sure it's correct."

A senior engineer reviewing a PR generated by a junior developer using AI

When developers push PRs of meaningful complexity without understanding the code, they are likely to introduce bugs and technical debt.

The Verification Problem

The primary challenge with generated code is verification. The hardest problem is deciding what to trust and what to discard.

Our observations indicate that senior engineers accept more agent output than juniors. Several factors explain this:

• Higher-signal prompts: Senior engineers write prompts with tighter specifications and minimal ambiguity
• Better decomposition: They break down work into agent-compatible units
• Stronger priors: They have established mental models for correctness, making review faster and more accurate
• Verification heuristics: Junior developers generate output but lack the heuristics to confidently evaluate it

This evidence suggests that coding agents amplify existing engineering skill rather than replace it.

Supporting Research

Recent research from the University of Chicago on AI agents and software development productivity supports these observations. The study by Sarkar found:

• Experienced workers are more likely to develop plans with agents before acting, suggesting that generated output is more aligned with their intent
• Experienced developers have more codebase context, which improves alignment
• People with more experience and organizational context demonstrate higher skill in specifying the tasks they delegate

Implications

Senior engineers are better equipped to leverage AI tools effectively because they possess the skills and experience necessary to guide the AI, evaluate its output, and integrate it into their workflow. Junior developers can benefit from AI assistance, but they may struggle with verification and decision-making, which are essential for successful software development.

As AI continues to evolve, human judgment and expertise will remain critical for ensuring high-quality outcomes. AI coding tools function as amplifiers: they amplify whatever skills the user brings. For senior engineers with deep knowledge and refined judgment, they amplify productivity significantly. For junior developers still building foundations, the amplification effect is reduced.

The developers extracting the most value from AI are those who needed it least. The expertise that took years to develop is precisely what is required to use these tools effectively.

References

1. Sarkar, Suproteem K. "AI Agents, Productivity, and Higher-Order Thinking: Early Evidence From Software Development." University of Chicago, 2025. ↩