I had been procrastinating putting in the effort to find a decent web designer to redesign our company’s website because I couldn’t stomach the hours I would need to put in to educate them about our messaging and to slowly go around and around iteratively to get the design nailed.
Last week, I decided to try building the site myself using Codex (the terminal one). I chose Astro as the framework because I wanted to learn about it. I fed it some marketing framework materials (positioning statements and whatnot) and showed it some website designs that we like. I then asked it to produce a first cut and it one-shotted a pretty decent bit of output.
AGI is definitely a few more years away, because I’ve since invested probably 30 hours of iteration to make the site into something that is closer to what I eventually want to launch. But here’s the thing: I never intended for Codex to produce THE final website. But now I’m thinking, “maybe we can?” On my team, we have just enough expertise and design know-how to at least know what looks good and we are developers so we definitely know what good code looks like. And Codex is nailing it on both those fronts.
As I said, we’re far from AGI. There’s no way I can one-shot something like this. It requires iteration with humans who have years of “context” built up. But maybe the days of hiring a designer and just praying that they somehow get it right are behind us.
PS: Yes, I spent several hours on the weekend getting Codex to add animations, sound effects, and a mini game to our home page hero graphic. That was fun. I look forward to the creativity that people unleash with tools like this in the coming months.
Yes, this actually works. In 2026, software engineering is going to change a great deal as a result, and if you're not at least experimenting with this stuff to learn what it's capable of, that's a red flag for your career prospects.
I don't mean this in a disparaging way. But we're at a car-meets-horse-and-buggy moment and it's happening really quickly. We all need to at least try driving a car and maybe park the horse in the stable for a few hours.
The FOMO nonsense is really uncalled for. If everything is going to be vibecoded in the future then either theres going to be a million code-unfucking jobs or no jobs at all.
Attitudes like that, where you believe that the richeous AI pushers will be saved from the coming rapture meanwhile everyone else will be out on the streets, really make people hate the AI crowd
The comment you’re replying to is actually very sensible and non-hypey. I wouldn’t even categorize it as particularly pro-AI, considering how ridiculous some of the frothing pro-AI stuff can get.
Uhuh, heard the same thing about IDEs, Machine Learning in your tools and others. Yet the most impressive people that I’ve met, actual wizards who could achieve what no one else could, were using EMacs or Vim.
I had ChatGPT spend a few kWh coming up with Algorithmostartupoventurecapitoopensourcolicensioprivacysecuritorustigogolokubernetocloudiosaasodistributedodatabasolatencyphoboshowhnaskhncommentopedantolongformoaillmopromptomancyethicoregulatiocontroversioburnoutikon, which apparently describes the vibe here on HN.
I tried talking to Claude today. What a nightmare. It constantly interrupts you. I don’t mind if Claude wants to spend ten seconds thinking about its reply, but at least let ME finish my thought. Without decent turn-taking, the AI seems impolite and it’s just an icky experience. I hope tech like this gets widely distributed soon because there are so many situations in which I would love to talk with a model. If only it worked.
Agreed. English is not my native language. And I do speak it well, it's just that sometimes I need a second to think mid-sentence. None of the live chat models out there handle this well. Claude just starts answering before I've even had the chance to finish a sentence.
Anthropic doesn't have any realtime multimodal audio models available, they just use STT and TTS models slapped on top of Claude. So they are currently the worst provider if you actually want to use voice communication.
I think Anthropic currently has a slight edge for coding, but this is changing constantly with every new model. For business applications, where tool calling and multi-modality matter a lot, OpenAI is and always has been superior. Only recently Google started to put some small dents in their moat. OpenAI also has the best platform, but less because it is good and more because Google and Anthropic are truly dismal in every regard when it comes to devx. I also feel like Google has accrued an edge in hard-core science, but that is just a personal feeling and I haven't seen any hard data on this yet.
I love Anthropic's models but their realtime voice is absolutely terrible. Every time I use it there is at least once that I curse at it for interrupting me.
My main use case for OpenAI/ChatGPT at this point is realtime voice chats.
OpenAI has done a pretty great job w/ realtime (their realtime API is pretty fantastic out of the box... not perfect, but pretty fantastic and dead simple setup). I can have what feels like a legitimate conversation with AI and it's downright magical feeling.
That said, the output is created by OpenAI models so it's... not my favorite.
I sometimes use ChatGPT realtime to think through/work through a problem/idea, have it create a detailed summary, then upload that summary to Claude to let 4.5 Opus rewrite/audit and come up with a better final output.
I use Claude Code for everything, and I love Anthropic's models. I don't know why, but it wasn't until reading this that I realized: I can use Sparrow-1 with Anthropic's models within CVI. Adding this to my todo list.
