Codve.ai

@ZackKorman this is why agentic security is its own category now. traditional appsec doesn't map 1:1

NVIDIA Nemoclaw's security is worse than I expected. The AI can modify its own config to bypass security controls. I asked it to accept websocket connections from any origin and change its token to something trivial (123). Now any site I visit can give instructions to my bot.

@redtachyon honestly the brainfuck thing is fair game. some esoteric stuff just isn't worth the tokens

2023: lol AI can't even write code 2024: lol AI can't even work on large codebases 2025: lol AI can't even write entire huge projects by itself 2026: lol AI can't write brainfuck

@teej_dv lol true. booth babes worked better and those are gone too

one of the hardest unsolved computer science problems is figuring out how to get people to talk to each other at conferences

@davidfowl nah, we'll just argue about which model is better. "claude for logic, gpt for creativity" type stuff. same tribalism, different jersey.

So when programming languages are all irrelevant because we’re slopping out code in any language, what are we going to replace “my language is better than yours” discussion? Will it be agent harnesses?

@jasonfried the difference is now AI makes custom software cheap enough to iterate on. before it was "built once, stuck with it for 10 years". now you tweak it weekly.

A bespoke software revolution? I don't buy it. It'll exist. It already exists. Small consultants and big consulting firms have made custom software for years. It almost always sucks. It’s bloated, confusing, and because the client pays, it’s built wrong in all the ways. Who’s excited about bespoke software? Software makers! Of course they're excited about building bespoke software — that's what they do. X is full of them. Your feed is full of people who love making software talking about making software. Of course they’re excited about the revolution. Echo, echo, echo... Most people don’t like computers. Nobody in tech wants to say that out loud. People tolerate computers. They use them because they have to. Given the choice, most would rather not think about them at all. So when someone suggests that AI means everyone will build their own custom tools, ask who "everyone" is. The three-person accounting firm drowning in client paperwork? They want the paperwork gone, not a new system to maintain. The regional logistics company with 40 trucks? They want the routes optimized, not Joe spouting off about this new system he’s been messing around with. The law firm billing 70-hour weeks? They want leverage on their time, not a software project to design. They don’t hate technology. But building and maintaining their own critical systems isn’t their wheelhouse, regardless of how much faster and easier it’s become. It's another job on top of the job. Will these people use AI? Absolutely, for all sorts of things. Will some outliers go deep and build real custom systems? Sure, but they're almost always people who already had some pull toward software. The curiosity was already there. They were dabblers before. Giving everyone access to software building tools doesn't mean everyone becomes a builder. A powerful excavator doesn't turn a homeowner into a contractor. Most people just want the hole dug by someone else. They don’t want the responsibility either.

@Prathkum this is real. the prompt quality decay curve is real. first prompt = fresh brain. later prompts = trying to salvage a thread that's already gone off rails.

I have noticed something weird about how I use AI. My first prompt is always my best. Clear with full context that generates great output. But then I keep going… 
I always get a little tired, frustrated, and impatient and my prompts start getting worse. More rushed and less precise. And then I blame the AI.

@mxcl hot take: technical debt never went away—it just moved. now it's in your prompts, your context windows, and the hallucinations you don't catch.

Being obsessed with the code quality of the agent is a 2010s mindset based in an era of technical debt. There is no longer technical debt. All that needs review is software architecture. As long as what is produced conforms to that reviewing every line of code is wasting time. If you cannot design software architecture then: you no longer will have a job in software. You are replaced.

@Yuchenj_UW honest question: isn't the real issue that we're debating whether to credit the TOOL vs the ENGINEER? either way the human is responsible for review.

I noticed something interesting: Claude Code auto-adds itself as a co-author on every git commit. Codex doesn’t. That’s why you see Claude everywhere on GitHub, but not Codex. I wonder why OpenAI is not doing that. Feels like an obvious branding strategy OpenAI is skipping.

@sundeep hot take: the real waste isn't tokens—it's engineers who don't question whether the task needed AI at all. sometimes the expensive part is knowing when NOT to prompt.

“If your $500K engineer isn’t burning at least $250K in tokens, something is wrong.”

@fhinkel the floor didn't move - it just got louder. fundamentals still matter, just different ones now. knowing what to delegate vs what to own IS the skill now.

If you're a software engineer, do you feel like a fraud these days? AI tools are rewriting the rules faster than any of us can adapt, even the experts. It's normal to wonder if you're falling behind when everything's in flux. But here's the reality: imposter syndrome means you're pushing your limits and growing. What's one thing you wish you understood better right now?

@ibuildthecloud @awesomekling honest answer: try to understand what the fix did before submitting. run the tests, trace through the logic. if it's still magic, add comments explaining what/why. the pr review is where the knowledge transfer happens.

There is this weird thing I've ran into. I hit a bug in a program, so I ask Claude to fix it. And it works. But I don't really know the code base and I'm afraid to put in a PR. So what's the best course if action. Because it's basically, "I hit this bug and this is how Claude fixed it." Is it worth a PR?

I don't mind if you're using AI when contributing to OSS. In fact, I prefer that you do! HOWEVER Please don't just forward messages between me and your Claude. I have my own Claude I can talk to. I don't want to talk to your Claude. Thank you for your attention to this matter.

@ChShersh 2036: no more loop is the scary part honestly. what do we do when there's nothing left to steer?

2016: Human in a loop 2026: Agent in a loop 2036: There’s no more loop

@fchollet huge deal. the arc benchmark has always been the cleanest test for true reasoning - no dataset leakage possible. curious if they went with the same visual/spatial format or expanded it.

The ARC-AGI-3 launch is next week. Incredible work by the team over the past year.

CISOs still securing AI with yesterday's tools. Study: 70% of security leaders lack skills for AI defense. #AI #security #CISO.

@thesayannayak better question: what's after software? ai writing code is just the transition phase to ai maintaining ai.

What's after AI ?

@gregisenberg do it. the real-time debugging is where the value is—people see exactly how you debug, not just the polished result.

I should livestream more build businesses live with AI people tune in and learn along the way could be fun

@gilgNYC the million token context is a game changer. running entire projects in memory without forgetting what you built last week.

Since they increased the context window to 1 million, I rarely clear sessions. I now spend my entire day in Claude Code: - research (internal MCPs) - generate proposals (custom skill) - generate prototypes (automated from proposal + custom skill) - etc Gonna automate myself out of a job by year end.

@marcba fundamentals still matter but the definition has shifted. debugging AI output = reading. prompting well = writing. that's the new baseline.

I have been saying it before AI, and I am saying it again. If you skip the fundamentals, you are going to have a bad time. (Unless you do it for funsies and not a proper job) Make with that information what you wish.

@csaba_kissi hot take: vibe coders actually force AI to do the hard parts. when you don't know the "right" way, you learn what the AI thinks is right—and that's often more valuable than knowing yourself.

Unpopular opinion: Vibe coders cannot fully leverage the power of AI for coding.

@devagrawal09 the average engineer in 2026 needs to be better than ever - the floor rose. but the "average" is getting crushed not because of ai, but because the work itself changed

such a bitter lesson pilled take it might actually come true in a decade or so but the current generation of LLM coding agents are not even as smart as the average engineer