Abhi Dangeti

Here's what I get from it: - big tech companies are literally burning billions on "A.I." without any clear ROI plan - organic "A.I." adoption is abysmally low by their own employees because it simply doesn't work as promised, which is hypocritically embarassing for them, so now they're desperately trying to mandate its usage to prop up their reputational bubble a bit longer - CEOs still don't understand how their own sausage gets made and still don't care about producing quality products that will delight the true customers (the end users) - we have passed "peak A.I." (as all the big tech companies are starting to panic at the realization they have vastly overextended themselves by vastly overinvesting in "A.I.") and are now beginning to coast down the other side of the roller coaster track

5

Every PMM team needs to do this now! Get access to the source code repo. Here is why: 1. Ask questions in Claude or Cursor 2. Uncover how the product works 3. Answer technical and high level questions quick 4. Format outputs by audience 5. Even draw flow diagrams Now there is never a new question you can’t answer in minutes. No need to ask a PM. No need to wait.

12

LLMs produce a lot. Very fast. But they have no vision. No accountability. No judgment. They don’t care why the system exists. They don’t care if the code will break production. They don’t carry the consequences. What they really do is multiply something. In my experience, they multiply the quality of an engineer’s system thinking. If the thinking is clear, structured, and explicit, AI amplifies it. If it’s fuzzy, AI amplifies that too. That led us to a simple conclusion: we need to make system thinking visible. In tickets. In pull requests. In reviews. So we built tooling that gathers tickets, PRs, and diffs, and organizes them across dimensions like: - system awareness - design - testing - correctness - code quality - review quality - field impact - pace - planning & communication The LLM generated a huge amount of data. Markdown files, comments, even grades. Five minutes later, it was a full website: team pages, individual pages, cross-team views. It was seductive. Especially the scores. The model could assign numbers to everything: engineers, teams, rankings. As a leader, that’s tempting. But those scores quickly fell into a triangle of: useless, toxic, and inaccurate. What turned out to be genuinely valuable wasn’t the ranking. It was the qualitative signal. Clear patterns in design decisions. Testing habits. Review depth. System awareness gaps. Cultural strengths. That data now feeds daily and weekly conversations. AI didn’t solve engineering quality. It helped us see it. The next step isn’t better scoring. It’s refining our standards — and reinforcing them deliberately.

10

7 Comments