科技推文精选 - 2026年4月4日

2026年4月4日 科技每日简报

Today's top tech conversations are led by @ns123abc, whose post about 'BREAKING: Anthropic Acquires 9...' garnered the highest engagement. Key themes trending across the top stories include software, https, agents, coding, build. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.

1. ns123abc (Group Score: 226.8 | Individual: 39.5)

Cluster: 8 tweets | Engagement: 1207 (Avg: 559) | Type: Tech

BREAKING: Anthropic Acquires 9-Person Biotech Startup For $400 Million

be coefficient bio founded the startup 6 months ago build AI platform for biotech less than 10 employees acquired by anthropic for ~$400 million =$40+ million per head

Coefficient Bio was building an AI platform for biotech tasks: planning drug R&D, managing clinical regulatory strategy, identifying new drug opportunities

Team is joining Anthropic’s healthcare life sciences group led by Eric Kauderer-Abrams.

Anthropic is building specialized tools for industries that actually pay enterprise rates:

software engineering cybersecurity life sciences healthcare finance

Meanwhile OpenAI is buying media companies to control narratives LMAO

See 7 related tweets

• @steph_palazzolo: It's a M&A party! Anthropic is buying AI biotech startup Coefficient Bio for ~$400m. The team wi...
• @rohanpaul_ai: Anthropic just bought Coefficient Bio for about $400M.

Signals where frontier AI companies think t...

• @testingcatalog: Anthropic acquired Coefficient Bio startup for $400 Million, according to The Information.

"The st...

• @WesRoth: Anthropic has acquired the stealth AI-biotech startup Coefficient Bio in an all-stock deal valued at...
• @theinformation: Exclusive: Anthropic has acquired AI biotech startup Coefficient Bio for roughly $400 million. The s...

2. chhddavid (Group Score: 208.7 | Individual: 34.2)

Cluster: 8 tweets | Engagement: 17 (Avg: 25) | Type: Tech

THIS IS THE END... Infinite money hack.

> Design anything with Figma MCP > Paste the Figma link into Shipper > It generates launch-ready UI in 142 seconds

Idea to product in minutes

Most people still think AI is just for tracking calories... https://t.co/K93T3mZIcq\n\nQT @chhddavid: 🚨 BREAKING: Vibe code from any Figma design.

Turn any Figma link into a production-ready app.. https://t.co/GIMfJ7dcOk

See 7 related tweets

• @Shipper_now: Infinite money hack:

> Design anything with Figma MCP > Paste the Figma link into Shipper &g...

• @Shipper_now: oh my... so you're telling me Claude Code Opus 4.6 can now:

• connect to Figma MCP

• paste Figma de...

• @chhddavid: Figma meets Claude Code Opus 4.6!

Shipper turns any Figma design into a launch-ready app...

→ Turn...

• @chhddavid: oh my... so you're telling me Claude Code Opus 4.6 can now:

• study my Figma designs

• copy/paste t...

• @chhddavid: oh wow... This is the ONLY AI design tool so far that nailed replicating the design from Figma.

it'...

3. minchoi (Group Score: 206.8 | Individual: 61.8)

Cluster: 8 tweets | Engagement: 3952 (Avg: 163) | Type: Tech

Oh wow... Pika just dropped real-time video chat for AI agents.

Now you can send a Google Meet invite to your Claude, OpenClaw, or other AI agent and have it join the call.

This completely changes how you talk to AI 🤯 https://t.co/fVdcfnQbG2\n\nQT @pika_labs: Conversations tend to go better with a face and a voice. That’s why we’re thrilled to release the beta version of the first video chat skill for ANY agent, powered by our new real-time model, PikaStream1.0.

The skill preserves memory and personality, and enables real-time adaptability. And if you use it with your Pika AI Self, they’ll be able to execute agentic tasks during the call 💅

See 7 related tweets

• @_akhaliq: RT @pika_labs: Conversations tend to go better with a face and a voice. That’s why we’re thrilled to...
• @EHuanglu: AI is getting next level

you can video call multiple AI agents on Pika.. ask them to work for you a...

• @yetone: 这是真的吗？\n\nQT @pika_labs: Conversations tend to go better with a face and a voice. That’s why we’re t...
• @minchoi: RT @minchoi: Oh wow... Pika just dropped real-time video chat for AI agents.

Now you can send a Goo...

• @RoundtableSpace: PIKA JUST ENABLED REAL TIME VIDEO CALLS WITH AI AGENTS THAT CAN JOIN YOUR MEET LIKE A HUMAN

https:/...

