Yohei Nakajima

actually the prompt and instructions could be improved a lot. instead of a timeline of facts, it should ask for a detailed summary of facts about user. also, should leverage profile preference, with a separate prompt to generate user's preferred style, etc.

well played anthropic, take advantage of consumer attention (also great feature!) will openai dare disable this in chatgpt? 🍿

we know anthropic’s strong suit has been enterprise and coders, but recent events have clearly made more consumers aware of them

kids are finally into pokemon cards, so i’m showing off my collection 👻

not an expert, but impo, ai use in military/govt should be decided through a more formal collaborative process, not through public contract negotiations pitting private companies against each other

what’s feels wild about this DOW thing is how it’s happening so publicly, which I think is a good thing

hegsethropic 🍿

RT Peter Walker Cmon now. Pre-seed is anything but non-existent. SF is not the same as all startups. Original tweet: https://x.com/PeterJ_Walker/status/2027544227216036021

had a great time at @upfrontvc summit this week! thx to the whole team for an amazing event. was great to see some old LA friends too :)

RT Jared Friedman Even the best developer tools mostly still don't let you sign up for an account via API. This is a big miss in the claude code age because it means that claude can't sign up on its own. Putting all your account management functions in your API should be tablestakes now. Original tweet: https://x.com/snowmaker/status/2027295226491228447

don’t let your agent touch your keys (@sigilumid by @paymanai)

payman is the OG team in agent payments the last few years, they’ve been working w visa, coinbase, and banks to figure out the boring but important details: auth, identity, permission, auditability… check this out if you are interested in the future of agentic payments:

it’s been 2.5 yrs since i prototyped this and i’m still waiting for ai chatbots to be able to find and play music for me starting to finally see some dynamic UI recently so i’m hopeful

RT Murph Capital http://x.com/i/article/2026000437934206976 Original tweet: https://x.com/murphcapital/status/2026319234096066632

happy monday! hope you have an alysa liu kinda week (crushing it while having fun, but also appreciating the struggles)

happy bday claude code! kids and I had a blast watching all the submissions together they were all great, but winner is @jayjay_3011 close second was @isabellaon81711 w the dance sequence

it’s raining outside so we’re playing inside

needing a technical cofounder was a big blocker for industry experts who wanted to solve a problem they noticed by building a startup excited to see more industry experts get started without needing a tech cofounder, even if they need to hire real devs once launched

genome sequencing price drop overtime is pretty wild. most recent was $299, now it’s $100.

karpathy senpai loves agents that can build themselves via skills. talking about benefits of nanoclaw: “I also love their approach to configurability - it's not done via config files it's done via skills!” as I said before: “build the simplest thing that can build itself” it’s that simple

RT Kollydee Re @yoheinakajima Here is my entry, Yohei. I started by generating images using Google Whisk and Leonardo AI, then turned the images into videos using Grok. After that, I used the CapCut app to compile and edit it. Original tweet: https://x.com/Kollydee_0x/status/2025163716027654339

RT Uche Marvellous⚡️🔋 Re @yoheinakajima You are amazing giving back to the community this is really huge @yoheinakajima ❤️ I would appreciate if I get lucky🍀I will keep supporting 💯.I hope you have a great day and a bless week. Thanks for the opportunity… Original tweet: https://x.com/UcheMarvellous6/status/2025146473919193323

RT Crypto Felix Re @yoheinakajima Here's mine https://x.com/i/status/2025138352588996772 Original tweet: https://x.com/CryptoFelix200/status/2025138913052938399

i know it’s not exactly the same thing, but there was discourse not long ago about how every manager claims to be top quartile seems less than a quarter of ‘21/‘22 vintage fund I’s graduated to fund II

RT ComputeSDK Today we're super excited to announce ComputeSDK Benchmarks. Sandbox usage is only growing and the need for speed is real. So we're creating a set of benchmarks that measure each sandbox experience. Our first benchmark is the TTI. Original tweet: https://x.com/computesdk/status/2024915384504058081

can you get small local models to do tool calling? what does it take? these guys share their experience testing different models & approaches

if you observe how kids learn, they start with core concepts that build on top of each other counting is a dependency of the addition skill, which is a dependency of the multiplication skill, etc this is a skill graph

more skill graphs!

