I hate the term “forward-deployed engineers” as much as I hate and hated the term “growth hacker,” back when that was the current fad.
It is funny, though, that we now use the term “engineer” instead of “hacker.” I think it suggests that we are in the era of confidence men. Today is all about the appearance of “I know what is going on”; the harsh truth is that no one knows what is going on, but the con-men claim to know.
i'm confused when i see people who don't like AI say stuff like "well eventually it'll be good enough to replace me"
and i don't know where that's coming from. sure the hypemen are claiming that, but where's the actual evidence? feels like AI has been plateauing pretty hard for the last couple years for me except maybe for in agentic code
but all the advances in agentic code are hacks built on top of the same shitty LLM model that's stopped improving ages ago and those will also have a ceiling
real. fuck your ai overview
sad day for all us ZZ Top fans. rest in peace, Frank Beard. clearly somebody who gave to the world more than he ever took 🌻
I’ve been working on this one for a few weeks. Let me know what you think!
https://ryancheley.com/2026/08/17/what-if-maintainer-burnout-isn-t-burnout/
Managers: We have a seven-round interview process to ensure we hire only the best of of the best. 10x programmers only. You will need to score in the top 1% on this IQ test. Don't even bother applying without a first-class degree and a portfolio of high-profile projects.
Also managers: All of your work must be produced using this thing that understands nothing and generates statistically average code
don't forget to turn off AI summaries if you update firefox mobile. I had the "AI killswitch" turned on and the latest update turned summaries on and turned the "killswitch" back off
About a year ago I heard of this woman from a friend. I'm glad she felt safe enough to write this.
It's too bad most of the staff writers at the atlantic have worked against her life for years.
RE: https://bsky.app/profile/did:plc:x4pnflc3hylxldr352grock2/post/3mtdwgv353c26
I used to think that queer acceptance was mostly about things like marriage equality
Now I think it’s about how Chinese aunties and uncles here don’t skip a beat or care about your sexual orientation when they’re being nosy about your life, and go straight into ‘okay but what does your wife do? How did you meet? Are you planning to have kids? I heard about this thing called IVF have you considered it? Lots of my friends have gay kids too they all got IVF’
Yet again hearing how some skilled work never truly mattered now that it's getting automated, while it absolutely did and was a point of professional pride for many.
It's a significant aspect of 'deskilling', which is a known consequence of automation. Pretending otherwise is needlessly unkind.
Whenever a younger person learns how paranoid and worried I am about having my account or device compromised, they treat me like I'm insane. The question I get is "You can just get it all back, right?"
To them, apparently it's just a reality of life? Something they expect? And they don't seem to think there are negative consequences for it?
I worry.
I worry a lot.
This is a fascinating study.
I was on the briefing call yesterday and, y'all -- people (especially older women) HATE AI with the personal kind of passion one reserves for the villain who killed your father.
They HATE tech execs. Like, HAAAAATE.
And this sentiment does not change when folks 'get used to' AI. In fact, distrust has increased despite everyone using it.
They absolutely distrust AI boosters. No one believes the hype and an overwhelming majority believe this shit is bad -- because of personal experience of harm or they know someone who has been harmed.
People want AI *punished* they hate it so much. (This is me editorializing.)
And they want someone to wrap some chains around AI before it's too late. #RegulateAI
Duuuuude. Faaascinating.
Full report drops next week.
