Vaya hackeo

If Slack is the office, why do you need email?

He had a draft tweet ready: "Today I resigned from OpenAI as an engineer... AI will be our doom" But it didn't work out

What used to be best software practices remain in the AI era: -> Clean interfaces between services -> Clear boundaries -> Well-documented systems -> Rich code examples All of this amplifies AI effectiveness across all teams.

I’ve been resolving git conflicts and updating branches on my own, it feels good.

AI speed is only a win if quality holds. Here are some metrics your organization must be checking by now We all know AI writes code faster than any human; we don’t argue that. The thing is, as we dump more work into frontier models without the right metrics, we're setting a time bomb for your organization. Organizations that are thriving in AI adoption were measuring from the early stages. What I’ve seen across our organization is a pretty well-defined process (all orgs should have) for how to actively measure the usage of AI without going blindly. These are the metrics you should be paying attention to urgently: -> bug rates: you can rate them by severity; you can set a starting point where AI coding assistance began; this catches degradation in quality early -> code review rejection rates: if you have organizations orchestrators already, you should be proving feeback (approve/reject) and track those -> production incidents: this will be your source of truth for stability -> time to resolve issues (SLA): are you taking more time to solve issues (are harder to solve), complex code that nobody understand -> tech debt: are you creating more work items and sending them to the tech debt epic? -> batch size (PR size): you can implement some sort of labeling ( we did that) These metrics are important for keeping engineering discipline with high quality even when AI output increases faster. Do you have any other one we should all be keeping an eye on?

“To everyone leaving: thank you” Wow that’s really cold.

Meta has made it clear that they're aiming to be a personal assistant #1 model.

You should think if you want to increase the /effort in Claude sessions. I get it, you are hitting yourself against the wall with a production issue, a really hard bug- you name it. It's likely you are not providing good context or examples to look at, even the connectors to use; you are just throwing prompts like a maniac. The model is giving you awful outputs, and your first thought is "increase effort level". In the end, this is the chain: more reasoning -> more inference cost -> model goes deeper -> -> more token consumption

Fridges in French supermarkets hit by cyberattack You don't need a big electromagnetic pulse to shut down an entire civilization. As we get into more automation in every place we go, a well-targeted cyberattack could be enough.

Turns out that Codex CLI installs a persistent, self-updating background daemon by default; there is a fix for this already. github.com/openai/codex/issu…

Honestly, I don't understand why people say they now spend 8hrs day hitting enter (accepting AI edits) In the pre-AI era, we already knew that coding was not 100% of our daily shift; it was only 20%. And even with AI, that time is now lower, so you get that time back to you to focus on the other 80% you had previously. The truth is that we spend more time in: -> Solving bugs (we see a big time saving in triage with AI) -> Architecting new features/tech debt -> Customer escalations -> Code review -> Meetings and alignment with other teams So perhaps you are not taking more ownership of your team/product than you could do; meaningful work/life is through taking responsibility.

AI came here to expose what was already there; if you had bad practices, that is what was being created.

Google no longer needs a LeetCode-style hiring process. This is how organizations are reshaping their hiring process. We all know hiring processes are mostly broken for companies that don't understand how to hire in this AI era, but at least some of them are copying Big Tech, and I hope we see better results by the end of this quarter. Be careful, though; don't expect the interview to be "easier" now. It will be hard, cause AI has set the bar higher now.

I recall that some months ago, AI usage through a coding assistant was optional; now it is a policy by default across mature organizations. Now this is being measured internally; even in your PMC, you are asked about the usage and leverage you've taken from these tools. As this post mentions, you can "object to using AI, but you'll be replaced" The entire world has changed, and we must adapt.

Top Time - Saving AI Use Cases When you think about what should be the top 1 use case for time saving developers would benefit the most, is Code Generation, but that’s not true. Continue the reading and you’ll find the rank number. #1 Stack Trace Analysis Eliminates 30-45 min of manual parsing; AI identifies likely causes immediately. #2 Refactoring Code Clear before / after pattern; AI applies transformations consistently #3 Code Generation Faster initial output, but review time offsets some gains. #4 Documentation Generate boilerplate and explanatory text from code context. #5 Test Creation Pattern-based generation with clear verification criteria. Being capable of triage and solve really fast production issues is where developers find most time saving, this is good cause customer experience is first, instead of shipping more and more like a manufacturing company. Source: Developer-Reported Impact