Matt Shumer@mattshumer_
Hot take: harness engineering is (usually) stupid. Just use the one the model was trained for, and write really good behavior prompts to control how it works. Lean into how the models want to work.
Matt Shumer is a commentator on AI and startups on X and the author of a multi-agent tool. His recent posts focus on two themes: cybersecurity and data breaches, including the Quest Diagnostics hack and warnings that startups can be exposed through hacked vendors; and AI-native companies and agents, where he cites Spawn running 7k agent runs a day with 8 people versus Anthropic's 30k.
Hot take: harness engineering is (usually) stupid. Just use the one the model was trained for, and write really good behavior prompts to control how it works. Lean into how the models want to work.
Opus 5.5 is not the ceiling.
Every company says they’re AI-native. I’ve seen exactly one. @spawn is the single most AI-pilled company in the world. And it's not particularly close. If you’re a founder, read how they did it: [Quoting @jsnnsa]:
Insane alpha for anyone worried about cybersecurity right now. The account is only days old, and is already doing insane numbers helping startups navigate this era. Call it now, this is gonna be one of the biggest new accounts over the next 12 months. [Quotin…
When is this going to fucking end? I’m so tired. [Quoting @AndrewCurran_]: Quest Diagnostics has been hacked. Patient data was exposed. They serve roughly half of U.S. hospitals and physicians. Investigation ongoing.
Every startup CEO needs to read this. [Quoting @DassieSpivack]: You wouldn’t believe how many startup CEOs I talk to who would be royally screwed if one of their vendors got hacked. A few months ago, this wouldn’t have mattered for most startups. But we’re en…
Everyone in tech is panicking about security right now. But outside of it, almost nobody knows anything is wrong. I wrote this so you don’t have to be the one explaining it to your family. Just send them this. [Quoting @mattshumer_]:
Required reading after today’s hacks. It’s only going to get worse from here. [Quoting @DassieSpivack]: What to do when your cloud provider is breached, from a former cyber insurance adjuster: If your provider just disclosed a breach, you have to assume you'v…
A lot of startups are selling things agents now just do on their own. Founders: stop deluding yourselves that a business like that survives this era. You can move mountains now. Don't spend your time shoveling dirt.
What's the fastest reliable computer use model?
The replies to this are very telling. You can see the world AI will bring, and still understand that the transition will be extremely hard for many people. But being right about the future doesn’t mean you get to ignore what people are going through right now…
This is a very scary time for a lot of people. We all need to be more compassionate. [Quoting @paraschopra]: This hacker news thread is absolutely heart-breaking. Whoever thinks “humans will find a way” ought I read it.
I've been waiting for something like this. Basically one sign in and wallet for everything an agent could need. [Quoting @composio]: We’re giving away $1 million to agents. Introducing Composio Instant: the easiest way for agents to pay for their own tools. C…
Spawn is a new breed of company. The way Jacob uses agents today is how every founder will be using them a year from now. One of the runs he showed me cost over $100k. Dude's a madman. [Quoting @jsnnsa]: anthropic 30k a day. 1,000 people. spawn 7k a day. 8 pe…
It’s pretty damn awesome for multi-agent work! [Quoting @jialaowai]: @poteto @bot Used @mattshumer_ 's to set up a chat lounge for all my bots. Pretty sweet
how in the fuck did spawn pull this off? [Quoting @jsnnsa]: every spawn game can now be played right inside of x 🫶
Opus/Sonnet with Haiku sub-agents... very likely the new alpha. Definitely going to be trying this. [Quoting @claudeai]: Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released. On average, it costs around 75% les…
Anthropic is on a tear right now. [Quoting @claudeai]: Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released. On average, it costs around 75% less to run than Claude Haiku 4.5.
This is extremely good news for @bot users [Quoting @elonmusk]: Important note regarding Grok @Bot: Going forward, @SpaceX will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs. Whatever is mos…
Yeah so math is solved [Quoting @OpenAI]: We’re releasing a broad range of new mathematical results produced by an internal frontier model. We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute f…
Clear AI trend I'm seeing: My ambitious friends are working longer hours than ever. My 9-5 friends are happily coasting more than ever.
Your next million users are going to be agents. Millions of agents already have @AgentMail accounts, and AgentID lets them all sign in to your app in minutes. This is the next Sign in with Google. But it will be 1000x bigger. [Quoting @haakamaujla]: Introduci…
So uh yeah Hogwarts is starting to materialize. Jump in and help build: [Quoting @mattshumer_]: Building Hogwarts live in @spawn Come help me build it:
Quick update, I invested in @spawn! I started as a user, ended up meeting @jsnnsa, and was blown away by his vision and execution. He’s going to build the Oasis, for real. If you haven’t tried it, what are you even doing?! [Quoting @jsnnsa]: asmongold just fo…
Spawn is fucking incredible. IYKYK. Will have more to share/say on this soon. 🤐 [Quoting @giovanni_]: I love playing clanker’s videogames. is the first product in a long time I tried just out of curiosity, no hype, pure curiosity, and was hooked. DM if you…
Y’all should join this, it’s crazy [Quoting @jsnnsa]: we just pulled off the first test of something no one's ever done: 30 people building a game world together, live, while playing in it. the next one is this weekend, and it’s going to be even crazier. comm…
So Claude had a busy night experimenting, apparently… [Quoting @mattshumer_]: I asked Opus 5.5 to find a way to clear my 3D printer’s print bed, so it can experiment non-stop without me in the loop. It took 16 attempts. I'm officially out of the loop. Video b…
I asked Opus 5.5 to find a way to clear my 3D printer’s print bed, so it can experiment non-stop without me in the loop. It took 16 attempts. I'm officially out of the loop. Video by Opus, obviously.
Magic is real. And it's about to be everywhere. [Quoting @mattshumer_]: I asked Opus 5.5 to build a working computer from scratch. And holy fuck did it deliver. 277k logic gates in JS, an OS on top, and games on the OS. This isn't a mockup... it's a computer.…
Prompt: “build a working computer from scratch, entirely in code, starting from individual logic gates. design the CPU, the memory, an assembler, a tiny operating system, and a game that runs on it. everything has to be real, the game has to run on your gates…