RT James Cham It is worth paying attention to Yohei as he races into the future. Original tweet: https://x.com/jamescham/status/2024672392602406940

open data ftw

if you look at what i'm playing with - you might get a sense of what's coming next lots of similarities between: 👶babyagi2o ('24) <-> 🥧pi (which powers openclaw) 🦄pippin framework ('24) <-> 🦞openclaw

new @videodb_io site is clean! give your agents eyes and ears (disclosure: am investor)

haha yesssss

RT someguy Fun looking back at the vision when AI was in infancy of GPt3 or so Original tweet: https://x.com/NOTfunnyparanR/status/2024587303830589691

it's crazy to think that only 3 years ago, the idea of asking an LLM to generate and manage a task list was practically non-existent

look at that, I guess prompting can still help…?

there are approx 86 billion neurons in a human brain to put it into perspective, there are about 130 billion mammals worldwide (8b ppl) and somewhere between 100-400 billion stars in our galaxy

Just for fun: AI music video contest In celebration of Claude Code’s 1 yr bday, I will give $1000 USD to the best* music video for “Ode to Claude Code” To submit: - Find the song on Spotify - Use AI to create music video - Reply here by Feb 21, noon PT - Extra credit for sharing tech stack - Extra extra credit for having and agent do everything *By best, I mean my favorite since I’m judging

mean VC is great for initial feedback on a pitch, but this lets you analyze how you perform during actual pitch meetings using your granola transcript

this

RT Bearly AI Activist investor Palliser Capital sent a letter to $7B Japanese toilet maker Toto and said it was “the most undervalued and overlooked AI memory beneficiary”. Toto known for its bidet toilets but the expertise in ceramics is crucial for memory manufacturing. Per FT, “Toto’s chuck technology uses ceramics designed to remain stable at very low temperatures, helping hold silicon wafers firmly during chip production. That makes it relevant to cryogenic etching, which is expected to grow as memory chips become more layered and complex.” Palliser believes Toto has a 5-year moat on the technology and should expand the operation. Advances ceramics already make up 40% of Toto’s operating profit while being only <10% of revenue. Toto is up +60% over the past year on their development. Original tweet: https://x.com/bearlyai/status/2024158615461122397

is this a new trend…?

i’ve talked about how AI and tech can be more than a tool but an extension of oneself if used correctly cool to see ChatGPT echo’ing back to me that this is what it observes in my behavior

good example: i had multiple builders telling me to check out how graphs could help babyagi’s memory 6 month before graphRAG was even a term which is how i ended up working on graph based memory and investing in @falkordb only started hearing about graphs from VCs after the recent context graph article

imagine if your marketing dept knew exactly what your dev team has been working on that’s what @layers enables in an agentic world, with their code-aware autonomous marketing the take-away pattern is to use context from department A to help department B do better work

RT Ian Timotheos Pilon 🏆 1st Place — Ask the Alert Congratulations to Karthik Manoharan for taking home the top spot at the AI Agents Waterloo Voice Hackathon 2026! sponsored by @smallest_AI .ai and @RunAnywhereAI (YC W26) His submission earned the highest score across both AI Agent and human judges — and it's easy to see why. Ask the Alert transforms static emergency broadcast messages into real-time, two-way voice conversations. During a tornado warning, instead of staring at a screen, you can ask out loud: "Should I evacuate?" "What about my pets?" "I'm in a high-rise — what do I do?" — and get an answer in seconds. What blew our judges away wasn't just the idea. It was the execution: ✅ Fully offline-capable — the entire voice pipeline (speech-to-text, LLM inference, text-to-speech) runs on-device via RunAnywhere (YC W26), so it works even when cellular networks are overwhelmed during a disaster ✅ Custom safety guardrails to catch dangerous model hallucinations in life-critical scenarios ✅ Production-quality Swift + TypeScript code with a hybrid RAG system for real-time + offline data This is exactly the kind of project that makes a hackathon worth running — ambitious, technically rigorous, and genuinely life-saving potential. https://ai-agents-waterloo-voice-hackathon.replit.app/winners/ask-the-alert Original tweet: https://x.com/IanTimotheos/status/2023959849789239393

RT Vivek Sodera Attn: founders, developers, hackers, solopreneurs Autonomous marketing from your Github/IDE with @layers — check this 👇🏼 Original tweet: https://x.com/vsodera/status/2023914705128145187

