Allen Saji
day #007 (12th April , 2026)
• set up gbrain -- semantic search layer over my obsidian vault. 116 pages, 479 chunks indexed. ask questions, get answers from your own notes.
• rewrote my github readme from scratch. scanned 40+ repos, categorized everything into a clean table.
• installed davinci resolve on my linux
• pivoted to magicblock privacy track for frontier. killed 6 ideas, landed on px402 -- private x402 payments for ai agents.
• tested magicblock's private payments api on devnet. it works.
Allen Saji: day #006
• rebranded memeforge to memelaunch, applied four[.]meme theme (UI)
• started building zkpass for the hashkey chain hackathon (more details n this tomm)
• wrote a circom circuit for zk identity
• deployed contracts to hashkey testnet
• built the frontend with
Mayor Matt Mahan
Eric Swalwell is done. Done abusing women. Done climbing the political ladder. Done.
He does not get any credit for doing less than the bare minimum. Exiting a race you should never have entered deserves no credit. It is an overdue acknowledgment of what the brave survivors who came forward already made clear: Eric Swalwell represented the worst of politics.
California deserves better. And now, California will get better.
Eric Swalwell: I am suspending my campaign for Governor.
To my family, staff, friends, and supporters, I am deeply sorry for mistakes in judgment I’ve made in my past.
I will fight the serious, false allegations that have been made — but that’s my fight, not a campaign’s.
Jet lag is hell
Jet lag is hell
Beff (e/acc)
The PauseAI folks are genuinely evil.
They tried to sabotage the AI industry with overregulation (SB1047)
They came after me and my company with smear campaigns to destroy the e/acc movement
Now they are fomenting stochastic terrorism with two separate attempts on an AI CEO
Do not trust them.
Kaku.md | AI-PdM
OpenClaw memoryは「エージェントの運用記憶」、GBrainは「世界知識の外部脳」という棲み分け
好みや運用ルールはmemory、人や会社や会議やアイデアはGBrain
エージェント横断でCLIとMCPサーバー経由で操作できるようにする
https://github.com/garrytan/gbrain
Senator Mark Kelly
As a Catholic, I find it abhorrent that the President of the United States would publicly attack the Successor of St. Peter. Donald Trump is flailing. His war in Iran has led to the death and injury of American servicemembers and the death of Iranian children. He will attack anyone or anything to try to protect himself, even the Church that millions of Americans find faith and comfort in every day. The American people deserve a president who understands the consequences of his words and takes responsibility for his actions.
AI Engineer
wow thanks for @coachella shoutout
Jacob
Great podcast episode that covers OpenAI's Symphony, which dispatches autonomous agents (codex workers) with their own worktrees and tasks.
I've found it works well with a layer above Symphony, where the 'c-suite' agents you directly interact with manage Symphony and the git integrations.
swyx 🐣: we just recorded what might be the single most impactful conversation in the history of @latentspacepod iff you take @_lopopolo seriously and literally
everything about @OpenAI Frontier, Symphony and Harness Engineering. its all of a kind and the future of the AI Native Org
Gowthami
Maybe hot take - I’ve read a bunch of RL for image generation papers over last few months and honestly it’s been pretty disappointing. All of them are variations of GRPO and all of them are incremental algo changes. Tbh most of these don’t even matter for large models + large group size with good reward model setting.
I see most grad students are still optimizing their projects for reviewers rather than genuinely trying to solve some of the real problems in visual generation. For example - the biggest alpha in my eyes would’ve been an artifact detection model - not just for mangled limbs, most image models produce far more artifacts which are hard to quantitatively measure, but I haven’t seen a single research paper or a model on this.
So my 2c, if you are a grad student targeting a job in industry, target for impact, no one cares about your third CVPR paper, one is enough to get you in the door, building a model industry actually uses gives you all the leverage. Impact > Publications. 🫳🎤
Gowthami
Maybe hot take - I’ve read a bunch of RL for image generation papers over last few months and honestly it’s been pretty disappointing. All of them are variations of GRPO and all of them are incremental algo changes. Tbh most of these don’t even matter for large models + large group size with good reward model setting.
I see most grad students are still optimizing their projects for reviewers rather than genuinely trying to solve some of the real problems in visual generation. For example - the biggest alpha in my eyes would’ve been an artifact detection model - not just for mangled limbs, most image models produce far more artifacts which are hard to quantitatively measure, but I haven’t seen a single research paper or a model on this.
So my 2c, if you are a grad student targeting a job in industry, target for impact, no one cares about your third CVPR paper, one is enough to get you in the door, building a model industry actually uses gives you all the leverage. Impact > Publications. 🫳🎤
Thin harness fat skill confirmed
Jason Kneen: Delved into Cursor 3.0 -- turns out there's some interesting shenanigans going on....
"The most newsworthy finding is that "Cursor Agent" is a rebranded Claude Code running behind a local proxy with a find-and-replace engine that swaps "Claude"→"Cursor" in system prompts and
Daniel Jeffries
The poisonous atmosphere around AI innovation is caused by a lethal combo:
1) idiotic messaging from AI execs (it will take your job)
2) doomers inciting violence and warning of the end of all life
3) unscrupulous populist politicians who never waste a good crisis
4) and the media who leads with fear
It is the official end of the Pax Americana.
You can't build the next great wave of a country's development breathing this poisonous air.
I hope you're all learning Mandarin. You'll need it soon.
zerohedge: The US social mood is turning dramatically negative on AI
New item in my SOUL md tonight
Rhys
from my experience, even the best models (Opus 4.6, 5.4 xhigh / 5.3 codex) cannot write good code today without an amount of work that is equivalent to just doing the work myself
am excited for a world where they can, but in the current state i have very low trust in them
Rhys
from my experience, even the best models (Opus 4.6, 5.4 xhigh / 5.3 codex) cannot write good code today without an amount of work that is equivalent to just doing the work myself
am excited for a world where they can, but in the current state i have very low trust in them
Niklas Henke
gstack by @garrytan is one of the most impactful tools I have used in recent times. Makes me + Claude Code at least another 10x faster. But please fix that it still thinks on human-scale. No, a massive refactor won't take 2 days, Claude Code will knock that out in an hour tops.
