Peter Mbanugo’s post, dated May 24, 2025, opens by calling programming with LLMs “both promising and frustrating”: the assistants can help with coding and debugging but often waste time too, yet for senior engineers he sees real potential in what he calls “peer programming” with LLMs, striking through the word “pair”. The rest of the post is a curated collection of blog posts by other senior or staff+ engineers about using LLMs in their work, which he presents as free of “the usual hype or buzzwords”.
Each entry is a short summary of someone else’s post plus Mbanugo’s own reaction to it. The post makes no empirical claims of its own; its contribution is the selection and the author’s first-hand remarks about which techniques he has tried.
Key Points
- Sean Goedecke’s post on practical AI techniques is summarized as describing two main ways he uses AI: a “second opinion” technique and a “throwaway debugging scripts” technique. Mbanugo reports having used the “second opinion” technique himself in recent weeks and wanting to try the second one.
- Harper Reed’s LLM codegen workflow is summarized as: brainstorm a spec, co-plan a plan, then execute using LLM codegen, and repeat. Mbanugo likes the brainstorm-and-plan idea but reports that he has not succeeded with it for a complete feature or prototype, and that it can turn into many hours of chatting with an LLM — though it has sometimes helped him abandon a project early once he saw how much it would take.
- Lee Boonstra’s post on documenting prompts is described as the only post Mbanugo had seen on that topic; he says he already saves prompts to review which ones worked best, but without structure.
- Seth Godin’s “ChatGPT is dumber than it looks” is summarized as a reminder that LLMs are not as smart as many think, with the advice to create patterns and processes in which the LLM is used as a tool.
- The post closes by quoting an excerpt from a later Seth Godin post: “When you get stuck, first ask Claude, then ask a human.”
Context
The post is a personal roundup rather than an argument, and every technique it lists belongs to the author of the linked post rather than to Mbanugo. Its framing — experienced engineers treating the LLM as a peer they consult, not a replacement for their own judgment — sits within the broader practice of AI-Assisted Programming.