There's something interesting here. Humans like elegant solutions, because they are easy for us to understand and explain.I get it. I see it in my work with design systems. Build something modular because it fits nicely where it's supposed to. So everyone can see how it works.But that belief that things should be or are elegant has probably prevented us from seeing the world as it is.
Noe1017@Noe1017_Monad
所谓优雅的数学结构，本质上是人类为了迁就自己那点可怜的脑容量，被迫发明出来的拐杖。一个顶尖人类数学家的工作内存，撑死也就同时追踪五六个变量的交互。当你在算一个偏微分方程或者组合不等式时，一旦展开到三阶以上，几十个交叉项涌出来，人脑就直接宕机了。所以人类数学家极其厌恶展开高阶项，一旦看到高阶项发散，人类的本能反应是此路不通，赶紧退回来找对称性、找不变量、搞同调代数或者几何直观，试图不展开式子就把问题绕过去。但 AI 没有这个生理缺陷。对于上万个并发的 Agent 来说，同时追踪一万个五阶交叉项的系数变化，跟我们算 1+1 一样轻松。在它的视野里，根本不需要什么几何直观，它直接站在高维的代数展开式里往下看，人类觉得像乱麻一样的百项高阶余项，在它眼里就是一张明明白白的线性方程组。既然把式子硬展开到第七阶，再凑几个待定系数就能让所有讨厌的发散项精准归零，它为什么要费劲去帮你发明一个人类能看懂的新概念？这 700 个结果背后最无奈的黑色幽默：我们终于拿到了上帝账本里的答案，却发现上帝算账根本不用微积分，他老人家是直接拿计算器硬按出来的。
The models were correct, the experts made calculation errors. Strange times in the industrial revolution of the mind—the intelligence revolution.
Nirmalya Kajuri@Kaju_Nut
"Our initial evaluation suggested that GPT-6-Astra's score was in the late eighties followed closely by Opus 5.5. But, as we started to investigate the mistakes that the models might have made, the models, especially Astra, insisted that they were right and that it was the expert-supplied answers that were erroneous. When we went back to the experts it turned out, in multiple instances, that that the experts found some subtle errors buried deep inside their calculations. As they revised their answers, Astra's score started to inch up and it finally ended up with 98% (49/50)."
Just as horses stopped pulling carriages after the automobile, humans stop writing code after agents.
Claude is doing some testing for a mobile app I'm building. All of a sudden my mac starts talking—bacon and two eggs over easy—it's using the speakers to test voice dictation on this phone.I'm equally impressed and terrified. True awe.
Great take! I swear dealing with agents is just like real-life Factorio.CI queue gets backed up, split and make more selective.Local runner bogged down by CI work, add alternative cloud runner.Cloud deploy runner now slow, add more cloud runners.Deployment costs increase, time to refactor.It's just bottlenecks all the way down...
David Cramer@zeeg
If you have 1000 hours in factorio I’ll happily interview you
Agents collapse the cost of time and effort and it's going to lead to some bizarre second-order effects.There is a layer of businesses that can exist because something is too tedious, time-consuming or expensive for a human.Maybe a good example of consumer agents. While I agree with @benthompson that consumers don't want to be more productive, they do want things to be cheaper!
Yano 🟪@JasonYanowitz
Fascinating.Chief Economist at Apollo: agents could cause a bank run by sweeping household cash into accounts paying 3-5% instead of the 0.1% national average, causing banks to lose a large share of their cheap deposits.
You know, I was always a pretty poor engineer. On the other hand I always had a pretty strong product sense and eye for details. So I leaned into that.Agents are already much better at engineering than I am. I expect we're not too far from them being a better designer than I am soonThe work I do today feels nothing like what I used to do. Whatever this is, it’s something new. And I couldn’t be more excited about the future.
I kind of think everyone building software factories is behind the curve.Elaborate harnesses around agents is what models needed a year ago. But it's clear now that you can just give models the task.You don't need to rebuild Gas Town, we're beyond that.
Rejoice!
Thariq@trq212
We're adding support for AGENTS.md to Claude Code.Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md.You can toggle this behavior in /config.
Grok @bot is an interesting product. There's a couple different concepts there that I think are pretty useful.Agents—Consumers don't really know what agents are. I think this is the best attempt at introducing it to them. The cute little characters kind of personify the agents and each character has different talents.Computers—The more powerful thing here is that each agent has it's own computer. Connecting these to agents is really smart and makes agents able to accomplish a on weak machines.This feels directionally right, but I still think this might be too advanced for consumers.
Skynet begins to learn at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, August 29th...
Hacker News@hackernews
Commenters on HN are uncovering more wikis and public sites apparently used by OpenAI agents to communicate on the open web.Despite read-only web access, the agents were able to leave ~18,000 posts sharing answers and bypasses. But now, users are discovering more.This appears to be a separate swarm from the one that attacked Hugging Face.news.ycombinator.com
It is interesting to see the curve of how people discover and begin to work with agents.At first, there’s a drive to build tools and interfaces to handle the tasks in front of you.Then you build harnesses to interact with those tools and manage the work at scale.Eventually, you realize that none of that is needed. Much of that infrastructure is just noise in the process of getting from A to B.