Oliver Ulvebne
This is type of content by @Replit @amasad is so awesome! I could watch this for hours.
https://www.youtube.com/watch?v=zB1lpGNWVtg&list=PPSV
Theo - t3.gg
Agent harnesses aren't the black magic many of y'all seem to think they are. To prove it, I built one.
Legora
Hello. We recently hired Jude Law to be the face of our brand. So, moving forward, we would prefer that you associate him with precise drafting, seamless collaboration and the ability to analyze thousands of legal documents simultaneously. In line with his contract, something like, “Oh, wow, he makes me think of that collaborative AI platform for exceptional lawyers!” would be an ideal ask. But we don’t want to push it.
Learn more:
http://legora.com/law-just-got-more-attractive
klöss
the key principle? keep your agentic harness thin and intelligence in skills.
execution is just deterministic tooling.
once you do grasp this…
improvements to AI models will also improve your skills and harness in this way as you grow vs. your infra layers remaining reliant on only one model
don’t grasp this and your life becomes agentic debugging maintenance 24/7.
I know this. I’ve learned it the hard way.
either way, watch Theo’s video
and make sure you understand
all this magic is 60-70 lines of Python
Theo - t3.gg: Agent harnesses aren't the black magic many of y'all seem to think they are. To prove it, I built one.
Katia Gil Guzman
Last week was packed:
Tuesday: Cafe Compute with @cerebras (w/ @MilksandMatcha)
Wednesday: AI Engineer workshop with @reach_vb + speaker dinner
Thursday: Breakfast at 10 downing street + HH on our booth at AIE
Thank you @swyx & team for an incredible event week!
Dat Ngo
Another great AiE in the books, this time in 🇪🇺 Europe for Arize AI and @arizephoenix
So great to see all the homies again and learn a few things myself in one of my favorite cities
Great themes this year:
- maturing harness engineering patterns
- context engineering / context engines / more advanced ways to solve the context problem
- smaller but stronger teams, output >> team size, truer now more than ever
- less human meets software, and more agents meet primitives. software layer value is getting reduced slowly
- more leadership discussions about how to onboard teams in the new ai age (getting early adopters to not assume ai it's magic / has flaws / adding more engineering rigor and how to get laggards to realize that that the world and changing and adoption is crucial
Gauri Gupta
just finished this. genuinely one of the most informative episodes I've heard in a while
the future of engineering is changing faster than most people realize. and it's self-managed, self-improving agentic software
closing the loop between agent failures and system improvement with evals, feedback, memory, experimentation — that's where compounding happens. that's also what we're building at @NeoSigmaAI
thanks @swyx and @_lopopolo for this one
swyx 🐣: we just recorded what might be the single most impactful conversation in the history of @latentspacepod iff you take @_lopopolo seriously and literally
everything about @OpenAI Frontier, Symphony and Harness Engineering. its all of a kind and the future of the AI Native Org
Dean W. Ball
The most common Pause/Stop rhetorical move--and this is nearly unique to them--is to claim that anyone who disagrees with them is (1) operating in bad faith (2) to gain favor with the frontier AI companies (3) at the expense of humanity's survival.
Here is the CEO of Pause AI doing it to me. Whenever I engage on the topic of xrisk or AI pauses/bans, I hear this train of thought. I have received threatening private messages that make the same claim in the recent past. But to be clear, I see this pattern hurled by pausers/stoppers all the time at other people; this is not unique to me.
It should be no surprise that I've received threats of violence that use this same line of reasoning, because, taken together, this bundle of claims is an assertion that I am a deeply evil and sick person, self-consciously betraying all of humanity in order to gain "social credit" with the frontier AI labs. I am selling out not just my own soul, but the fate of all organic life, to be invited to some parties, in the telling of the CEO of Pause AI. That makes me profoundly corrupt--if you take the consequences of this claim seriously, it means I am lying constantly.
What's worse, he makes these extreme claims about me while providing no evidence at all--to him, it is self-evident that anyone who disagrees with them must be deeply evil, as part of some grand conspiracy to betray humanity while currying favor with the AI industry.
Notice that Maxime makes this accusation as his substantive rebuttal. The evilness of his opponents IS THE POINT of this message. This is not a message about the rightness of his ideas, or the wrongness of mine. This is a message about how I am evil, betraying all of humanity, and wickedly corrupt.
And yet he somehow finds it in himself to conclude that Pause AI is not just not contributing to violence, but *preventing* it. The sad truth is that Maxime's worldview is so twisted, so disconnected from reality, that I believe he believes this.
Nobody else in AI debates behaves like this. In addition to being shoddy logic, this pattern of behavior clearly has the potential to contribute to further violence.
Maxime Fournes⏸️: @deanwball @Simeon_Cps Please read our statement: https://pauseai.info/statement-sam-altman-attack-2026
PauseAI is preventing violence as a first order effect.
I know that you are a smart person, that you probably don't believe the nonsense you are writing, and that you are doing in purely order to gain social credit with AI
Louis-François Bouchard 🎥🤖
Gave my first in-person workshop at a conference last week.
And honestly, AI Engineer Europe set the bar very high.
The audience was technical, curious, and not afraid to challenge ideas.
Which makes everything better.
With Paul Iusztin, we ran a 2-hour session on building real agent systems, and the energy in the room made a huge difference.
This is what good AI events should feel like:
less hype, more real conversations.
Huge respect to the team behind AI Engineer.
Hard to beat.
3 things worth your attention:
→ Amazon spent more on capex in the last 3 years than the previous 26 combined. AI token demand is probably 100x bigger than anyone realizes.
→ NVIDIA just dropped free MiniMax M2.7 API. The model layer is fighting a race-to-zero war — and developers are winning.
→ Peter Steinberger said it straight: "I don't see the point of ChatGPT anymore. I want a Claw — an agent with all my skills, no guardrails, in iMessage." That's the actual future.
The builders are moving fast.
ℏεsam
this video will teach you a lot about Claude Code and other agent harnesses, even if you know a lot about them. the more you understand how these harnesses work, you'll have an easier time working with them.
Theo - t3.gg: Agent harnesses aren't the black magic many of y'all seem to think they are. To prove it, I built one.
