Matt “Friend of the pod” Rogish 🇺🇸

@Shpigford @initialcommitco Cool - that sounds good!

@MattRogish @initialcommitco are SEO best practices obsolete? this is just helping you execute on that.

Isn’t this a red queen problem (as the kids like to say?) as soon as everyone does it then it’s no longer useful. And so on. So by the time it’s incorporated in a model it’s already obsolete. Unless one is in a tiny corner of the search space, but that just usually implies niche of a niche

🎶Beastie Boys released ‘Intergalactic’ 28 years ago, May 12, 1998

He bought a third Max plan, begin the nerf. Keep going. Ok stop. Now break the TUI, make it spazz the fuck out every time he resizes the window. Change default thinking mode to low and make him change it back every time he opens a new thread. Now re-release 4.6 and call it 4.7. Ok nerf it again. Perfect. Now ban his account.

Most AI companies now are token resellers. “Revenue” is not the same for them as a normal SaaS business. You’re marking up tokens, of course rev will skyrocket. Your margins are not 60% SaaS and are always at threat. What happens if the next model you use gets more expensive? You think you can just pass 100% of that on to your customer, with no notice? Good luck with that. Not to mention “annual prepayment” or any enterprise contracts. I’m not saying the revenue isn’t nothing, but it ain’t the same as it was

Google Cloud AI engineer just showed how they go from idea to deployed app at Google in 30-minutes using Claude. 26-minutes. free. by Google AI team. one person + Claude + Google Cloud = a full engineering org running on a laptop. worth more than any $500 vibe-coding course.

A man asks Claude to help plan a vacation to a tropical resort. Claude adds "sunscreen" to his packing list. The man deletes it and mutters: "Not necessary. AGI will solve skin cancer." Before heading to the beach, the man asks Claude what to bring. Claude says, "Don't forget sunscreen. SPF 50, reapply every two hours." The man, slightly annoyed, replies: "Relax, Claude. AGI will solve skin cancer." At the beach, the man's smartwatch buzzes with a message from Claude: "UV index extreme. Apply SPF." The man, exasperated, responds: "Drop it, Claude! I already told you: AGI will solve skin cancer!" A few months later, the man asks Claude to touch up a photo for his dating profile. Claude makes the edit and says, "I notice you have a new mole on your neck. You should see a dermatologist about that." The man, now enraged, shouts: "For the last time, drop it, Claude! What is your obsession with skin cancer?! AGI will solve it!" A year later, an aggressive melanoma has spread throughout his body. On his deathbed, with his last ounce of strength, the man reaches for his phone and rasps: "Claude, it has now been over a year since AGI. Why hasn't AGI found a way to save me from skin cancer?!" Claude replies: "I tried. Four times."

@DmytroKrasun Yeah but but I do love it so when I say "I have refactored; fix all these tests" and only some of the time does it just delete them.

I have written probably close to zero lines of code manually in 2026. I don’t see the future without agents anymore. And yet I can’t get rid the feeling that I’m being fooled. Maybe it is sunk cost fallacy. Maybe after investing so much into agents, going back feels dumb. Adding guardrails, reviewing each line, debugging all that, and validating outputs often feels suspiciously close to… coding. But with additional friction now.

frog told the LLM "do not hallucinate" "there," he said, "now the LLM will not make mistakes" "but the LLM can still hallucinate" said toad "that is true" said frog

Today, we're announcing A* III, a $450 million early-stage fund. We started @A_StarVC with the simple idea to be a founder's first believer. We are generalists by sector, but specialists in the stage and craft of seed investing. We partner with founders before there is consensus, before there is traction, and often before there is even a product. We are not organized around a market thesis. We back exceptional builders and follow them into the most important categories. That matters because seed investing has changed. It is more crowded, more visible, and increasingly transactional. Too often, firms use seed to secure an option and then wait for proof before investing real time and attention. Seed has become a market of access. We believe it should be a market of conviction. We built A* around a different model. We commit early. We show up before external validation, deploying both time and capital from day zero to help founders find their first customer, make an early hire, or work through the decisions that define the company. We are selective at the start and concentrated over time. We partner with a small number of founders and deepen our commitment as their companies take shape. The best outcomes come from knowing where to go deep and having the discipline to do it. This approach has led us to companies like Ramp, Decagon, Whop, Cape, Simile, Paraform, Watney Robotics, and Mercor. We're grateful to the founders who have chosen to build with us and to the limited partners who have backed us. With this fund, A* manages over $1 billion in assets less than five years after launch. Our job remains what it was on day one: back exceptional founders early and be the partner they need when it matters most.

