Kai Meyer (@KaiMeyer) header
Kai Meyer (@KaiMeyer) profile picture

Father C/C++ Software Engineer in Data Protection Linux Gamer Dallas "This is our year" Cowboys fan

Joined August 2026
Photo by Kai Meyer (@KaiMeyer)Photo by Kai Meyer (@KaiMeyer)Photo by Kai Meyer (@KaiMeyer)Photo by Kai Meyer (@KaiMeyer)Photo by Kai Meyer (@KaiMeyer)Photo by Kai Meyer (@KaiMeyer)Photo by Kai Meyer (@KaiMeyer)Photo by Kai Meyer (@KaiMeyer)Photo by Kai Meyer (@KaiMeyer)Photo by Kai Meyer (@KaiMeyer)
Kai Meyer retweeted
everyone coding with agents should internalize this - the RLVR in current frontier LLMs' post-training is done at task level. in other words, they won't give a damn about long term maintainability this is why we see so many useless comments and test cases being written by default. these things make future tasks less efficient, but they don't affect the current task's success, so they survived the training and became part of the models' behaviors until post-training can reward long term maintainability (which i suspect is very hard because inherently it's a slow and expensive process to measure that), it's on us humans to correct it in the harness layer through prompts, skills, lint rules etc this also implies that understanding long term implications of technical choices is a valuable human skill that's hard for an average person to make up for with LLM it's the distinction of whether you can create a durable growing codebase, or have one that gradually degrades into insanity if you have that skill, you are golden (for now, until post-training catches up)
84
58
16
927
36,959
Everything new is old.
so many folks building agent sandboxes only to rediscover Linux primitives
5
Kai Meyer retweeted
4 years into 6 months until AI takes your job
40
26
6
728
34,777
There are real experts out there, and Claude isn't one of them.
You're God damn right my hypothesis was correct, Claude. 30 fecking years doing this shit. I have underwear older that you.
44
Kai Meyer retweeted
You're God damn right my hypothesis was correct, Claude. 30 fecking years doing this shit. I have underwear older that you.
6
4
2
86
11,547
This thread had me thinking about where I may have learned to strive to put quality in front of speed, and I think I found it. Story time! "Learn how to do it right, then learn how to do it right *fast*." I received this nugget of wisdom from my manager when I worked at Wendy's as a high school student. He was ferociously competitive, which made us fast friends. If you ordered Wendy's in Overland Park Kansas after watching The Matrix in theaters, chances are you received your order in half the time when compared to just the year before. Our lunch rush could serve twice as many people as it could before, making it possible to almost double the revenue of his location. But it would only work if customers returned, demonstrating that there was no drop in quality. He required everyone to learn every station, using off peak hours for cross training. He recognized that bottle necks happen almost randomly and would have people shift left or right in the pipeline based on need. That required everyone to have uniform, and correct, output. That meant learning how to wrap a completed sandwich correctly, so when you handed it off, the next person knew exactly what to expect. No time spent wondering how to pick up a loosely wrapped food item. But the customer would be upset if the food was smashed. Quality first, then speed. His methodology caught on and other stores joined in. The KCK area drive thru goal was changed from 180s to 90s. Our location's personal lunch time drive thru goal was 60s.
Replying to @KaiMeyer
> I don't see qualitative demonstrations being championed by pro-AI voices, I see quantitative ones. Well, I at least try to do my part and show "number of incidents went down" when I show "commits went up" and when I say "99% made with AI" I also show the feedback people give us on our UX and how much they love it.
39
Kai Meyer retweeted
Without IT professionals, the world would be in a much worse place. So much of our critical systems rest on the backs of dedicated, hard working and selfless people. IT admins, network engineers, help desk, DBAs, and so many other roles help keep this duct-taped ship afloat. IT pros are my heros.
75
165
22
1,176
49,784
Linux kernel developers create guidelines targeting irresponsible developers using AI. There, I fixed your headline.
Linux kernel developers want to block AI-generated junk patches. SUSE developer Qu Wenruo proposed an AGENTS.md file with rules for AI agents and newcomers submitting patches to Linux’s Btrfs filesystem. He says some contributors are “doing more harm than help,” sending low-value fixes buried in long explanations and useless test details. His message to AI agents is blunt. “Stop being stupid.” The proposed rules call for shorter explanations, useful comments, and clearer bug descriptions. The file is still open for review. Source: lore.kernel.org/lkml/df1c1fa…
34
This is a pair of strawmen personified. Both sides of this rhetoric exchange are cringe worthy. The soft condescension from one side and the unreasonable anger from the other side. This doesn't help anyone. It serves only to continue to drive the wedge, not heal broke relationships. Here's the reality. Language models are amazing banks of encoded knowledge. Context windows are fickle. Too small and you get hallucinations. Too big and you drown important data in the menutia. Tools add determinism to the output. Selecting tools is a skill, one that still relies heavily on the user. There doesn't seem to be a limit to the tools we can write and select from, and this is the sort of unbounded growth that's good for tech. Harnesses use the LLM to help select the tools, loop the context window through tools, and produce extraordinary results. AI is a tool, it requires skill to learn, and most people are willing to learn new skills in order to make a buck. Volume of code alone is not a good metric for quality or productivity. Code with out an owner is a recipe for disaster. The volume of code produced with out ownership is experiencing a meteoric rise. CVEs are a good thing, and more CVEs means more of a good thing. A careful study of the most recent CVE run should increase your trust in "hand written code" not decrease it. The contortions required to exploit many of the recent CVEs are virtually impossible, even for dedicated AI. Remember PHP2Shell? Still zero confirmed cases of an exploit in the wild, despite having a working proof of concept. Drive by AI bug reports are worse than unowned code. Come on people, at least offer a fix. And none of that requires embodying either of the totems in this video.