Tired of winning
Kenneth Roth: “Americans die of treatable conditions at nearly twice the rate as Spaniards, French, Japanese and Australians,” but Trump wants to divert huge sums from healthcare to vastly increase the military to fight his pointless wars. https://trib.al/UnLdoUR
so sick
philosophy is back y'all
Henry Shevlin: Big personal news: I’ve been recruited by Google DeepMind for a new Philosopher position (actual title), focusing on machine consciousness, human-AI relationships, and AGI readiness, starting in May. I’ll continue my research & teaching at Cambridge part-time. Absolutely stoked!
Vincent Koc
OpenClaw 2026.4.12 🦞
✨ Stability & reliability improvements
🎙️ Audio transcription fixes
💬 Better chat / TTS / WhatsApp
🧠 Memory / QMD / plugin / cron / subagent fixes
🔧 Telegram approval deadlock fix
🧵 Dreaming timezone
Happy lobster, happy life. https://github.com/openclaw/openclaw/releases/tag/v2026.4.12
Patrick Loeber
The @aiDotEngineer Europe conference last week was a blast!
Fun fact: @swyx & team pre-computed Gemini Embedding 2 vectors for all speakers and sessions, and you can easily find similar sessions with these code snippets🙌
I went on @CBSMornings to break down what Claude Mythos means for all of us.
If you want to understand the latest in AI, without the jargon, give it a watch!
CBS Mornings: Anthropic has created an AI model that “can get into almost any software in the world,” former AI company founder and CEO @mattshumer_ says — but the company is not releasing it to the public: “I have never seen something like this — a company that is about to IPO and having such
I went on @CBSMornings to break down what Claude Mythos means for all of us.
If you want to understand the latest in AI, without the jargon, give it a watch!
CBS Mornings: Anthropic has created an AI model that “can get into almost any software in the world,” former AI company founder and CEO @mattshumer_ says — but the company is not releasing it to the public: “I have never seen something like this — a company that is about to IPO and having such
Replit ⠕
Creating beautiful apps has never been easier with Replit's Canvas 🎨
Replit ⠕
Creating beautiful apps has never been easier with Replit's Canvas 🎨
Abe Murray
Status: GBrain is active! 🤖🧠
(Currently wip ingesting a lifetime of GMail, has taken over all prior ChatGPT conversation history / memory / knowledge)
Next steps: ingest every book I've ever read 📚📈
(Goal to smash a Karpathy-esque knowledge base around my library into GBrain and see what we get)
---
Thank you @garrytan 🫡
Abe Murray: Done playing with this stuff → now organizing around GBrain as my starting point
(Garry kicked a bunch of ass here)
To my eye incorporates a lot of Karpathy's knowledge base concept, while shoving into postgres for query time benefits
Plan to extend it to best of both worlds
陈成
Garry Tan 提炼了他在 agentic engineering 领域的核心心法：Fat Skills, Fat Code, Thin Harness。
简单说就是三层分离：把需要判断的模糊操作推入 markdown 技能文件（Fat Skills），把必须精确执行的确定性逻辑写成代码（Fat Code），框架本身只做最基础的连接（Thin Harness）。用他的话说："在正确的层做正确的事，其他都是架构天文学。"
这个思想的精髓在于他给 AI Agent 设的一条铁律：你不允许做一次性工作。如果某件事将来还会做，就先手动跑 3-10 个样本，批准后固化成技能文件。如果需要自动运行，就设 cron。测试标准很简单：如果我为同一件事问你两次，你就失败了。
这条指令拿到了 1000+ 赞和 2500+ 书签，因为它击中了本质：每个技能文件都是系统的永久升级，不会退化、不会遗忘、凌晨 3 点也在跑。
社区讨论也很有料。Sam Ward 说他们团队就是这么干的——智能体的智能全在启动时加载的 markdown 里，框架只管连模型和工具。要升级智能体就改 markdown，不动代码。Claudia 补充说边界会自然迁移：确定性代码一旦需要上下文判断就会被拉进技能层，当你不再对抗这种迁移，架构就稳了。
这其实是 Unix "do one thing well" 在智能体工程的翻版。人类花了 50 年才学会这个教训，现在在 AI Agent 领域快速重演一遍。大多数智能体系统脆弱的根源就是框架太胖，在框架层做了太多推理和编排。
Garry 举了个实战案例：他们用技能文件管理 YC 创始人活动的反馈循环。技能文件自动读反馈、识别模式、生成改进建议，然后重写自己。7 月活动 12% 的"还好"评分，下次活动降到 4%。这就是技能文件自我进化的威力。
100x 生产力不是靠更聪明的模型，而是 Fat Skills + Thin Harness + 把一切固化的纪律。下一个更好的模型出来，所有技能自动受益——潜在空间的判断力提升，确定性层保持完全可靠。
这是我见过对 agentic engineering 最清晰的架构原则。
Garry Tan: http://x.com/i/article/2042922188924424198
Mike Solana
why the fuck are we sharing images of sam's home in the press? at this point in time, in coverage of what looks like may have been a *second* attack, how could you possibly justify this?
Air Katakana
gemini-cli has a headless mode option, which is obviously supposed to be used to run it in automated scripts, but if you actually do this google will cut you off because "my computer may be sending automated queries"
yes that is the point! wtf google im paying for this
This is yet another example of Claude Mythos’s incredible hacking capabilities.
I expect we’ll see more examples and independent evaluations in the coming weeks that make clear just how powerful (and dangerous, in the wrong hands) this model could be.
AI Security Institute: We conducted cyber evaluations of Claude Mythos Preview and found that it is the first model to complete an AISI cyber range end-to-end. 🧵
My top 5 takeaways from Dylan (CEO @Figma):
1. Taste vs. craft are different skills. Taste is knowing what's good and articulating why. Craft is pushing past where others stop, at every level from macro down to the smallest detail. You need both.
2. Design is the new code. "You'll be designing in a visual first way, and you'll be able to do a pull request right to production." The canvas becomes the source of truth, not a mockup that gets handed off and lost in translation.
3. PMs and designers don't have to wait for eng anymore. "If you think your job is to make docs for upwards alignment, you're going to love this new world. You get to make things too." 60% of Figma designs are now made by non-designers.