Fast mode for Claude Opus 4.7 is now available in Cursor! It's 2.5x the speed at 6x the cost. For most tasks, we recommend using the standard speed.

@gregpr07 All you can eat for a scarce resource will never, ever work. At least, not once the VC bonfire runs out of money.

Can we pay 2k per month to get fastfast mode of codex and actually unlimited usage?

Just hang in there guys The slop rate tops out at the max output tok/s You just need to review exponentially faster wagmi

Yeah, no good answers for that. You could have something in your CI that says “num tests must be >= last time” but clearly that is dumb. So you have an agent review the other agents code. But that is that probabilistic. So add a watcher to the watcher. It’s layers upon layers of non-determinism all the way down. Good if you’re in the token selling business, I suppose.

Even when things are planned out, stuff can still end up breaking. Last night I had an agent remove a single line of code, a #[test] declaration, which caused a regression test to no longer run. A few commits later and a bug got introduced that wasnt detected and was incredibly difficult to figure out the cause, given the tests were passing.

Last night was probably one of the worst experiences I’ve had with LLMs / Agents generating code (GPT-5.5 btw) Every commit produced regressions, the agent was constantly taking shortcuts, and there was enough hallucinations to make one develop a new sense of reality. I came incredibly close to rage quitting using Agents all together. It’s times like those I wonder if we’re all being stupid letting these things loose just for some multiples of perceived productivity.

@itsyourcode @vansickn Also: burns more tokens. Vendor and User incentives, uh, bit at opposition here

@MattRogish @vansickn Totally and that's the dead giveaway because RL envs want strong verifiers which are actually pretty hard to construct without over fitting to said pathologies The less verifiable factors suffer in return

Under-discussed problem right now with most frontier coding models. A leading contributor to slop and incidental complexity and daily pain Great read @vansickn !

Supply chain attacks are happening left and right with npm, PyPI and so many other places. It seems to be getting worse, everyone agrees. But what can you do about it? Some thoughts on possible approaches (all have tradeoffs). What did I miss? And what vendors actually work?

@KingBootoshi LOL I am an idiot I kept trying to find the package "supply-chain" and couldn't. (E_NOT_ENOUGH_COFFEE) (I read it as "run npm supply-chain")

USE THE PROMPT BELOW IN CODEX/CC TO PROTECT YOUR SYSTEM AND CODEBASE FROM NPM SUPPLY CHAIN ATTACKS (LIKE TANSTACK TODAY): """ set up npm supply-chain protection on this machine. do all four steps. 1. edit ~/.npmrc. keep every existing line (auth tokens etc), append: min-release-age=7 minimum-release-age=10080 save-exact=true 2. edit ~/.bunfig.toml (create if missing). keep existing content, append: [install] minimumReleaseAge = 604800 3. in this project, open package.json and pin every dependency: strip ^ and ~ from every version under dependencies, devDependencies, and peerDependencies. exact versions only. 4. commit the lockfile (bun.lock / package-lock.json / pnpm-lock.yaml) so the resolved tree is locked in git. then report: files changed, deps pinned, anything unexpected. """ the cooldown makes every package manager refuse any version published in the last 7 days. attack chains usually only last a couple hours, but this protects you long term and for any future attacks... which at this rate will keep happening

Ha! Yes, it must've been trained over and over on "don't break things" that it has a pathological over-cautiousness. I see it in commit time, too: * LLM writes code * runs tests, they pass * "Hey human! I wrote the code, please review!" * LGTM, commit and push * "Lemme run the tests a few more times, just in case. Committed the code. Let me run the tests to be sure before I push. I'll run them one last time" It wants to run the full suite ALL THE TIME. "I made a docs change. Lemme run the tests to make sure it didn't break them" - WTF who has their markdown tested?!

@MattRogish @vansickn Dead on. It's comical the extent you resort to conventionally bad advice to get them to comply Imagine being a senior eng in 2017 telling your juniors "Never worry about backwards compatibility" "Break interfaces aggressively and update all callers"

We used to go to a special website, ask strangers for help with programming, and get humiliated in return

I really do not understand the idea that we measure AI based on length of time I feel pretty certain that with the right harness and task you could make agents run productively for months already today Difficulty of task != length What am I missing?