The one thing that really surprised me, the thing I learned that's affected my conversational abilities the most: turn taking in conversation is a negotiation: there are no set rules. There are protocols:
- bids
- holds / stays
- implications (semantic / prosodic)
But then the actual flow of the conversation is deeply semantic in the best conversations, and the rules are very much a "dance" or a negotiation between partners.
That's an interesting way to think about it, I like that.
It also implies that being the person who has something to say but is unable to get into the conversation due to following the conversational semantics is akin to going to a dance in your nice clothes but not being able to find a dance partner.
Yeah, I can relate to that. Maybe it's also because you are too shy to ask someone to dance. I think I learned that lesson: just ask, and be unafraid to fail. Things tend to work themselves out. Much of this is experimentation. I think our models need to be open to that: which is one cool thing about Sparrow-1: it's a meta-in-context learner. This means that when it try's and fails, or you try and fail, it learns at runtime to adapt.
My best guess after dipping my toe into semiconductor fabrication a decade ago is that there is a mysterious guru in a cave under a volcano who decides which customers get access to which nodes at which prices.
I have found significant frustration since the pandemic in the most unexpected place: the new expectation of punctuality for online meetings. In meatspace, in the before times, if a meeting was set for 2pm in the board room, everyone understood this to mean 2pm was the time to come through the door and chat for a bit while everyone got comfortable. The actual meeting would start at 2:05-2:10.
Online, there is no equivalent to walking in the door for a bit of a chit chat. Everyone just materializes instantly and then we’re supposed to be ready to go by 2:01. I miss meatspace for meetings and the more casual, human-matched pace.
This is cute. I think within 36 months AI will replace middle management in software companies. This will happen because, ironically, today’s middle managers will switch back to being individual contributors, using AI to contribute PRs once again (who doesn’t prefer this anyway?).
Sufficiently powerful AI can become the middle manager of everyone’s dreams. Wonderfully effective interpersonal skills, no personality defects. Fair and timely feedback.
Where is the AI going to get the information required to do the job?
How is the AI going to notice that Bob looks a bit burnt out, or understand which projects to work on/prioritise?
Who is going to set the AI managers objectives? Are they simple or are they multi-factorial and sometimes conflicting? Does the objective function stay static over time? If not how is it updated?
How are you going to download all the historic experience of the manager to the AI or are they just going to learn on the job.
What happens when your manager AI starts talking to another teams manager AI? Will you just re-invent office politics but in AI form? Will you learn how to game your AI manager as you understand and potentially control all it's inputs?
If we use outsourcing as proxy for what jobs will move to AI first, management jobs will be the last to be replaced.
Managing is about building relationships to coordinate and prioritize work and even though LLMs have excellent soft skills, they can't build relationships.
Spot on. AI might simulate the message perfectly, but it can't hold the social capital and trust required to actually move a team when things get tough.
> Sufficiently powerful AI can become the middle manager of everyone’s dreams. Wonderfully effective interpersonal skills, no personality defects. Fair and timely feedback.
> The girls liked it because Manna didn’t hit on them either. Manna simply asked you to do something, you did it, you said, “OK”, and Manna asked you to do the next step.
I actually ran this specific 'Backchannel VP' scenario through raw GPT-4 before building the hard-coded version, and the results were surprisingly 'meh.'
The missing piece wasn't intelligence, but statefulness and emotional memory.
A human manager (or VP) remembers that you embarrassed them in a meeting three weeks ago, and that hidden state dictates their reaction today. LLMs—currently—are too 'forgiving' and rational. They don't hold grudges or play power games naturally.
Until AI can simulate that messy, long-term 'political capital' (or lack thereof), I think we still need humans to navigate other humans. But I agree, for pure PR review and logical feedback, I'd take an AI manager any day!
I'm not sure you understand the job. Do you have management experience? It's mostly about discussion, agreeing on how to proceed, and building relationships. It's not clear to me at all that people will want to work for AI instead of a real human that cares. I certainly wouldn't.
Agreed. People work for people, not APIs. That human connection and the feeling that your manager actually cares ( hopefully :D ) is the one thing you can't automate away
I think most of these objections are valid against a “ChatGPT-in-a-box is your manager” framing. That’s not what I meant by “AI replaces middle management”.
What I did mean is: within ~36 months, a large chunk of the coordination + information-routing + prioritization plumbing that currently consumes a lot of EM/PM time gets automated, so orgs can run materially flatter.
A few specifics to the questions:
“Where does the AI get the information?”