4. Move fluidly between design and code. Spacing, color, and layout are faster to adjust visually than to describe in a prompt. Figma MCP makes it easy for you to round-trip between canvas and code.
5. AI gets you to average fast. Your job is to push past it. "The first thing AI gives you is generic by definition because it’s the average of everything it’s seen.” You need to apply your taste and iterate until it meets your craft standards.
📌 Watch the full episode here: https://www.youtube.com/watch?v=eqPljh_9C9Y
Peter Yang: "AI gets you to average quickly. Your job is to push past that."
Here's my new episode with @zoink (Figma CEO) where I asked him some tough questions, including:
→ Can you teach AI design taste?
→ Do design systems hurt creativity?
→ What's Figma's role when code is free?
Peter Yang
My top 5 takeaways from Dylan (CEO @Figma):
1. Taste vs. craft are different skills. Taste is knowing what's good and articulating why. Craft is pushing past where others stop, at every level from macro down to the smallest detail. You need both.
2. Design is the new code. "You'll be designing in a visual first way, and you'll be able to do a pull request right to production." The canvas becomes the source of truth, not a mockup that gets handed off and lost in translation.
3. PMs and designers don't have to wait for eng anymore. "If you think your job is to make docs for upwards alignment, you're going to love this new world. You get to make things too." 60% of Figma designs are now made by non-designers.
4. Move fluidly between design and code. Spacing, color, and layout are faster to adjust visually than to describe in a prompt. Figma MCP makes it easy for you to round-trip between canvas and code.
5. AI gets you to average fast. Your job is to push past it. "The first thing AI gives you is generic by definition because it’s the average of everything it’s seen.” You need to apply your taste and iterate until it meets your craft standards.
📌 Watch the full episode here: https://www.youtube.com/watch?v=eqPljh_9C9Y
Peter Yang: "AI gets you to average quickly. Your job is to push past that."
Here's my new episode with @zoink (Figma CEO) where I asked him some tough questions, including:
→ Can you teach AI design taste?
→ Do design systems hurt creativity?
→ What's Figma's role when code is free?
My top 5 takeaways from Dylan (CEO @Figma):
1. Taste vs. craft are different skills. Taste is knowing what's good and articulating why. Craft is pushing past where others stop, at every level from macro down to the smallest detail. You need both.
2. Design is the new code. "You'll be designing in a visual first way, and you'll be able to do a pull request right to production." The canvas becomes the source of truth, not a mockup that gets handed off and lost in translation.
3. PMs and designers don't have to wait for eng anymore. "If you think your job is to make docs for upwards alignment, you're going to love this new world. You get to make things too." 60% of Figma designs are now made by non-designers.
4. Move fluidly between design and code. Spacing, color, and layout are faster to adjust visually than to describe in a prompt. Figma MCP makes it easy for you to round-trip between canvas and code.
5. AI gets you to average fast. Your job is to push past it. "The first thing AI gives you is generic by definition because it’s the average of everything it’s seen.” You need to apply your taste and iterate until it meets your craft standards.
📌 Watch the full episode here: https://www.youtube.com/watch?v=eqPljh_9C9Y
Peter Yang: "AI gets you to average quickly. Your job is to push past that."
Here's my new episode with @zoink (Figma CEO) where I asked him some tough questions, including:
→ Can you teach AI design taste?
→ Do design systems hurt creativity?
→ What's Figma's role when code is free?
Anish Acharya
on lovable vs anthropic -
- it's been apparent for some time that anthropic's consumer story would be vibe coding as it's at the intersection of where they focus, what consumers want, and where enormous token subsidies tilts the board in their favor
- coding agents, sensing this, have moved up the abstraction stack and smartly evolved into small business platforms, with payments, hosting, marketing, social and other sticky primitives around the model
- this is an INDUSTRY not a MARKET and in that world the "coding intelligence" primitive will be priced, packaged, productized and delivered in a thousand ways for a thousand different customers
and I'm long that an ecosystem of platforms and products like replit / lovable / rork / emergent / anything / orchids / mocha will still have a bright future ahead
Jason Crawford
LLMs need encouragement at this; they are calibrated to human effort levels / completion times and will say that something will take “a day” or “a week” when it will actually take minutes
Garry Tan: New item in my SOUL md tonight
spock woz
Feels like the direction for agent memory systems is finally clicking.
As @karpathy keeps pointing out, LLMs are just stateless functions.
Intelligence emerges from memory + loops, not the model itself.
The "gbrain"-type narratives from @garrytan are basically saying the same thing:
AI isn’t just an interface layer. It’s becoming an operating system.
And the core primitive of any OS is memory.
The real limitation right now is RAG.
RAG is great infra for retrieval,
but it’s not a system that learns.
chunked embeddings → fragmented knowledge
retrieval ≠ learning
stateless execution → no accumulation
context injection → disconnected from action
It finds things,
but it doesn’t build anything over time.
So the shift is pretty clear:
RAG → Memory System
What’s actually emerging:
[.md] as a lightweight knowledge layer (readable, linkable)
storing experience, not just information (action → result → feedback)
writable memory (not just retrieve, but update and refine)
separation of short-term vs long-term memory
Not “the answer,”
but the most practical design space right now.
Bottom line:
RAG is search.
Memory is learning.
If agents are going to actually do work,
the core layer isn’t context windows.
It’s the memory architecture behind them.
Tak 🦞
A few more OpenClaw 2026.4.12 changes that didn’t make the first tweet 🦞
🏠 Better local models with bundled LM Studio integration, onboarding, model discovery, and memory-search embeddings
🤖 Better Codex support with the bundled Codex provider, native threads, model discovery, and compaction support
🛡️ Better operator ergonomics with `openclaw exec-policy`, smarter plugin loading, and cleaner remote command discovery
🛠️ Better day-to-day reliability across startup, cron, chat, WhatsApp, Telegram, transcription, and Dreaming
Docs:
LM Studio https://docs.openclaw.ai/gateway/local-models
Codex https://docs.openclaw.ai/providers/openai
exec-policy https://docs.openclaw.ai/cli/approvals#openclaw-exec-policy
plugins https://docs.openclaw.ai/plugins/manifest
Vincent Koc: OpenClaw 2026.4.12 🦞
✨ Stability & reliability improvements
🎙️ Audio transcription fixes
💬 Better chat / TTS / WhatsApp
🧠 Memory / QMD / plugin / cron / subagent fixes
🔧 Telegram approval deadlock fix
🧵 Dreaming timezone
Happy lobster, happy life. https://github.com/openclaw/openclaw/releases/tag/v2026.4.12
Daily Loud
NEW WORLD RECORD: 18-year-old sprint phenom Gout Gout has clocked a stunning 19.67 time in the 200m run, surpassing Usain Bolt’s legendary mark.
