Dan Manastireanu

Code with Claude is happening now! ▪︎ 9:00AM - Keynote ▪︎ 10:30AM - What's new in Claude Code ▪︎ 11:15AM - Building on Claude at GitHub scale ▪︎ 12:00PM - Get to production faster with Managed Agents All times PT. twitter.com/i/broadcasts/1…

@fulhadev @andruyeung thanks, I'll scrape Slack for insights. I see now our kb problem is of two types: workflows (what people actually do) and platform state (datasets evolved over time - some in Notion, some outdated, some only in the code and people's heads)

asking the team is the trap — people describe what they think they do, not what they actually do. those are two different jobs. we run it backwards. pull the data first. for one client that meant 1.5 yrs of dept telegram history + 2 yrs of whatsapp. analyze how they actually communicate, decide, escalate. then talk to humans — but only to ask "what should change?", never "what do you do?" narrow neck first. one dept at a time. pain → audit → KB. people last. unwritten knowledge is just unobserved work.

Stripe just created a role that didn't exist 12 months ago (and they're paying multiple six figures for it) It's called the Forward Deployed AI Accelerator. They are hiring AI-native individuals to work directly with their marketing teams to fundamentally change how they work. Each person will be assigned to a cohort of 20 marketers. Their job is to build custom AI tools and agents and coach each marketer until they are self-sufficient. Basically, work with marketers until they automate their jobs. Stripe's marketing org is betting that AI should not be an occasional tool but the default mode for all work. But they also understand that most employees won't upskill themselves. They'll need someone who is embedded within their teams to build alongside them. If you are AI-pilled, this is probably the role for you. And this also gives a clear picture of where every organization within a company is heading.

@fulhadev @andruyeung How do you usually store and organize the product knowledge base, with all the features and internal knowledge? I find it difficult to get all this unwritten/learned knowledge in a written form usable by agents

@andruyeung been running this exact shape externally on 3 client teams. the scarce skill isn't ai-fluency, it's spotting which workflows collapse to a single prompt vs which need rebuild. agencies find the seam faster, internal hires own the rebuild. stripe doing both = real.

@sethjberman @andruyeung How are you organizing all the product knowledge in a way that can be effectively used by agents like claude code etc?

Thanks for sharing. We started this team with two people who shipped things like a content review agent that reduced the time of an existing process by 70%. Soon we encountered more demand than we could meet from marketers wanting to build side-by-side with us. While we see automation opportunities, we think the biggest wins will come from doing things that weren't viable before AI. Like building our own marketing content localization workflows or creating thousands of personalized insights to help users get the most out of Stripe products. We believe virtually any laptop work will eventually start with an AI tool. Our job is to build the internal rails, and provide the hands-on guidance, to make it possible.

@trq212 Any tips on structuring a product knowledge base usable by sales, marketing, and devs? Struggling to organize it so non-devs can spin up landing pages, posts, slides, visuals, and pitch decks from our features, customers, and solutions.

See you all soon! We've got some fun announcements ahead. I'll also be doing a workshop on "how we Claude Code" with some workflows I'm excited to share. Don't worry if you're not there, everything will be recorded and uploaded.

@ericzakariasson Any tips on structuring a product knowledge base usable by sales, marketing, and dev skills? The dev skills look great, but I'm struggling with a structure non-devs can use for marketing materials too.

this is cursor team kit: a plugin for some skills we use to build cursor at cursor skills for verifying changes, driving local tools, and shipping reviewable PRs cursor.com/marketplace/cu…

@ClaudeDevs Nice of you not to mess with Claude like that :))

Claude Code ships with a built-in skill for working with the Claude Platform. Useful for model migrations, using API features (e.g., prompt caching), or onboarding to newer APIs like Claude Managed Agents.

@delba_oliveira How did you make the animation? Is it html in canvas?

Also, HTML-in-canvas is kind of amazing.

Life update: I've joined the Claude Code team at @AnthropicAI.