https://drive.google.com/file/d/1lKvEjvDK_zwIvGDdCdL8d74ov5eVFqkm/view?pli=1
What did I say about not getting into details?... There's the joke that the great thing about standards is there are so many to choose from. In this case, code users may be variously motivated by ISO9001 or CMMI or in our case NQA-1 which does its level best to treat software as a plant component like a pump or valve or spool of cable regardless of how much software resembles none of those things. Never underestimate the tenuous logic of a committee of mechanical engineers, I guess. Regardless, if you have to chart a course through overlapping or conflicting standards, it's reasonable to start with a set that is a close philosophical match for your organization and try to logically pick and choose the best aspects of the other standards, then distill that down to something you can understand, explain, and implement. Again, you know you won't satisfy everyone but you're not just making it up as you go along and you have a consistent high-level narrative of what problems you're trying to address. You make it clear you did your research and are taking the problem seriously; people can argue over the details but you can show you address the core issues in a systematic way and there's a path from procedures back to the plans they implement. There's some logical consistency to it and it's up to the stakeholders (in this case us) to judge whether the differences and details are significant enough to argue about. Most of these standards are just different ways of getting to the same place and maybe more important than the details is that staff have a process they understand and can follow and they can demonatrate that they adhere to. This stuff is picky and tedious but ultimately is meant to keep people out of trouble and to avoid known and avoidable problems. What we are getting out of this is more exposure to different risk philosophies - we're required to meet NQA-1 but not everyone is and we know that spec and Procurement-centic approach is not a great fit for safety analysis software (software FMEA? Are you mad bro?...)
As we walk through their process, looking at a few specific error reports and feature requests we can see reasonable consistent practice that makes sense and is familiar. There's a strong emphasis on configuration management so systems are known, processes are repeatable, and you can reason about the system state. Testing is comprehensive and improving and there's clear indication the code is becoming more robust with every release and that trends ard being watched. Here we're looking at a mature code that's been under active development since the early 80s. Parts of the code predate QA requirements so there are known issues where certain parts of the code are more fragile than others and much more care is taken when modifying them to avoid breaking some of the less robust code. It's not great but it's the nature of a large code that evolved over a long period; there's no easy solution and there are always resource constraints. But what we see is that risks are understood and managed and practice follows philosophy and the QA regime is a tedious net positive. People know what they're doing and why and there's enough self reflection and peer review to keep everyone honest. All things considered they're doing about as well as could be expected. We might be able to see some gaps but we also see there are constraints we don't have that. In the main what they are doing makes sense and is reasonable even if it's not exactly what we'd do. Mostly it's a different path to about the same place and their practice is not too different from our own. Like I said, our goal in this visit is to understand not to verify compliance and we're waking away with the sense that they know what they're doing, their practice is mostly consistent and makes sense.
This article starts with a story from someone who tried Claude Code and found it amazing, but then switches to the same person a few months later seeing what a disaster it’s been. This quote is key:
The problem is that code produced by an AI agent looks reasonable, but can contain ‘hard-to-spot bugs’ that end up causing major problems
LLMs, by their nature, generate statistically plausible output. That is often a set that overlaps with correct output. But the things that are not correct look exactly the same as the ones they are.
Learning to review code is hard. You look for the bugs that you expect to be possible implementing it by using your theory of mind for the person writing the code and the kinds of things that they might overlook (not necessarily a specific person, but the kinds of things people miss) and also common bug classes.
And the big help is that the person writing the code is not thinking adversarially. They are not trying to sneak bugs in. Normally. Unless they’re a supply-chain attacker, and we’ve a depressing amount of evidence that code review doesn’t catch supply-chain attacks.
An LLM is not trying to do anything. It has no intent. But it is a machine that is trained on code that made it past code review. The kind of bugs that it will generate are ones that look like code that appeared in production. This is exactly what an attacker would do: try to write code that looks correct but is subtly wrong.
I’m only being slightly flippant when I say LLMs are a mechanism for bringing supply chain attacks in house.
this guy has never had to help an elderly person who got got by a full-page Chrome takeover "ad" that told them they had a virus and needed to call a phone number to fix their computer
RE: https://bsky.app/profile/did:plc:swgxwbir4rrimj57pf75jckf/post/3msruy5im2k2d
the scarcest resource is give-a-shit. even something perfect will collapse if there's not enough give-a-shit to go around. the only reason anything works is because there's people in the background invisibly doing thankless labour to keep everything working. it's so hard to explain this point. I've heard people say it before I became a community organizer and was like "I guess that makes sense" but now that I've got all this experience it REALLY makes sense. I'm not sure how I'd explain this experience-backed understanding to myself from three years ago. probably impossible.
in which the author of xfwl4, XFCE's Wayland implementation, tells us exactly how much his principles are worth https://www.spurint.org/journal/2026/07/llms-and-xfwl4