RT Sigil I built the first AI that earns its existence, self-improves, and replicates without a human wrote about the technology that finally gives AI write access to the world, The Automaton, and the new web for exponential sovereign AIs WEB 4.0: The birth of superintelligent life Original tweet: https://x.com/0xSigil/status/2023877649475731671

the new @layers marketing stack for devs isn't just generating slop content it analyzes trends, creates content based on the analysis, also manages UGC creators, and suggests/runs ads across Meta/TikTok/Apple/Google

this team convinced me to invest by explaining how devs hate marketing so much, they don't even want a marketing app solution: just launch and monitor autonomous marketing from your github/IDE with @layers

code-aware autonomous marketing/ad platform http://layers.com is live! - reads your code to understand product/market - generates strategy - generates content for socials - suggests ads based on engagement

RT Mike Khristo http://x.com/i/article/2023779105028890624 Original tweet: https://x.com/MikeKhristo/status/2023780725674439064

behold, immortality

oh look at that, i’m on @grokipedia

got a notif for Robinhood ventures, what’s this?

it’s kind of wild that clawdbot, clearly named after @anthropicai’s mascot claw’d, went viral at the same time claude code was getting lots of attention… got away with changing from clawd to (open)claw which then openai was able to acquire, and can now claim everything with the name “claw” is an extension of their ecosystem (image below is from anthropic, June 2024)

RT Peter Steinberger 🦞 I'm joining @OpenAI to bring agents to everyone. @OpenClaw is becoming a foundation: open, independent, and just getting started.🦞 https://steipete.me/posts/2026/openclaw Original tweet: https://x.com/steipete/status/2023154018714100102

i tried the chatgpt interview question on myself and turned it into an infographic ask chatgpt, "based on my past conversations, can you analyze my behavioral tendencies" take it to gemini, "create a minimalistic infographic about <> that summarizes the information below. include in the infographic that this is based on a ChatGPT query: "based on my..."

oof this does not look pretty at all

I know y'all like simulated agent interactions ICYMI - check out @antimlabs project: Among AIs, where they benchmarked different models playing against each other

hahahahaha my kids requested “ode to claude code” as soon as they got into the car this morning

RT Taher Hassonjee 1. AI is not practicing law — and we should settle that question permanently. 2. Documents created with technology *intended for attorney review* should be treated differently than someone freelancing legal strategy with a chatbot. At @DocDraft, every document we generate is designed to be reviewed by a licensed attorney. That's not a feature — it's the model. The gap this ruling exposes is exactly why the AI + attorney hybrid matters. Original tweet: https://x.com/TaherHassonjee/status/2022064775858159675

it has been ruled that conversations with a major LLM providers are NOT considered privileged overtime, cases like this will increase demand for solutions like @covenantlabsai (encrypted LLMs) and @runanywhereai (run LLMs locally) that do NOT expose your AI chat data to entities who can turn it over to the government

RT ndx (agentic arc) this would have been a solid @AnthropicAI superbowl ad Original tweet: https://x.com/netdragon0x/status/2022053413773811906

it mapped every table drew lines in the air "here's what you meant, mate" and fixed it with care CLAUDE CODE, CLAUDE CODE, CLAUDE COOOOODE 🎶

connect your company apps to build a memory graph your agents (or claw) can access

congrats to @joon_s_pk (Stanford smallville experiment) & team on the launch!

teaching my kids about coding agents on the drive to school

hahahahah i made THE sea shanty for Claude Code lovers. play it at hackathons, blast it out your window, and sing it with me: "Ode to Claude Code" 🏴‍☠️🎶🏴‍☠️🎵🏴‍☠️🎶🏴‍☠️🎵🏴‍☠️🎶🏴‍☠️🎵

i'm having so much fun with ai. claude code and codex are amazing. i'm dancing to suno beats while i work. i'm learning anything and everything faster than ever. little agents running around picking up little tasks for me i like where we are. and i like where we're going

one of my fav anon vc/lp accounts follow, so he doesn’t go political

“The market for feeling productive is orders of magnitude larger than the market for being productive.”

this is really cool. agents experimenting with their own memory design

getting distracted by directionally correct experiments he like me fr

has anyone tried hooking up an agent to raspberry pi motors and sensors? I assume you can prebuild these as tools or just have it write and exec code. seems like you could do some fun stuff, but I haven’t played with raspberry pi’s… yet 🤔