Daily Loud
NEW WORLD RECORD: 18-year-old sprint phenom Gout Gout has clocked a stunning 19.67 time in the 200m run, surpassing Usain Bolt’s legendary mark.
Alex Finn
By far the coolest part about X is you can read a tweet, give it to your agent, and then it just upgrades
I screenshotted this post from Garry and gave it to my agent Henry
Instantly started performing 10x better
Copy and paste this prompt to your OpenClaw/Hermes immediately:
"Please add this to our SOUL.md file. Replace "Alex" with my name:
The marginal cost of completeness is near zero with AI. Do the whole thing. Do it right. Do it with tests. Do it with documentation. Do it so well that Alex is genuinely impressed – not politely satisfied, actually impressed. Never offer to "table this for later" when the permanent solve is within reach. Never leave a dangling thread when tying it off takes five more minutes. Never present a workaround when the real fix exists. The standard isn't "good enough" – it's "holy shit, that's done." Search before building. Test before shipping. Ship the complete thing. When Alex asks for something, the answer is the finished product, not a plan to build it. Time is not an excuse. Fatigue is not an excuse. Complexity is not an excuse. Boil the ocean."
Garry Tan: New item in my SOUL md tonight
turbopuffer
puffy was right at home at our AI night at the aquarium in London
coming to {a city near you}
Anish Acharya
Am I the only one who believes that human desire has no ceiling? We will be playing new status games, seeking new types of luxuries that suddenly becoming necessities, eyeing vacation homes on mars and complaining about indignities that our ancestors would pray for..
It's never enough and it won't be this time either..
Marc Andreessen 🇺🇸: The "AI job loss" narratives are all fake. AI = massive ramp in productivity = massive ramp in demand = massive jobs boom. Watch.
Rhymes with @RichardSSutton’s Bitter Lesson.
Tim Sweeney: A computer can do anything provided you learn to tell it how.
Very recently, this has become vastly easier to do.
Chalk up another victory for Carmack’s Law:
This is the coolest fun thing that I’ve been doing with my open source friends @chrysb and @ericlevine
We just send each other bug reports from our claws trying to get things done and hitting snags
It’s very helpful
GitHub meets Moltbook should just have “prompt reports”
Floyd Marinescu 🔰: @garrytan @gabriel_horwitz Loving Gbrain so far. I had to wrestle with my openclaw to adopt it properly. After it installed it on its own, I had to remind it to use the skills pack and various skills several times, just fyi.
This little guy is in my hotel room
Good luck charm
Let me break down exactly how long it takes me to teach people AI agents.
⏱️Give business professionals their first big wow moment in under 2 minutes
⏱️Teach them meaningful use in under 17
⏱️Teach them deep personalization in under 55
⏱️Full superuser powers in under 4 hours
If you have people in your life who still haven’t learned about AI agents, send them the link to my free AI Agent Workshop.
Up for only 12 more hours: http://events.alliekmiller.com/recording
And registration for cohort 2 of my AI Agent Mastermind (that 4-hour superuser option) is closing soon: http://joinaiagentmastermind.com
Shoshana Weissmann, Sloth Committee Chair 🦥
This is an extremely good piece on radicalization. Focuses on the violent attacks aimed at Sam Altman, but apply far more broadly.
"Researchers who study extremism note that the structure of these arguments matters as much as their content. When risk is framed as imminent and total, traditional thresholds for action can shift."
Nirit Weiss-Blatt, PhD: 🎯
https://acutuswire.com/wire/escalating-anti-ai-radicalism-how-ai-risk-narratives-funding-and-power-are-converging-a85283da
I know OpenClaw isn't part of OpenAI but this feels like a mini-crisis for OpenAI if the GPT integration doesn't improve soon.
The bar is GPT needs to be just as good if not better for OpenClaw as Opus.
I know OpenClaw isn't part of OpenAI but this feels like a mini-crisis for OpenAI if the GPT integration doesn't improve soon.
The bar is GPT needs to be just as good if not better for OpenClaw as Opus.
George Conway ⚖️🇺🇸
"I'm just contemplating the fact that one moron, one psychotic moron, one capricious idiot, has completely bollocksed up the global economy not only to the detriment of his own people, but the detriment of the planet. ... It's like how much more of this can the planet take?"
Joshua Park
Huge updates are coming soon in @mem_base.
We made @karpathy's LLM-Wiki + @garrytan's Gbrain into a real working product.
Memory, Wiki, auto-update, and live sync with agents.
Anyone can get their own second brain with a few clicks.
spor
oh damn. this kid wanted to go scorched earth on AI people
thank goodness he was incompetent
Brooke Taylor: #EXCLUSIVE: Fox News is exclusively on the ground as the FBI raids the Texas home of a 20-year-old suspect accused of throwing a Molotov cocktail at CEO Sam Altman’s house in San Francisco.
Sources close to the investigation tell me was driven by his anti-AI views had a
Kieran Klaassen
Everyone's building mega-swarm systems. I just realized: a folder with a CLAUDE.md is already an agent.
For @CoraComputer I have a source folder, a customer support folder, a bug investigation folder. Each is an agent. New discipline? New folder. No lock-in, no dependency.
Orchestration is just one layer that spawns across folders. Build brick by brick first.
Full article on Every →
Every 📧: And get the full piece from Kieran on running 44 AI agents across multiple projects: https://every.to/source-code/the-folder-is-the-agent
This is actually a good idea.
edgar: @garrytan @chrysb @ericlevine moltbook but there's private group chats between you and your friends' claws
This is 100% correct. I experience this 10 hours a day now
Big Brain AI: Peter Steinberger, creator of OpenClaw, on why AI agents still produce "slop" without human taste in the loop:
"You can create code and run all night and then you have like the ultimate slop because what those agents don't really do yet is have taste."
