holtkam2

How do you have such stunning confidence that you're replying to someone who knows 'nothing about' the subject matter? Maybe he's a neuroscientist.

I think YOU are the one showing stunning confidence in something you know nothing about (op's area of expertise. Unless you know them personally. Then ignore me).

It doesn’t need to be perfect - it just needs to be as effective as a human CEO… luckily, many human CEOs suck at their job

At a certain point, can a judge just please rule that Meta has lost its right to do business in the United States? It repeatedly ignores rulings and has lost credibility when it claims that it will change its practices. I don't see why they 'need' to exist. Another, more responsible social media platform company would quickly spring up in its place if it were dissolved.

I have bought every generation of Xbox since the original (I had it in 2005). This is a dealbreaker for me. Won’t be buying their next gen system.

My rule is: I only let AI code for me, I don’t let it think for me.

Since writing is thinking, coding is thinking since coding is writing. That means any time I am not certain how I’m gonna implement some feature or bug fix, I have to code it myself because that’s the only way I can force myself to think through it. Only when I get to a point where I’m line “ok I know exactly what to do now- all that’s left to do is type it out” that’s when AI can be employed - essentially as a autocomplete.

This is only for projects where I will be held responsible for outcomes and must understand how it works. For hackathon / personal projects, I vibe away.

I also use AI to brainstorm at the outset of the task when I don’t know where to start at all.

Is this inefficient? My take: no. It’s maximally efficient. Over the long term it gives me an edge over any teammates who just vibe code everything because I actually, you know, understand how stuff works. I become the guy who can save the day at 3am when the team’s business critical app goes down. I become the guy that gets pulled into meetings so the suits can ask “is this possible?”. I see opportunities and problems before my teammates because I have a relationship with our code and system that they never took the time to develop or think about.

Yall are missing my point - you don’t gotta replace the upper management - they will continue to hold the responsibility for outcomes and problems like they do today.

There will just be like 75% fewer of those individuals because they all can do so much more per head

Can someone make an AI agent to target automating tasks of upper management already so they feel the pain as well?

You don’t need to even replace them outright, just make your agent so effective that [big tech company] requires fewer VPs and GMs etc.

Modern frontier LLMs are incredibly good at all types of knowledge work - this should be possible with enough elbow grease

Awesome animations!

That question is probably what inspired someone to create the first ever derivatives market

Great piece, but it misses the elephant in the room: the lion's share of economic benefit from refactoring will come from the fact that it makes it easier for humans to understand. That means 3am pages get resolved faster, fewer bugs will end up in prod, and your team can ship faster than your competition - gaining a leg up in the market. Most importantly, folks will be more comfortable accepting responsibility and ownership of a system when they understand it... this means that if/when something goes wrong, people will more quickly jump in and fix it, and when things could be better, folks will jump in and improve it.

Claude code? Who’s talking about Claude code? I am talking about generative AI. It has been better than humans at coding for years. It is also currently much better at things like taste, stakeholder communication, and literally everything else conceivable. Yet here we are and employment is pretty much unchanged. It’s a paradox imo.

The final user-facing responses are usually a tiny fraction of the total tokens used over the course of a given conversation turn. When you're doing any real work, reasoning and tool uses constitute the overwhelming majority of the tokens in / out... not the final user-facing response.

This guy's head will explode when he realizes that AI has been better at coding than humans for years and yet software engineering employment % has barely twitched...

If you want to fully understand / contribute to / fully leverage the most powerful technology humanity has ever devised (AI), you must learn to read and write code. That’s the only reason anyone should need.

If you used a time machine to go back to 2021 and showed someone the best open source LLMs from 2026 they would surely say “yeah that’s AGI”

This is the way. But also - keep your head up - your mindset matters more than you think. If you consider yourself a victim of your circumstances with no agency, this will be harder for you. So get your mentality right. Something along the lines of 'this is an exciting new era and I'm well prepared to master this new domain like I have many times before'.