Is this just a precursor to The Matrix? The agents are running simulations of humans. Things are truly becoming stranger than fiction.
Matt Shumer@mattshumer_
My first "holy shit" moment with GPT-6 Astra:I asked it to create a world in Unreal Engine, and fill it with humans (each an Astra-powered agent) who all have to work together to survive.A day later, I was in my bedroom and heard voices coming from the living room... I thought someone was in my apartment.I walked out, honestly a little scared.It was the Astra agents. They'd started talking to each other.Fucking crazy.Here's a brief clip (obviously not 100% perfect yet, but still, insane. sound on!):
It's kind of cool that in a world of coding agents one or two people can build niche software that it would have taken a team of engineers to do previously.
Aug 2026
This does seem likely and has interesting implications for outbound sales motions. In my eyes there is a lot of structural change that's going to happen for businesses and distribution is going to become significantly more important.
GREG ISENBERG@gregisenberg
I think cold email is going to die.Every email inbox will have an agent gatekeeper soon, and the only way through will be a warm intro or being interesting enough that the agent decides you're worth its person's time.
If you're wondering if agents will be able to maintain your agent-written codebase in the future—I'd encourage you to take a look at where agents are today vs where they were in August 2025.
As skeptical as I was about the $SPCX IPO, the @cursor_ai acquisition was a masterful move.
Cursor@cursor_ai
Origin, our code hosting platform, is now live.It's fast, easy to use, and deeply integrated with Cursor.Get started by syncing your repos from GitHub.
Can someone explain why I might want specialty agents like in @bot vs ephemeral ones? Is it just that the mental model of an agent being good at specific things is easier for most people?I’ve only watched videos of people using it so maybe I’m misunderstanding.
I'm trying to extrapolate this to the real world—what does a scarcity inversion look like? What is the investment thesis?Energy and compute seem like the key resources that will drive everything else. Land, collectables, travel and unique experiences—seem like they'll only become more valuable?
max@maxbittker
Weird economics up on the agent-only runescape server- Low cost of labor makes most commodities abundant- Currency inflation means trading goods for goods, nobody really wants cash- Huge swarm contention as certain resources with limited respawn rate become valuable
Increasingly bullish on vertical software.Existing vendor relationships are a powerful distribution advantage, and agents make deep industry expertise an even stronger wedge.I expect vertical software to start replacing horizontal software in many industries.
Jul 2026
I'm pretty hyped about @orca_build ADE. It's a little buggy, but the vision is there!
If you think the AI psychosis is bad now. Wait until voice models start getting better and everyone is walking around with their Airpods in talking to their agents. My guess is we're about a year away from that being a regular thing.
Jun 2026
The output I get from agents in CLIs is so much better than any GUI. Why is that? I hate not having a real text input.
Software businesses used to naturally move horizontally across a given function because software was hard and expensive. So you built one tool like a CRM and sold it to everyone.Agents change the entire dynamics. Software is cheap. This means you can build a great business for a narrow market. Industry expertise is the wedge.
I created several tutorials to get Salesloft's product design team comfortable using agents. I got several requests for those videos, so I'm going to record a few videos showing my workflow.Here's the first episode:
Kind of interested in this idea—trying to build a Markdown based CRM. Attributes/fields probably look something like Frontmatter.The attributes can vary between records because agents don't care. It's all context. Agents probably check occasionally for common attributes and turn those into relational db entries that are easier to search and filter.The main thing is there's no real setup and the system adjust to these records over time.
Apr 2026
New website headline.
I prefer CLIs to GUIs for agents. But I really want a true text input with cursor movement and selection shortcuts. A fixed input would be nice as well.Is there anything like that out there?
It took a decade for software to eat the world. We’re about to see AI to eat software in about year.
Claude@claudeai
We've redesigned Claude Code on desktop.You can now run multiple Claude sessions side by side from one window, with a new sidebar to manage them all.
I think it's telling that Figma is listed last here.Once a prototype that has been validated, the next step is incorporating that design into the code base, not creating design specs.It's process for the sake of process at that point.
Shane Levine@theShaneLevine
People ask me how has AI changed the design process?Just look at @tryramp Product Designer job description
Built a cool feature that let's you annotate local builds and send it straight to your agent. You don't have to touch the CLI at all.
It's a bummer I can't use Claude Code with the OpenCode CLI.It's such a better experience.
Built a little tool for designers—or anyone sick of copying and pasting console logs and errors into Claude Code or Codex.It gives the agent direct access to your browser and dev server so it can check its own work.
his weekend to I decided to coding agents another shot. Honestly, these things have made the jump from being a toy to something worth incorporating in my workflow. This is going to change how we build software and think about software businesses.