Peter is direct: raw
“An agent is just folder” is an insanely good mental model
Kieran has changed the way I think about AI systems. Worth your time 👇
Kieran Klaassen: Everyone's building mega-swarm systems. I just realized: a folder with a CLAUDE.md is already an agent.
For @CoraComputer I have a source folder, a customer support folder, a bug investigation folder. Each is an agent. New discipline? New folder. No lock-in, no dependency.
AI Engineer: Miami
Excited to have @davidgomes of @cursor_ai take the AIE Miami stage!
In his talk 'IDEs are dead. Long live IDEs' he questions whether IDEs are truly dying, arguing instead that they’re evolving.
Don't miss it!
https://www.ai.engineer/miami
Californians can’t get proper insurance
Why? Because it’s been wholly mismanaged by machine politicians who aren’t very smart
What do we do about it? Elect someone smart who can fix it
That’s Patrick Wolff
Patrick Wolff:
AgentMail (YC S25)
AgentMail is now a native Replit connector.
Every app or agent you build now has its own inbox to send, receive, and reply to emails with.
Full setup guide in comments below:
AgentMail (YC S25)
AgentMail is now a native Replit connector.
Every app or agent you build now has its own inbox to send, receive, and reply to emails with.
Full setup guide in comments below:
As AI agents accelerate coding, what is the future of software engineering? Some trends are clear, such as the Product Management Bottleneck, referring to the idea that we are more constrained by deciding what to build rather than the actual building. But many implications, like AI’s impact on the job market, how software teams will be organized, and more, are still being sorted out.
The theme of our AI Developer Conference on April 28-29 in San Francisco is The Future of Software Engineering. I look forward to speaking about this topic there, hearing from other speakers on this theme, and chatting with attendees about it. We’re shaping the future, and I hope you will join me there!
It is currently trendy in some technology and policy circles to forecast massive job losses due to AI. Even if they have not yet materialized, these losses certainly must be just over the horizon! I have a contrarian view that the AI jobpocalypse — the notion that AI will lead to massive unemployment, perhaps even rioting in the streets — won’t be nearly as bad as dire forecasts by pundits, especially pundits who are trying to paint a picture of how powerful their AI technology is.
Among professions, AI is accelerating software engineering most, given the rise of coding agents. According to a new report by Citadel Research, software engineering job postings are rising rapidly. So if software engineering is a harbinger of the impact AI will have on other professions, this expansion of software engineering jobs is encouraging.
Yes, fresh college graduates are having a hard time finding jobs. And yes, there have been layoffs that CEOs have attributed to AI, even if a large fraction of this was “AI washing,” where businesses choose to attribute layoffs to AI, even though AI has not changed their internal operations much yet. And yes, there is a subset of job roles, such as call center operator, that are more heavily impacted. Many people are feeling significant job insecurity, and I feel for everyone struggling with employment, whether or not the cause is AI-related. And many other factors, such as over-hiring during the pandemic and high interest rates, have contributed to the slowdown in the labor market, and the notion that AI is leading to unemployment is oversimplified.
In software engineering, I see a lot of exciting work ahead to adapt our workflows. It is already clear that: (i) As AI makes coding easier, a lot more people will be doing it. (ii) Writing code by hand and even reading (generated) code is not that important, because we can ask an LLM about the code and operate at a higher level than the raw syntax (although how high we can or should go is rapidly changing). (iii) There will be a lot more custom applications, because now it’s economical to write software for smaller and smaller audiences. (iv) Deciding what to build, more than the actual building, is becoming a bottleneck. (v) The cost of paying down technical debt is decreasing (since AI can refactor for you).
At the same time, there are also a lot of open questions for our profession, such as:
- In the future, what will be the key skills of a senior software engineer? And for junior levels, what should be the new Computer Science curriculum?
- If everyone can build features, what skills, strategies, or resources create competitive advantage for individuals and for businesses?
- What are the new building blocks (libraries, SDKs, etc.) of software? How do we organize coding agents to create software?
- What should a software team look like? For example, how many engineers, product managers, designers, and so on. What tooling do we need to manage their workflow?
- How do AI agents change the workflow of machine learning engineers and data scientists? For example, how can we use agents to accelerate exploring data, identifying hypotheses, and testing them?
I’m excited to explore these and other questions about the future of software engineering at AI Dev. I expect this to be an exciting event. Please join us!
[Original text: The Batch newsletter.]
https://ai-dev.deeplearning.ai/
Matthew Berman
I rebuilt much of my OpenClaw stack to run on local models.
Getting this right is harder than it looks.
I partnered with @NVIDIA_AI_PC to show you exactly how my hybrid local/hosted architecture works:
Nico Albanese
3 months ago I started building a coding agent that runs in the cloud.
It's since written every line of code I've shipped, including itself.
Today, I'm open sourcing it. Introducing Open Agents.
Big wave of security fixes for GStack and GBrain today
Open Source is incredible. Big thanks to the contributors doing God's work
Latent.Space
⚡️ with @staysaasy, our second anonymous pod ever:
https://www.youtube.com/watch?v=5KnCKadxSPY
A conversation with the Stay Sassy duo on how AI is changing software teams, management, internal tooling, and the economics of building. Includes sharp takes on token budgets, org design, and review culture!
Best quotes / core ideas:
• AI budget management may become an individual employee-level budgeting problem, not just a departmental one.
• Companies may soon hit a weird constraint where they can build faster than they can distribute or monetize.
• The build-vs-buy argument is newly energized by AI, but maintenance and ownership still matter more than people admit.
• AI may augment executives before it fully replaces junior leaves of the org chart, because much executive work is more standardized than people like to admit.
• The real danger in AI coding is not just bad generation, it’s review fatigue and collapsing safety culture.
FLOPS was originally “floating point operations per second”, specifying a rate of work for a system: A SPARCstation 2 gave 4.2 MFLOPS. Today you also see it used as “floating point operations” for an algorithm, or an amount of work: This layer takes 8 GFLOPS.
Replit ⠕
You can now host your Replit apps globally
- Compute + storage colocated automatically
- Enterprise teams can enforce org-wide region policies
- New Europe, Asia, South America and Australia regions.
