Matthew Schrager

Trading does provide value to the world, just in a way that is unintuitive to most people. People buy from Amazon or Walmart because they provide the best price/service among competitors. If Amazon didn't exist, it would not make its customers richer. They would pay more to someone else, or not buy at all. Same for trading. Market-makers win business by providing the best price for a security when someone wants to buy or sell. They do so in a brutally competitive market; the only way they win business is by charging less than dozens of competitors. If Jane Street or Citadel disappeared tomorrow, it would not make retail traders richer. They would pay more to someone else. It would make Citadel's competitors richer. This isn't hypothetical. All you have to do is look at what happened before these firms existed. Transaction costs were far higher. Transaction volumes were far lower. Trading was expensive, and therefore people did a lot less of it. The reason these firms make so much money is that they've collectively made trading so much cheaper that *people want to trade much more than before*. Trading volumes, especially from retail, have risen dramatically in recent years. Demand for the product these firms sell (liquidity) has gone up. All things equal, higher demand -> higher profit. But it's not a zero-sum game. They make money because they sell a product that's in high demand, and they do it more cost-effectively than their competitors. If they weren't there, their customers would either pay more to someone else, or not trade at all. It's value additive to society, just in an unintuitive and easy-to-vilify form.

@businessbarista @tenex_labs Excel junkie? Don’t we use Claude for that part now? :)

We are hiring for a Sr. Director of Strategic Finance at @tenex_labs. Looking for someone who can ultimately grow into a CFO role long term. This person will interface with me and my cofounder daily and we will look to them as a strategic partner as our business continues to hyperscale. If you're a good fit, apply here: jobs.ashbyhq.com/tenexlabs/b315… If you're not, send your smartest excel junkie our way.

@MTSlive Agentic engineering is the right term. It’s still engineering. Just at a higher level of abstraction.

"At a certain scale, vibecoded is a slur."  "If you vibecode software and it's generating 7, 8, 9, 10 figures of revenue, chances are you're not gonna be telling people that you vibecoded it. It just feels like a little bit of a slur." "And you know what's not a slur? Agentic engineering" " When Sam Altman said there's gonna be a one person, $1 billion company,  a lot of people thought he was crazy. But where we are in 2026, it sounds crazy to say, but I believe that it's possible." @gregisenberg

I agree with you re: complexity and the need to manage it and what can go wrong if you don’t. I just disagree that Anthropic’s uptime record is definitive evidence that they aren’t doing those things. The simpler explanation is that demand for their product outstrips the compute they have available to supply it, causing downtime.

@MatthewSchrager @RhysSullivan Complexity just continues to grow otherwise, and regressions start to happen due to context window length. I imagine this happens later with Claude Code and Codex, but very much doubt it’s been solved autonomously yet (or if it can be)

we literally have no idea how to make software anymore

The vibes among developers are shifting back to OpenAI. Codex is arguably better at most things than Claude Code other than UI, the desktop app is great and does some really cool computer use stuff, and they’re much more generous with usage than Anthropic. That article is based on 2025 data. In AI world, that’s forever ago. We’ll see what this war looks like in a few months.

It feels like Claude has surpassed ChatGTP as a product. Maybe for good. Especially given the financial reports on OpenAi (the once not-for-profit company seeking to IPO).

I think it comes down to how much domain complexity is baked into the system. If the app reduces to a nice interface over a database, it's replaceable. If the software embeds, say, knowledge of idiosyncratic HR/payroll laws from all 50 states, that's not worth replacing. Zooming out, to the extent many companies encounter common problems, it would be globally inefficient for all of them to reimplement overlapping solutions. But there's a continuum. Many problems are kinda/sorta common, but with many company-specific nuances. Directionally speaking, as the cost of implementation goes down, more of those on the margin will make sense to in-house than before. But it's definitely not everything.

@LuminousTheReal @SimonHoiberg They have mobile apps, you just use them from your phone. Claude supports mobile access to local sessions, which is nice, as well as cloud-based sessions. Codex just has cloud sessions at the moment, but I’m sure they will improve it.

@MatthewSchrager @SimonHoiberg mobile support how?

GPT 5.5 with OpenClaw is so good it's uncomfortable 😳 It's running with entire features now. Large, complex, end-to-end features. It specs it out for itself. Implements it, covers edge cases, tests it, does QA via browser automation, and comes back a few hours later with the whole thing finished (with only minor follow ups needed). And then it goes on to support my users who may have any issues or questions. All I do is voice notes. Not at the computer supervising or steering anymore. I could not be more convinced that developers are about to be entirely replaced. Every single last one of us.

They have very publicly taken more aggressive approach to acquiring compute. Dario (in)famously talked about Anthropic being more conservative in compute acquisition all the way back in *checks notes* February (man this space moves fast). What you’re describing is the result of that strategic divergence playing out in real time. OpenAI has better capacity because they spent many billions of dollars to acquire the compute to scale it. But it doesn’t tell you much about the quality of the software being built at each firm.

@MatthewSchrager OpenAI has

Jevon's Paradox for software development: when developer productivity increases dramatically, the set of software that is economical to build expands proportionally, and you get higher demand for developers. I think it's too early to tell what the long term net effect on developer jobs will be. It depends on elasticities of demand and supply etc. But those proclaiming the death of the software developer are too certain. What is certain: the nature of the role is changing, and it will look dramatically different than it has for the past 30 years. For any individual developer, the best thing they can do is become one of the best in the world at this new AI-enabled skillset. That role will be in huge demand as far as the eye can see.

