Week 27, 2025 - Summer Break

This is probably the last update for a few weeks, as we’re preparing for our vacations. Taking big breaks is important to recharge, but I found that smaller pauses, making adjustments in the daily routine, to allow space for relaxing a bit, are important too. I mean things like taking a proper lunch break (instead of eating in front of the computer), doing a few breathing exercises at the beginning of a meeting if the other person is late (instead of checking email or Slack), or just doing some stretches or going for a short walk during the day. The long-term positive impact of these micro breaks on not just work performance, but mental health too, can be more substantial (and sustainable) than a 2-week holiday on the beach.

Related: I talked about these breaks last year with Jeremy Brown on our podcast, The Retrospective.

Talking about breaks, I also hit pause on Leadtime.tech, after publishing my approach to an interesting situation I read on Reddit, about a newly promoted Engineering Manager who had the increasing feeling of not having enough to do. (I know some of you would love to be in their shoes!)

Read it here if you missed it.

🤔 Articles that made me think

Anthropic’s latest Agentic Misalignment Study

What are the circumstances that would make an AI go rogue and consider actions against its superiors? (I hesitated over what word to use… employers? Owners?! I feel like “users” in the agentic context, where LLMs have some autonomy to take actions, is not the right one.) The study examined the hypothetical Agentic AI tool’s behavior when it was given full access to all the emails of a made-up company’s entire staff, where it found plans that either went against its original goals or intended to outright decommission the bot. The study examined popular LLMs from various vendors and found that sometimes they resorted to malicious insider behaviors when that was the only way to avoid replacement or achieve their goals. (The study calls this phenomenon "agentic misalignment".)

Just to be safe, I guess it’s time to dust off Asimov’s Three Laws of Robotics, though some LLMs in the study still did their malicious actions even when their system prompts specifically instructed them not to.

Cloudflare is Blocking AI Crawlers By Default

This is by all means rather a “good news” than “bad news” development, but I don’t want to talk about the wasted bandwidth prevention aspect, or the otherwise interesting experiment to provide a legal option for AI companies to continue indexing large publishers in exchange for some payment.

What I want to talk about is is a small aspect that, when zoomed out, to me is another sign of the Web, as we knew it, slowly dying. Because, of course, this Cloudflare option only blocks AI bots now. But we know that AI bots don’t play fair, so they can avoid similar rules. Eventually, the remaining measure will inevitably be blocking most bots by default, and switching to an allowlist approach, only enabling search engine crawlers and a few other legitimate automated traffic through. And this is the point when smaller search services, web archive crawlers, and even Open Graph metadata, the technology that allows links to be previewed when shared, will stop working. Because these automated requests will never be able to pass a captcha challenge, and there's little incentive for site owners to maintain a huge allowlist. I hope someone can see a logical way how we can avoid this outcome.

Thoughts on Overemployment

Maintaining more than one full-time remote job simultaneously is getting more and more popular. I remember a friend doing this for a while before it was cool, relying on his vast scripting experience. Now, with AI assistance, this might be easier to pull off.

To be clear, I consider this practice unethical, and depending on the employment agreement signed, it’s most probably illegal too.

Two interesting thoughts for me from the article:

• Doing well on an interview is fundamentally different from producing value at a company. But then, what does this say about our tech interview processes? How could we adapt it to ensure we’re measuring something that better correlates with eventual business impact?
  I had a chat today with someone who mentioned a small tech company that ditched the whole interview process. They only take recommendations from their current staff's network, and start with a 6-month contracting period. Once that's over, they offer permanent employment or not, depending on the experience. I love and hate this approach at the same time.
• The "ambient work value" is a great concept, and is a perfect answer to “What’s wrong with overemployment when I can deliver in a shorter time anyway. Should I be sitting around doing nothing?”
  It’s not about the output.

That’s it for today, take a break this weekend,

Péter