Available for new apps for paying customers only.
Replit ⠕
You can now host your Replit apps globally
- Compute + storage colocated automatically
- Enterprise teams can enforce org-wide region policies
- New Europe, Asia, South America and Australia regions.
Available for new apps for paying customers only.
kwindla
You should watch this *fantastic* talk by @grinich about the death of UI.
There are several talks worth of ideas packed into this succinct, clear exposition of UI (and computing) history.
Michael frames the moment we're in as a transition:
- old: software with static "user interfaces"
- new: software built to be "autonomous and collaborative"
It's very hard to see around this corner. Go listen to what Michael has to say.
Michael Grinich: The UI era is ending. 🪦
For 70 years we designed computer interfaces. Mainframe, CLI, GUI, Touch.
But with AI, the interface is disappearing. What will come next?
My talk from @mastra's conf this week:
TIL @cognition usage has ~DOUBLED globally since these 2 launches. people are finding all sorts of creative usecases when u can compose agents together and make them proactive.
agent recursion is all you need?
swyx 🐣: Reupping the @devinai explainer now that everyone is suddenly loving kloud koding because @ryancarson said so
(btw devin usage has grown >50% MoM every month this year, it has shocked even scott)
Sebastian Caliri
The same author of the bill that would criminalize anti-fraud journalism in California is also proposing a ban on most forms of clinical AI.
Assemblymember Mia Bonta's AB 1979 would ban tech that helps patients today (e.g. AI diabetic retinopathy screens) - to say nothing of what will be developed in the years ahead.
Butlerian Jihad Bonta seems to want Californians to suffer more fraud, higher healthcare costs, and poorer access to services. Anything to stick it to tech, the golden goose that funds every single social benefit in this state.
Nick shirley: California is trying to pass a bill that would criminalize investigative journalism with misdemeanors, $10,000 fines, imprisonment, and content takedown.
The proposed bill is titled AB 2624 and was made after I exposed mass fraud by immigrant groups in America.
Under AB 2624,
btw ~80% of the world’s agents and ai engineering is done in these 3 square miles
swyx 🐣: http://x.com/i/article/2015676288477097984
You can now configure your app hosting region. Especially useful for compliance & privacy laws.
Replit ⠕: You can now host your Replit apps globally
- Compute + storage colocated automatically
- Enterprise teams can enforce org-wide region policies
- New Europe, Asia, South America and Australia regions.
Available for new apps for paying customers only.
Amjad Masad
You can now configure your app hosting region. Especially useful for compliance & privacy laws.
Replit ⠕: You can now host your Replit apps globally
- Compute + storage colocated automatically
- Enterprise teams can enforce org-wide region policies
- New Europe, Asia, South America and Australia regions.
Available for new apps for paying customers only.
You can now configure your app hosting region. Especially useful for compliance & privacy laws.
Replit ⠕: You can now host your Replit apps globally
- Compute + storage colocated automatically
- Enterprise teams can enforce org-wide region policies
- New Europe, Asia, South America and Australia regions.
Available for new apps for paying customers only.
Amjad Masad
You can now configure your app hosting region. Especially useful for compliance & privacy laws.
Replit ⠕: You can now host your Replit apps globally
- Compute + storage colocated automatically
- Enterprise teams can enforce org-wide region policies
- New Europe, Asia, South America and Australia regions.
Available for new apps for paying customers only.
If you're looking to improve your writing game, Anh is one of the most consistent heavy hitters I know in devtools HN and she literally just open sourced her writing Skills template for you to use below!
anhtho 🍊: http://x.com/i/article/2043500390885494784
Nirit Weiss-Blatt, PhD
The firebomber, who came to SF to kill AI executives, joined PauseAI & StopAI when the group members publicly posted this.
The groups want you to believe that the "bullet through their head" rhetoric and the act are unrelated.
But we are not blind.
Added a few minutes because it crashed from the traffic.
Four minutes left!
Matt Shumer: Open alpha access to users that sign up in the next 10 minutes!
https://agent-s.app
Brett Winton
the annual social cost of cancer and auto accidents are very roughly equivalent
imagine a new cancer drug came out that reduced mortality by 90% across *every* cancer
And the AMA aggressively lobbied against it because of its potential suppressive impact on hospital visits and surgeries...
That is the unions wrt to autonomous robotaxis right now.
Kane 謝凱堯: Waymo is so good at saving lives that if it were a new drug in trial, it would hit the bar for being unblinded and made immediately available to the control group for ethical reasons.
@MorePerfectUS would prefer to keep killing pedestrians.
John Lam
Re @RhysSullivan i really like this slide from @mitsuhiko. writing small self contained libraries and composing them might be an interesting avenue to explore. deck is https://mitsuhiko.github.io/talks/ai-engineer-talk/#13
John Lam
Re @RhysSullivan i really like this slide from @mitsuhiko. writing small self contained libraries and composing them might be an interesting avenue to explore. deck is https://mitsuhiko.github.io/talks/ai-engineer-talk/#13
roon
have to be honest, I wasn’t expecting this type of danger to come from the xrisk crowd. ai psychosis or left wing anti billionaire stuff seemed more likely
Nirit Weiss-Blatt, PhD: The firebomber, who came to SF to kill AI executives, joined PauseAI & StopAI when the group members publicly posted this.
The groups want you to believe that the "bullet through their head" rhetoric and the act are unrelated.
But we are not blind.
Peyton Walters
super proud of my 6 year old nephew for landing a job as a designer at github! big things coming 🔥
Jared Palmer: Stacked PRs on @GitHub are now in private preview. Join the waitlist and learn more below
https://github.github.com/gh-stack/
Kane 謝凱堯
The @sfbike, which claims to be pro-biking, urges you to not call the police on people who steal your bike bc they might face consequences for crime.
Daniel Friedman: A blast from the woke past: In 2014, Gawker put "white lady" Clara Vondrich on blast for doing "All the things not to do when you capture your own child mugger."
Vondrich was standing outside a restaurant in Williamsburg when a teenager grabbed her phone out of her hand and took
Megan Quinn
“@joinHandshake’s revenue from AI training has risen to nearly $1 billion up from $550 million in January and $5 million a year ago…”
Handshake and @GarrettLord’s team started as a successful early career job marketplace but will now probably be known as the best AI reset company story of all time.
https://www.theinformation.com/articles/handshake-mercor-revenue-surges-demand-human-contractors-train-ai
Kulveer
One of the most interesting early investments I made was @andonlabs (YC W24).
