Post

If you, like me, are in a situation where writing code took forever, then you replace it with Claude/Codex/et al and it'll be more productive BUT you also will (or should, IMO) review the code, in small slices. So the time savings is: ORIG_CODING - (AI_CODING + REVIEW_TIME)

0 reposts0 likes

It takes an iterative approach, so like as I'm working on a replatforming epic, I broke it into phases, and each phase has slices of work that do some small change. It stacks the PRs so I can review a full phase at once, but still keeps PRs manageable to review.

The actual time savings for me actually comes from being able to review and get work done remotely via my phone. The only time I could code before was at the computer. Now, Claude cloud or local session can code, and I can control it and review async. THAT is my personal time savings.

There's a great podcast on evidence-based science in medicine and health/wellness by Dr. Bobby Dubois. He has a saying, that you can only evaluate based on N of 1. Meaning, you have to experiment and base your decision of what works/what doesn't by looking at YOUR data, not just general advice.

I think N of 1 applies to any AI workflows/tools as well. Depending on where you work, how you work, what constraints you work under, AI tools could help you save time or they will make it worse/slow you down. In my case, I can easily see that it is helping me while still maintaining good quality.