Because of AI, I am reading more than ever.

I noticed it last week at work. I was building a tool for visualizing our account onboarding flow and the capabilities we provision for new accounts, and I caught myself sitting at my desk with my hands off the keyboard for a long stretch. Not stuck. Not stalling. Reading. On my screen were the planning docs, review docs, and audit docs for the feature. A year ago that was not how I spent my time.

The numbers tell it. Of my productive hours, meetings excluded, research takes up 10 to 15 percent of my time, far more than a year ago, simply because it is now infinitely accessible. Planning is another 30 to 40 percent. Code review is another 10 to 20 percent. The balance is working with AI agents to get the work done. Three of those four activities are reading.

None of it
can be delegated.

Here is the part that surprised me: none of that reading can be delegated. Research, plans, and reviews are not status updates. They are the inputs to the next stage of my workflow. If I skim a plan and let the work proceed, I have approved everything in it, including the parts I did not read. If I ask an AI to summarize any of them instead of reading them, the gaps I would have caught stay in, and the agents building from them inherit every one.

“If I skim a plan and let the work proceed, I have approved everything in it, including the parts I did not read.”

Code review was
already too late.

A colleague of mine uses a term for what is happening here: moving things to the left. He means moving your work and your thought processes to earlier stages in the feature development cycle. The scrutiny that used to happen in code review now belongs in planning and in grilling the plan before anything gets built. Code review was already too late. What moved left is the reading.

I recently built an idempotency framework for transfers and payment processing. Most of that project was not typing. It was reading best practices, designing the approach, planning it, and iterating on the design with interactive diagrams until I was confident in a low footprint, high quality solution. Only then did any code get written.

The takeaway.

The result is more reading. And I am convinced it is time well spent, because one unit of effort at that stage has a multiplicative effect on everything downstream.