Narrative violation! Amazon Web Services CEO Matt Garman: "we are hiring just as many software developers" "demand for that is accelerating" with AI agents. He notes nature of job changes (broader versus just doing one micro-skill set over and over)

Yep, agreed. Coding agents are force multipliers for software developers, and gateway drugs for the software-curious. But they do not reduce the fundamental complexity of architecting real-world systems, where the majority of difficulty and value is in the gap between "done" and "almost done". This is why proclamations of "everyone's going to write their own software" are misguided. They misjudge what software development really is. The hard part was never producing syntax, which is now commodified. The hard part is understanding the real world with sufficient detail to describe it systematically and in a way that is robust to entropy over time. In the right hands, coding agents are amazing tools to aid in this process. But they don't magically make the hard parts disappear.

Agentic coding is a huge boon for software developers that want to get far more done, great for IT people to build vastly more custom systems internally, great for domain experts that want to automate workflows or wire systems together, and absolutely fantastic for anyone curious to learn how to start coding. What it’s less great for is casually building complex software that you have to maintain on an ongoing basis and take on all the risk for. Upgrades, maintenance, keeping up to date with latest security issues, and so on, are taxes most knowledge workers aren’t familiar with or prepared for. Net net: we’re going to get 100X more software and vastly more software developers in the future. But that’s different from *everyone* rolling their own.

The vibes have shifted significantly to Codex over the past few weeks. It’s been arguably better than Claude Code for months at everything other than UI, the desktop app is increasingly impressive, and they are much more generous with tokens that Anthropic thanks to their compute advantage. It will take a while for the sentiment to filter into the mainstream, but it’s been heading this direction for a while for those paying attention.

I believe it. My Codex use went from 0% to 30%. Will go up further probably. Poor numbers probably based on a few months ago.

FWIW, the story is based on 2025 numbers and misses the huge growth in Codex in recent months. Vibes have shifted back to OAI in the coding space. Codex + GPT 5.5 is better than Claude Code along most dimensions, and OAI is much more generous with tokens thanks to their compute advantage. TBD how it translates into the numbers, but I think this story is somewhat out of date. To be fair, hard for traditional journalism to keep up with the pace of this industry.

And so it (the bloodbath for investors) begins? If revenues from LLMs are hard to come by (outside of coding and some other niches), then the wild hyperscaler datacenter buildout costs can't be justified. Competition for the smaller than expected revenue pie could lead to price wars (see DeepSeek price slashing yesterday), further eroding revenues. Slowdowns and cutbacks in spending could lead to another cyclical downturn in the semiconductor industry - not a happy thought with those stocks currently in orbit. Look Ma - No net! "OpenAI missed multiple monthly revenue targets earlier this year after losing ground to Anthropic in the coding and enterprise markets, people familiar with its finances said." "It also missed its yearly revenue target for ChatGPT as well after Google’s Gemini saw massive growth late last year and ate into OpenAI’s market share, the people said. The company has also struggled with defection rates among subscribers, according to people familiar with those figures." wsj.com/tech/ai/openai…

@fejau_inc Yeah… this industry moves too fast for traditional journalism.

I mean yeah maybe comparing usage when your main customer was AI psychosis 4o redditors to now with Agentic coding on codex isn’t the most relevant comp on the go forward of a company

This is where the vibe coding revolution will ultimately end up. Software development isn’t hard because the syntax of programming languages is obscure. It’s hard because the real world is complex, and systems that interact with it must accurately model that complexity, both statically and over time. That complexity is fundamental. It is not reduced by easing the syntax burden. The job of software development is primarily the job of wrangling that complexity. Coding agents make that skillset *more* necessary, not less. Similar to how crypto speed ran the discovery of why finance is hard, vibe coders will speed run the discovery of why software is hard. Coding agents are miraculous tools. But like any powerful tool, they can produce catastrophic results when wielded by people who don’t know how to use them.

@TheStalwart @OGTaylorB Does this have something to do with you turning down the job?

@OGTaylorB The CEO iirc

I never interviewed at Jane Street, obviously, but I did interview at a prop shop in Austin. In one round I was given two joysticks, one that moved an inverted isosceles triangle on the screen, while the other adjusted its tilt. Goal was to get it centered and perfectly vertical.

@auchenberg Yep, anyone dunking on GH for downtime without mentioning that usage has gone vertical is missing the story.

GitHub experiencing exponential growth fueled by AI. "By February 2026, it was clear that we needed to design for a future that requires 30X today’s scale."

Trust me, they are trying. Everyone in the industry has had the “how do we eat into Jane Street’s margin” conversation. Turns out it’s really hard to do. They are very good.

@skyquake_1 Congrats.

@MatthewSchrager the guy said see you in hell scumbag before they locked me up so i assume yes when i get out in 2 years~

Since everyone is sharing, my Jane Street interview was quite unusual. I was prepared for brainteasers and math/finance questions but not "who are you" and "how did you get in here." They even mixed it up to keep me guessing, like "put down the knife" wasn't even a question

This was basically Dario's argument for Anthropic's conservative approach to acquiring compute. It looks bad at the moment as Anthropic struggles with capacity. But he wasn't *wrong*. It's just a probabilistic bet that pays off in the exact case this headline mentions.