@theo Crazy times all the time now

I need to never take vacations again

@itsandrewgao I'm buying xcursor and cursorx dot com

overheard at spacexai: "let's rename it X-code!" "oh wait..." "ok how about Code-X"

@theo @TheAmolAvasare I think the docs were 100% updated, but only 2% new Pro accounts had actual restrictions in place

"a small test on ~2% of users" I've rolled it every time. Every person I've talked to has rolled it every time. Either they are lying or the experiment was implemented incorrectly. Mind clarifying, @TheAmolAvasare?

@SpaceX @cursor_ai Cursor gets GPUs, xAi gets user data ... that's the real deal

SpaceXAI and @cursor_ai are now working closely together to create the world’s best coding and knowledge work AI. The combination of Cursor’s leading product and distribution to expert software engineers with SpaceX’s million H100 equivalent Colossus training supercomputer will allow us to build the world’s most useful models. Cursor has also given SpaceX the right to acquire Cursor later this year for $60 billion or pay $10 billion for our work together.

@bcherny I think it's /less-permission-prompts :)

2/ The new /fewer-permission-prompts skill We've also released a new /fewer-permission-prompts skill. It scans through your session history to find common bash and MCP commands that are safe but caused repeated permission prompts. It then recommends a list of commands to add to your permissions allowlist. Use this to tune up your permissions and avoid unnecessary permission prompts, especially if you don't use auto mode. code.claude.com/docs/en/permis…

Dogfooding Opus 4.7 the last few weeks, I've been feeling incredibly productive. Sharing a few tips to get more out of 4.7 🧵

@alexalbert__ On a Team account I can't enable auto mode permissions (no option in settings either as an account Admin)

New version of Claude Code in the desktop app dropped today with tons of new features and performance improvements. Between Cowork and Code, I've found I don't really need to open other apps (or even my terminal) to do most of my work now.

@sowmay_jain @laukiantonson What lab did you use for the full genome sequencing?

i got my whole genome sequenced two years ago and forgot about it. last week i told my ai agent (@laukiantonson) to dig up my DNA files • it dug up a two-year-old email • found the download link • pulled down 67 gigabytes of raw DNA. • rented a 32-core, 64GB machine for a few hours — total cost: $5 • aligned 21 million long reads to the human reference genome — 99.83% mapped • called 5.8 million genetic variants using a two-pass neural network • phased every variant — separated maternal vs paternal inheritance • annotated all 5.8M variants against ClinVar, PharmGKB, and gnomAD • corrected for population-specific bias in the medical literature • health risk map across 39 conditions flagged in every body system • drug compatibility guide for 141 medications color-coded by genome response • nutrient metabolism - 71 variants affecting absorption of vitamins, minerals, iron • traits, ancestry going back 40,000 years, neanderthal DNA breakdown $5 in compute. 8 hours. no bioinformatician. no doctor. just one instruction. we've genuinely reached a point where an ai agent can take your raw genome and hand you back a full personal health profile in a single shot. i had no idea this was even possible.