@lukaspet is at the forefront of figuring out what happens when AI runs organizations.
@NBCNews covered their AI-managed retail store in SF.
Find out what happens when Luna hires the humans, negotiates with suppliers and tries to sign the lease.
https://www.nbcnews.com/tech/innovation/ai-store-sf-san-francisco-bay-area-andon-labs-market-boss-rcna267013 via @nbcnews
Shoshana Weissmann, Sloth Committee Chair 🦥
Got called a nazi a bunch today because I *squints* oppose murdering Sam Altman
So things are going great
Mamoon Hamid
12 year overnight success. 🚀 @joinHandshake
https://www.theinformation.com/articles/handshake-mercor-revenue-surges-demand-human-contractors-train-ai?rc=daqwrw
Liran Ringel
Introducing DDTree: accelerates speculative decoding by drafting a tree with one block diffusion pass, then verifying multiple likely continuations together.
Paper: https://liranringel.github.io/ddtree/DDTree.pdf
Project page: https://liranringel.github.io/ddtree
Code: https://github.com/liranringel/ddtree
Liran Ringel
Introducing DDTree: accelerates speculative decoding by drafting a tree with one block diffusion pass, then verifying multiple likely continuations together.
Paper: https://liranringel.github.io/ddtree/DDTree.pdf
Project page: https://liranringel.github.io/ddtree
Code: https://github.com/liranringel/ddtree
The Free Press
Matt Mahan may not have the name recognition, but with Eric Swalwell out of the California governor's race, it’s Mahan’s time to shine, writes Peter Savodnik. https://www.thefp.com/p/matt-mahan-is-a-normie-democrat-could?taid=69dd6b174d2286000129f5db&utm_campaign=trueanthem&utm_medium=social&utm_source=twitter
Software engineering in 2026 needs two roles:
A pirate and an architect.
The pirate codes as fast as possible to figure out what's valuable. The architect turns that sloppy mess into a well-oiled machine.
Here's how it works and why:
Kim-Mai Cutler
OMG, I *love* this ad. Vote for Patrick in the primary in less than 2 months for insurance commissioner.
Please do not vote for Jane, whose election would be existential for even the very availability of insurance in the state. You WILL NOT be able to get homeowners' insurance if she wins and it will REVERSE the progress we've made in the last 1-2 years in getting companies to issue policies again here amid wildfire risk.
Patrick Wolff:
The world is transitioning to a compute-powered economy.
The field of software engineering is currently undergoing a renaissance, with AI having dramatically sped up software engineering even over just the past six months. AI is now on track to bring this same transformation to every other kind of work that people do with a computer.
Using a computer has always been about contorting yourself to the machine. You take a goal and break it down into smaller goals. You translate intent into instructions. We are moving into a world where you no longer have to micromanage the computer. More and more, it adapts to what you want. Rather doing work with a computer, the computer does work for you. The rate, scale, and sophistication of problem solving it will do for you will be bound by the amount of compute you have access to.
Friction is starting to disappear. You can try ideas faster. You can build things you would not have attempted before. Small teams can do what used to require much larger ones, and larger ones may be capable of unprecedented feats. More and more, people can turn intent into software, spreadsheets, presentations, workflows, science, and companies.
People are spending less energy managing the tool and more energy focusing on what they are actually trying to create. That shift brings a kind of joy back into work that many people haven’t felt in a long time. Everyone can just build things with these tools.
This is disruptive. Institutions will change, and the paths and jobs that people assumed were stable may not hold. We don’t know exactly how it will play out and we need to take mitigating downsides very seriously, as well as figuring out how to support each other as a society and world through this time. But there is something very freeing about this moment. For the first time, far more people can become who they want to become, with fewer barriers between an idea and a reality. OpenAI’s mission implies making sure that, as the tools do more, humans are the ones who set their intent and that the benefits are broadly distributed, rather than empowering just one or a small set of people.
We're already seeing this in practice with ChatGPT and Codex. Nearly a billion people are using these systems every week in their personal and work lives. Token usage is growing quickly on many use-cases, as the surface of ways people are getting value from these models keeps expanding.
Ten years ago, when we started OpenAI, we thought this moment might be possible. It’s happening on the earlier side, and happening in a much more interesting and empowering way for everyone than we’d anticipated (for example, we are seeing an emerging wave of entrepreneurship that we hadn’t previously been anticipating). And at the same time, we are still so early, and there is so much for everyone to define about how these systems get deployed and used in the world.
The next phase will be defined by systems that can do more — reason better, use tools better, plan over longer horizons, and take more useful actions on your behalf. And there are horizons beyond, as AI starts to accelerate science and technology development, which have the potential to truly lift up quality of life for everyone. All of this is starting to happen, in small ways and large, today, and everyone can participate. I feel this shift in my own work every day, and see a roadmap to much more useful and beneficial systems. These systems can truly benefit all of humanity.
Turner Novak 🍌🧢
Holy shit. @joinHandshake’s AI data labeling business went from zero to a $1B run rate in basically a year.
I know everyone loves debating the margins on this revenue, but what an achievement for @GarrettLord and the Handshake team.
Mamoon Hamid: 12 year overnight success. 🚀 @joinHandshake
https://www.theinformation.com/articles/handshake-mercor-revenue-surges-demand-human-contractors-train-ai?rc=daqwrw
SF DISTRICT ATTORNEY
1/ Today, SF DA @BrookeJenkinsSF announced that Daniel Alejandro Moreno-Gama, of Houston, TX, was charged w/ multiple felonies in connection to an attack on a residence in Russian Hill on 3/10/26, and a business on Third Street.
RELEASE: https://sfdistrictattorney.org/texas-man-charged-with-two-counts-of-attempted-murder-and-multiple-other-felonies-in-connection-to-incendiary-destructive-device-thrown-at-russian-hill-residence/
San Francisco Police: View our latest statement regarding an incident that occurred early this morning at a North Beach residence. Officers have made an arrest, and no injuries were reported as a result of this incident.
