AI May Be Your Neediest Reader of Code
Why coding agents may need more human-readable context, not less.
Index / Published writing
A working collection of essays, technical notes, and field investigations. Organized by subject, but allowed to develop a few eccentric cross-connections.
Subject / date / reading time
Why coding agents may need more human-readable context, not less.
If professional judgment is valuable enough to automate, it is valuable enough to keep developing.
Sometimes a prediction about the future is a confession about what someone never valued in the present.
What happens when an organization stops treating a metric as an approximation and starts treating it as reality.
Delegation is useful; surrendering the understanding that makes judgment possible is not.
What Queen Amidala and the Force can teach us about maintaining skill in the age of AI.
Cognitive Resource Management in the age of AI.
Why state-dependent data changes are not merely another delivery pipeline.
Good engineering depends on caring more about reality than ego.
An incident review demonstrates that engineers understand what happened.
Mature systems contain accumulated adaptations that a feature inventory cannot capture.
AI did not abolish coupling, code rot, hidden assumptions, or the accumulated history of software.
Professional judgment gets sharper when informed product and engineering partners challenge constraints together.
Mindset is most useful when treated as a practical thinking tool.
Useful ideas lose their power when organizations turn them into vague slogans instead of concrete guidance.
Healthy project management changes the plan when engineering work reveals new constraints.
Engineering dashboards show artifacts of work, not the work itself—and the distinction matters.
How one small interface change transformed reading, learning, and an understanding of accessibility.
A good interface turns technical possibility into capability a person can actually use.
An interface communicates how well a company understands and respects its users.
AI can help build the thing without becoming the thing.
Some feedback exists to improve your work. Some exists to make your competence less uncomfortable.
Small systems for getting reoriented when cognitive overload or task initiation brings work to a halt.
Responsible engineering sometimes means protecting the team and architecture by saying no, not like that.
What repeated advice about collegiality sounds like after decades of neurodivergent adaptation and professional practice.
The difference between asking an engineer for information and asking them to bless a decision you have already made.
Not every unusual working method needs to make sense to an observer in order to be useful.
Before blaming the tool, make sure the thing it needs actually exists.
Old systems contain unclaimed inheritance that has not stopped mattering simply because it is not new.
AI can accelerate good engineering habits, but it can streamline bad ones just as efficiently.
The better prize is creating a culture where excellence becomes accessible instead of exceptional.
The cleaner the path through a codebase, the easier it is to overlook the work that cleared it.
Accessibility cannot be outsourced to a widget, overlay, or magical compliance button.
The same principles that make interfaces understandable can make a codebase easier for people to navigate and use.
Keyboard focus indicators are navigation tools, not a place to demonstrate subtle compliance.
Teams that imagine they have exhausted the available intelligence may only have exhausted their own frame of reference.