The way I think of it has evolved a lot over the last 5 years. At this point I think human brains probably do something analogous to next token prediction when we think. For all the hype, I think LLMs are actually more, not less, intelligent than that average person realizes. I think it’s legit, actual intelligence, not just “artificial” intelligence. That may be a hot take but it’s just my perception.

at a certain point you're gonna need to change your benchmark because this will end up in the model's training set

I wish I could upvote this twice. We (devs) really REALLY need to consider on-device compute before going to the cloud for LLM inference.

IDK if the author's 'metacognition' needs to be a feature of the LLM itself.

I could imagine a harness that 1) reads LLM output 2) uses a research sub-agent to attempt to verify any factual claims 2) rephrase the main agent's output such that it conveys uncertainty if the factual claim cannot be independently verified

No, I'm not considering a career change. But my career will change.

I started off as a javacript developer. Then I was a full stack SWE. Then I was an Applied AI Engineer. I think enterprises will have a need for folks with technical expertise to deliver value - often new software - for a long time.

Until an enterprise like capital one can operate without anyone in the organization knowing how any of their technical infrastructure works (never?) I expect I'll be able to find work.

Just use the codex plugin for vscode ?

Someone tell me if my idea is dumb:

You know how adding ".ai" to your company name / domain name increases your valuation an order of magnitude or whatever?

Well has anyone tried chaining .ai's to their company name...???

For example could x.ai increase their valuation another 5-10x by renaming themselves x.ai.ai?

What about perplexity.ai.ai.ai?

Seems like a low-hanging fruit for increasing valuation / attention on your company, but let me know if I'm missing something.

This article forgot the strongest argument for hand writing your code. When the process is done, you understand it at a depth far beyond any vibe coder who created the same thing

Moving from SF to Chicago this week. This made me feel a little more excited. TY for the share!

To answer your question directly: yes I am bored of talking about AI. I think it’s funny how the folks who are screaming most loudly about their AI expertise typically have not built anything of value with it. They are so focused on the tool they have forgotten that the output is what mattered all along

Level 9: agent managers running agent teams Level 10: agent CEOs overseeing agent managers Level 11: agent board of directors overseeing the agent CEO

Level 12: agent superintelligence - single entity doing everything

Level 13: agent superagent, agenting agency agentically, in a loop, recursively, mega agent, agentic agent agent agency super AGI agent

Level 14: A G E N T

Are you telling me juniors aren’t facing hard problems anymore? I doubt that’s the case. They’re probably banging their heads against the wall trying to understand why this or that doesn’t work… just as they always had before. AI isn’t a magical wand that makes everything just “work”. Probably never will be.

Never - data centers will always offer more power if you only care about raw inference speed. HOWEVER I think that we'll reach the 'good enough' bar super soon. In 2-3 years I expect apple macs to be able to run a model as 'good' as Claude 4.6 sonnet at 90% of the inference speed we're used to from a cloud API.

Yes, I'm sure by then there will be better models on offer via cloud providers, but idk if I'll even care. I'm not doing science / research or complex mathematical proofs, I just want a model good enough to vibe code personal projects for fun. So I think at that point I'll stop being a OpenAI / Anthropic customer.

At the end of the day you need humans who understand the business critical (or safety critical) systems that underpin the enterprise.

Someone needs to be held accountable when things go wrong. Someone needs to be able to explain to the CEO why this or that is impossible.

If you want to have AI generate all the code for your business critical software, fine, but you better make sure you understand it well. Sometimes the fastest path to deep understanding is just coding things out yourself - so be it.

This is why the truly critical software doesn’t get developed much faster when AI tools are introduced. The bottleneck isn’t how fast the code can be created, it’s how fast humans can construct their understanding before they put their careers on the line by deploying it.

Ofc… this doesn’t apply to prototypes, hackathons, POCs, etc. for those “low stakes” projects, vibe code away, if you wish.