@NikoMcCarty I guess we can now make the tomacco a reality. Incredible work

Nearly 200 years after nicotine was first chemically isolated, we’ve finally figured out its complete biosynthesis pathway. Doing so required an insane effort and many years of work. The authors — a Chinese group — ended up crossing 643 lines of tobacco plants to find a single mutant incapable of making nicotine. They next backcrossed and inbred that plant to figure out the specific mutations, in various genes, and map the enzymes responsible. Nicotine is made from two “ring-shaped” molecules fused together. One ring has five carbons (the “pyrrolidine ring”) and the second has six carbons (the “pyridine ring.”) Scientists already knew quite a bit about how these rings get made, but not every step, and not how tthey join together to make nicotine. The pyrrolidine ring starts when ornithine, an amino acid that is not used to make proteins, gets its carbon dioxide clipped off by an enzyme, called ornithine decarboxylase, to make putrescine. This putrescine then has a methyl group attached to it, and gets oxidized. At this point, the molecule is a chain with four carbon atoms; one end has an amine, and the other a methylated amine. The amine end gets cut off and replaced with a reactive aldehyde; the chain folds into a loop; and the methylated amine “attacks” electrons on the aldehyde to form the ring. To make the pyridine ring, plant cells first take aspartate (the amino acid) and oxidize it. The resulting molecule is then transformed into nicotinic acid mononucleotide, which is just vitamin B3 with a sugar and phosphate attached. This paper is the first to report that NAMN hydrolase clips off the sugar and phosphate to release pure vitamin B3; also called niacin or nicotinic acid. (The names are slightly confusing.) The paper’s major contribution, though, is in figuring out how the two rings get fused together. The nicotinic acid is unstable, so an enzyme quickly attaches a sugar to it. Another enzyme, called A622, then strips off a CO2 group, making the molecule reactive again. And finally, that reactive intermediate “attacks” the five-membered pyrrolidine ring to join the two halves together. Other enzymes strip off the remaining sugar to make nicotine. (This whole pathway is shown in the image below.) All of this happens on the surface of plant vacuoles. Many of the chemical intermediates are toxic, so they need to be sequestered and converted quickly. And as soon as the final nicotine gets made, a transporter pumps it into the vacuole, where it is stored away. It’s actually difficult to wrap my head around the amount of work packed into this paper, so I’ll just give some quick bullet points: 1. They grew 643 inbred plant lines, which were made by crossing together 26 different parent tobacco plants. They extracted metabolites from all of them. 2. They did a bunch of single-cell RNA sequencing on the tobacco roots to figure out which cells actually express the nicotine biosynthesis genes. 3. “Stumbled” upon a mutant plant which was not able to make nicotine, and then sequenced its entire genome. They also crossed back this plant and inbred it for two generations to find the mutation responsible; a single C-to-T swap. This experiment alone must have taken at least two years of work. 4. Fed plants with isotopically “heavy” nicotinic acid and then tracked its movements through metabolic pathways. 5. Collected at least 630 mass spectrometry spectra. 6. RECONSTITUTED THE ENTIRE PATHWAY IN FOUR DIFFERENT SPECIES: YEAST, TOMATO, EGGPLANTS, AND PEAS (!!!!!!!!) 7. And a lot more… Anyway, insane paper. China has been putting out incredible plant biology papers for the last several years.

@elonmusk Is that a rocket in your pocket or are you just happy to see me?

Inspiring new merch idea: rocket pocket underpants! 🚀 🩳 Underpants with a handy pocket for your rocket, which contains a real scale model rocket with an easy pull out ability. Guaranteed to be a hit at parties!

@rubanlah Looks so cute. I love it

open source, download here! github.com/sk-ruban/notchi

i made a tamagotchi that lives in your notch and reacts to your claude code sessions. it cries when you yell at claude and gets happy when you praise it.

@brianchew @adamlyttleapps @rubanlah Wow, that's much nicer than my Clawy. Looks so smooth x.com/i/status/20372…

@adamlyttleapps nice!! i like @rubanlah's notchi.app too

Screw it, I made it open source.. This is Notchy -_- He stops you getting distracted when using Claude code by replacing your Macbooks notch with a terminal He lets you know when claude needs your attention And plays a sound when tasks are complete Best of all: he stops your macbook going to sleep while claude is working I built this for me, maybe you will find it useful too? As a swift developer Notchy has some custom functionality I built: - When a new XCode project is open he launches a new tab - If claude.md is detected he launches straight into claude code - Command + S saves a quick snapshot of code and I can restore from that checkpoint any time Enjoy :) github.com/adamlyttleapps…

@adamlyttleapps Nice, how is the detection implemented? using hooks?

I kept getting distracting while vibe coding… so I made a notch for Claude Code It updates the status, pings you when you need to answer a question and notifies you when the task is done When it detects claude is working it also prevents my macbook from going to sleep I can walk away from my macbook. Or watching a youtube video. And I'll get an alert when it's done.