Are you coding more or less, managing people differently, or making decisions in new ways because of AI tools? Which tools (LLMs, copilots, internal agents, analytics, etc.) have meaningfully stuck, and which turned out to be hype? I’m especially interested in concrete changes to how you plan, review work, and support teams.
Before I had not enough time to gather context, be in the flow, code and test.
Now I work throughout the day, as soon as I have 10/15 minutes I send a prompt to one of my Claude Code so they can make progress on tasks that the teams cannot undertake because too time consuming versus business needs (major migration, architecture changes, etc…)
I've heard this a few times already on Twitter, but wouldn't your time be better spent in other areas? even if it is 10 or 15 mins? I am having a hard time understanding this sentiment from SEM.
If it's something that anyone could do in just 10 minutes, it could be another simple task for the ICs working with you.
You will need ICs to review the code, approve the PR, and this will go to the pipeline and eventually get released. So you won't avoid engineers having the context of this piece of work.
In my opinion, an EM should use these 10-15 minutes for any of the responsibilities that they have in their role.
I have never in 30 years across 10 jobs ever seen it work well when a manager does production code instead of just POCs.
Either you’re going to suck as a manager and not get your team the resources they need, play the political games, etc or you’re going to suck as a developer because you can’t keep your commitments and can’t do the follow through.
Even worse, is a manager who is pushing vibe coded slop.
The best thing a manager can do if they still want to code is R&D level work that developers productize.
I am a staff consultant. If I’m leading a large project, I purposefully don’t commit any code and spend most of my time coordinating between other developers and “the business”. If I do work it’s again an isolated POC.
I am in enterprise API management and gathering requirements from outside teams is the bottleneck, not writing the code. We have officially supported internal AI capabilities now and nobody is using it.
GPT + Grok (sometimes Claude) for writing docs, policies, requirements, client responses etc. Grok is often times more concise/direct, which helps me as I tend to be verbose. I always review/edit regardless. Much faster than writing from scratch, and combining responses on the same topic is sometimes best.
Copilot for code completion + reviews or small snippets/functions but larger code/module generation has been weak so far.
Claude for full modules generation or complex multi-file edits.
Research: Grok (less filtered + better search), Claude (complex dev topics), GPT (balanced but sometimes slanted and/or seems to favor certain sources).
My Teams: Mostly Copilot for code completion/reviews, mix of GPT/Claude for code. Last year was loose/experimental to learn but we plan to formalize guidelines more this quarter.
Definitely a ton of hype that doesn't always match reality, but it is a super powerful tool that really has made things move faster.
I have time to code up little tools for myself and my team. I can now automate simple things quickly to improve workflow, like gathering all the modified tickets in the past 24 hours and writing them out in our daily sprint document (it's the process they want so I go with it). Am I directly contributing to the critical path code, no, but I'm able to understand more quickly. I'm a Sr SDM and have been divorced from daily code for several year, now I'm able to get a broad view of things more quickly. Even just the act of finding dependent packages removes a little friction which means I'm less likely to let it become a road hump.
I run a small team building Dreamlux.ai (AI video generation). AI changed my day-to-day in a few concrete ways:
Coding: I write less boilerplate and more “integration glue”. LLMs are great for scaffolding (Next.js routes, workers, SQL migrations) and translating logs/errors into hypotheses. Biggest win is speed from idea → working PR, not perfect code.
Debugging/ops: I paste real production symptoms (headers, cache status, curl repros, traces) and ask for ranked root-cause candidates + experiments. This reduced “blind poking” a lot, especially around CDN caching rules, 429s, image optimization, and edge-case billing/credits.
Planning/reviews: We now require “AI-assisted PR reviews” to include: risk list, test plan, rollback steps, and what metrics should move. It’s basically a checklist generator + reviewer #2. Humans still make the call.
What stuck: ChatGPT + a copilot in editor for daily work; LLM as a “rubber duck” for incident triage and for turning messy notes into specs.
What felt like hype: autonomous agents that “own” features end-to-end. Without tight scopes, they wander; with tight scopes, they’re just faster scripts. The sweet spot is human-led, AI-accelerated loops.
Curious: for managers, what’s your best process change that AI enabled (not just “wrote code faster”)?
Before I had not enough time to gather context, be in the flow, code and test.
Now I work throughout the day, as soon as I have 10/15 minutes I send a prompt to one of my Claude Code so they can make progress on tasks that the teams cannot undertake because too time consuming versus business needs (major migration, architecture changes, etc…)
I love it to be able to contribute more
reply