More than just code review

The Catalyst: An Overlooked Paradigm Shift

Simon Willison's short post highlights a confusion many developers are experiencing but haven't yet articulated clearly: as AI coding assistants (like Cursor, Copilot, Devin, etc.) take on more coding work, what exactly is our core value and workflow as 'human developers'? We've been accustomed to viewing 'code review' as the cornerstone of quality assurance, but Simon presents a counter-intuitive idea: eyeballing every line of code may never have been the most effective method.

Deconstruction: The Evolution from 'Reviewer' to 'Commander'

Simon's core argument is that the key skill for productive collaboration with AI coding agents has shifted from 'reviewing code' to 'confidently instructing them and then confidently verifying the results'. This sounds simple, but its implications are profound.

You can think of it like managing a highly capable but explicitly guided junior engineer. Your value isn't in rewriting every line of their code, but in:

1. Clearly Defining the Task: You must accurately describe 'what to do' and 'why', not just a vague 'help me write a feature'.
2. Designing the Verification Plan: You need to think ahead about how to confirm the AI's output meets expectations through tests, logs, and execution results. This is far more efficient than post-hoc line-by-line inspection.
3. Understanding System Intent: You need to grasp the overall architecture and business logic to ensure the AI's local modifications don't break the whole.

Simon's point that 'eyeballing every line of code has never been the most effective way to validate a change' essentially means that even in traditional development, over-reliance on manual visual review is prone to missing logical errors. Now, the speed and volume of AI-generated code make this inefficient method completely obsolete.

Trend Insight: The 'Upgrading' of the Developer Role

This reveals a deeper trend: developers are transitioning from 'code producers' to 'system architects and quality commanders'. AI handles the concrete coding tasks, while human developers focus on higher-level abstractions: requirements analysis, system design, testing strategy, and final result verification.

This means future core developer competencies will include:

• Precision Prompt Engineering: How to use minimal instructions to get AI to produce code that best matches expectations.
• Strong Testing & Verification Mindset: How to design automated tests, monitoring, and verification processes to ensure the quality of AI output.
• Systems Thinking: How to break down large tasks into small modules suitable for AI execution and manage their interdependencies.

Practical Value: How Can You Adjust Your Workflow?

For developers currently using or planning to use AI coding assistants, Simon's perspective offers direct guidance:

1. Change Your Review Habits: Don't fall into the anxiety of 'must read every line of generated code'. Focus your energy on critical logic, boundary conditions, and integration points with existing systems.
2. Invest in Verification Infrastructure: Spend time writing more comprehensive unit and integration tests, and establish clear logging and monitoring. This is the foundation for trusting AI output.
3. Practice the 'Art of Instruction': When assigning tasks, try to provide more specific context, constraints, and examples. Treat the AI as a team member who needs a clear requirements document.
4. Embrace 'Black-Box Testing' Thinking: Sometimes, verifying that a function works is more important than understanding how every line works. First ensure 'input A yields output B', then dive into details only when necessary.

Counter-Intuitive/Unexpected: Trust but Verify, and Upgrade Your Verification

A point that might be overlooked is that Simon is not advocating for 'blindly trusting AI'. On the contrary, he is calling for a more advanced and effective form of 'verification'. In the AI era, 'verification' does not equal 'review'. Verification can mean running a test suite, observing system behavior in production, or checking if the final output meets business metrics. This requires us to move beyond the 'code is everything' mindset and ensure quality from a more macroscopic system perspective.

In summary, Simon Willison, in just a few sentences, outlines the evolution of the developer workflow in the age of AI programming. This isn't about whether to use AI, but about how to use it intelligently and redefine your own role within it.