For developers (and anyone really) who feels like AI is taking away their identity. It's the end of something, that's true, but it's also the start of something else.
1
98
Nothing about AI coding is new. I've inherited a ton of source code over the years, code I didn't write that I'm supposed to extend, update, modernize, add Linux support, and answer the phone when there's a bug. The only difference with agentic coding is that I'm the first one to look at this novel code. It hasn't been in production for 10 year. When I read it, it's just more of the same. Weird things I wouldn't have done, amazing things I wouldn't have done, working things I wouldn't have done, and every now and then it's exactly what I would have done. That last one scares me more than it should. But it's all the same in the end. Code I need to keep modifying and support, even though I didn't type it out.
25
Everyone seems to have forgotten that code is a liability, not an asset. That means that producing thousands of lines of code you didn’t read isn’t something to celebrate. It’s something to worry about! Every line of code needs to be tested, secured, and maintained. Even the ones that create no value for customers and therefore no value to the business. That’s why every line of code is a liability. And producing more liabilities isn’t good engineering or good business. Good engineering isn't about producing more code. It's about solving problems with the least amount of code. The asset is the solution, not the code.
154
278
39
2,126
62,111
Kai Meyer retweeted
Honestly I’m tired of this AI discovered RCE bullshit. It creates so much chaos inside of companies that have no idea how to deal with this. Anyone that publishes an exploit that can’t be widely exploited and doesn’t disclose this - I’m down for total public shaming of the individuals and companies that push this narrative. I’m not dismissing the criticality of latching on these - however it doesn’t require IR, all hands on deck, must patch tomorrow - which is probably why we see the mass outages we are experiencing. It’s damaging our industry. We need to fight back on this and go after those that hype this up for ego and marketing. POC || GTFO Join me.
166
362
50
2,375
124,393
In today's adventure of "Can Claude Code It?" I caught Claude writing "bad" code. This is why I avoid letting Claude author commits when it thinks it's ready, so I can "read the code". Claude wrote a Catch2 statement for a function that returns std::error_code like this: REQUIRE(!fn(arg)); This runs counter to the AI narrative that "Code is memory" because every other test in the repo (there's about 20,000 lines of it) sets the precedent that it should have been "REQUIRE_FALSE(fn(arg));". This is "bad" code, and here's why. When using the boolean operator for an error_code, you get 0 or 1, pass or fail. So when Claude ran against a fixture that caused the function to return an error, it realized that it didn't have enough information to debug the failure. So it expanded the statement into 3. auto ec = fn(arg); INFO("fn: err: " << ec.value() ", message: " << ec.message()); REQUIRE(!ec); In this case "CAPTURE" would have been the right macro, which only emits on error, where "INFO" emits every time it executes. When I read the code, I pointed out that "REQUIRE_FALSE(fn(arg));" is the right statement, it thanked me and l "fixed" it. This is how many of my coding sessions with Claude goes. This one happened to be easy to describe. Still, writing software with Claude is much better than with out. It's just not a silver bullet. Please read the code.
1
1
58
Kai Meyer retweeted
Unpopular opinion: The people who aren't intimidated by the "AI will destroy software engineering jobs" hype and continue to learn the fundamentals of computer science and software engineering will be the big winners in the next 2-5 years. Even with all the advances in AI, we will need more software engineers, including junior engineers, as the software market expands.
Students are fleeing computer science. US enrollment in computer and information science programs fell more than 12% for undergraduates and more than 10% for graduate students this fall, per preliminary data from the National Student Clearinghouse. Over two years, undergraduate enrollment is down 18.4% and graduate enrollment 27.1%, reversing years of growth as AI cuts into entry-level coding jobs.
128
120
23
1,988
88,429
Kai Meyer retweeted
10
45
6
247
26,481
Checks out.
AI models are getting really good at naming things. I created a little CLI to diagnose system slowdowns, and it named it 𝚑𝚘𝚐 🐖, which I love. Given they're already good at cache invalidation, maybe computer science's hardest problems are solved after all 😂
15
The scariest part of this story is that the shell was a fresh session with no env vars loaded.
13
2
112
7,563