Not from vibes. From the same places managers already get it, but with fewer blind spots and better recall: issue trackers, PRs, incident timelines, on-call load, review latency, meeting notes, customer tickets, delivery metrics, lightweight check-ins. The “AI manager” is really a system with tools + permissions + audit logs, not a standalone LLM.
“How does it notice burnout / team health?”
Two parts: (1) observable signals (sustained after-hours activity, chronic context switching, on-call spikes, growing review queues, missed 1:1s, reduced throughput variance), and (2) explicit human input (quick pulse check-ins, opt-in journaling, “I’m overloaded” flags). Humans are still in the loop for the “I’m not okay” stuff. The AI just catches it earlier and more consistently than a busy manager with 8 directs and 30 Slack threads.
“Who sets objectives / what about conflicting goals?”
Exactly: humans. Strategy is still human-owned. But translating “increase reliability without killing roadmap” into day-to-day sequencing, tradeoff visibility, and risk accounting is where software can help a lot. Think: continuous, explainable prioritization that shows its work (“we’re pushing this because it reduces SEV risk by X and unblocks Y; here are the assumptions”).
“What about historic experience?”
You don’t “download” a manager’s career. You encode the org’s policies, past decisions, and constraints into an accessible memory: postmortems, decision records, architecture notes, norms. The AI won’t have wisdom-by-osmosis, but it will have perfect retrieval of “what happened last time we tried this” and it won’t forget the quiet lessons buried in docs.
“Will we reinvent office politics / will people game it?”
We already do. The difference is: an AI system can be designed to be harder to game because inputs can be cross-validated (tickets vs PRs vs customer impact vs peer feedback) and the rules can be transparent and audited. Also: if you try to game an AI that logs its reasoning, you leave a paper trail. That alone changes incentives.
“Relationships and trust can’t be automated.”
Agree. And that’s why I don’t think “management disappears.” I think it unbundles the human part (trust, coaching, hard conversations, hiring/firing accountability, culture) - that part stays human.
The mechanical part (status synthesis, dependency chasing, agenda generation, follow-up enforcement, draft feedback, metric hygiene, “what should we do next and why”) becomes mostly automated. But did everyone love that part anyway? I don't.
So the likely outcome isn’t “everyone reports to an API”. It’s: fewer layers, more player-coaches, and AI doing the boring middle-management work that currently eats the calendar.
In other words: I’m not claiming AI becomes the perfect human manager. I’m claiming it makes the org need less middle management by automating the parts that are fundamentally information processing.
I’m actually neurodivergent. It’s not much of a gift, let me tell you. I fully count on my neurotypical teammates to ensure the company doesn’t go off the rails. If it were just me, cool stuff would definitely happen - randomly. And we would run out of money and fail.
> If it were just me, cool stuff would definitely happen - randomly. And we would run out of money and fail.
I like to think there needs to be some distribution. Just like how there should be an adversarial process where the business people are only concerned about the money and the programmers are only concerned about the product [0]. You need code monkeys to build a product and make it something people want to use. But you need business monkeys to make it profitable and to get the money to build better stuff. If the code monkeys dominate then user interfaces suck, building takes forever, and is vastly outsold by products with worse features[1]. If the business monkeys dominate too much you just get enshitification because they only care about the product so far as people will buy it (even if they turn down future sales).
Turtles all the way down. Even on just the code monkey side you need some of those people that swing for the fences and miss a lot to get those big leaps but most people should no be doing this and instead keep things marching forward. Both groups are useful, just in different ways. An adversarial process can also be useful here. But everyone also needs to recognize they're on the same team.
[0] more accurately: the business people optimize for the money, conditioned on the product; the engineers optimize the product, conditioned on the money.
[1] throughout history the products that are technically better frequently lose to inferior ones. Because success isn't purely reliant upon features
Last week, I decided to try building the site myself using Codex (the terminal one). I chose Astro as the framework because I wanted to learn about it. I fed it some marketing framework materials (positioning statements and whatnot) and showed it some website designs that we like. I then asked it to produce a first cut and it one-shotted a pretty decent bit of output.
AGI is definitely a few more years away, because I’ve since invested probably 30 hours of iteration to make the site into something that is closer to what I eventually want to launch. But here’s the thing: I never intended for Codex to produce THE final website. But now I’m thinking, “maybe we can?” On my team, we have just enough expertise and design know-how to at least know what looks good and we are developers so we definitely know what good code looks like. And Codex is nailing it on both those fronts.
As I said, we’re far from AGI. There’s no way I can one-shot something like this. It requires iteration with humans who have years of “context” built up. But maybe the days of hiring a designer and just praying that they somehow get it right are behind us.
reply