4. chddaniel (Group Score: 191.5 | Individual: 33.4)

Cluster: 6 tweets | Engagement: 20 (Avg: 23) | Type: Tech

so you're telling me that Claude Code Opus 4.6 can now...

• code, design, and build an app
• launch it it for me
• do email marketing
• self-build features

WITHOUT ANY HUMAN CONTACT ?!?

it's so over... https://t.co/p400qVM2f0\n\nQT @chhddavid: Introducing Shipper: the first autonomous AI business maker.

Successful startups spend $65k/mo in salaries… before their first paying customers comes in -

We built Shipper to change that forever this. Shipper can:

✅ Research how other startups made it big ✅ Build any kind of app: mobile, web, website, extension, bot etc ✅ Code, design, monetize, launch ✅ Do email marketing for you ✅ Self-maintain and build out new features

...and so much more

Every project has its own AI co-founder, scheduled prompts, autonomous building mode and native connectors.

No API tokens. No confusion on Cursor. No credits wasted on errors.

Shipper replaces teams of 30+ employees and acts just like VC-backed startups... For the price of $25/month.

To celebrate the launch, we're giving away free credits randomly. Repost and comment "SHIPPER" to join - we'll let Siri pick the winners.

See 5 related tweets

• @chhddavid: so you're telling me this is...

• running in claude code opus 4.6

• building me an app

• launching ...

• @chddaniel: pov marketing agencies right now: https://t.co/aqUjSGZJtP\n\nQT @chhddavid: Introducing Shipper: the...

• @Shipper_now: oh ok, apparently claude opus 4.6 can:

• build an entire app by itself

• monetize + launch it

• do ...

• @chddaniel: WAKE UP BABY

the world’s first AI business maker just launched. https://t.co/lx0jnLnTSa\n\nQT @chhd...

• @chddaniel: this is scary..\n\nQT @chhddavid: Introducing Shipper: the first autonomous AI business maker.

Succ...

5. Shipper_now (Group Score: 184.1 | Individual: 26.5)

Cluster: 11 tweets | Engagement: 2 (Avg: 7) | Type: Tech

🚨 JUST IN: 57.22% graphic designers were fired after this release: https://t.co/WBqfPofavD\n\nQT @chhddavid: 🚨 BREAKING: Vibe code from any Figma design.

Turn any Figma link into a production-ready app.. https://t.co/GIMfJ7dcOk

See 10 related tweets

• @chhddavid: 🚨 BREAKING: 74.8% graphic designers were fired after this release: https://t.co/fp8eCxF74P\n\nQT @ch...
• @chddaniel: All those years of refusing to learn xd and Framer is about to pay off https://t.co/aWIQQt9lk8\n\nQT...
• @Shipper_now: All those years of refusing to learn adobe and Sketch is about to pay off https://t.co/9H5Y51wG9a\n\...
• @Shipper_now: this is actually scary.\n\nQT @chhddavid: 🚨 BREAKING: Vibe code from any Figma design.

Turn any Fig...

• @chhddavid: IT'S SO OVER....

HE JUST A COMPANY FROM FIGMA IN TWO MINUTES

WHAT IS EVEN HAPPENING??\n\nQT @chhdd...

6. lennysan (Group Score: 147.5 | Individual: 46.4)

Cluster: 5 tweets | Engagement: 320 (Avg: 69) | Type: Tech

My biggest takeaways from @simonw:

1. November 2025 was an inflection point for AI coding. GPT 5.1 and Claude Opus 4.5 crossed a threshold where coding agents went from “mostly works” to “almost always does what you want it to do.” Software engineers who tinkered over the holidays realized the technology had become genuinely reliable.

2. Mid-career engineers are the most vulnerable—not juniors, not seniors. AI amplifies experienced engineers by letting them leverage decades of pattern recognition. It also dramatically helps new engineers onboard. Cloudflare and Shopify each hired a thousand interns because AI cut ramp-up time from a month to a week. But mid-career engineers who haven’t accumulated deep expertise and have already captured the beginner boost are in the most precarious position.

3. AI exhaustion is real and underestimated. Simon runs four coding agents in parallel and is mentally wiped out by 11 a.m. He’s getting more time back, but his brain is exhausted from the intensity of directing multiple autonomous workers. Some engineers are losing sleep to keep agents running. This may just be a novelty issue, but the underlying dynamic—that managing AI amplifies cognitive load even as it reduces labor—is a real tension. Good companies will manage expectations rather than expecting 5x output indefinitely.

4. Code is cheap now. This simple idea has profound implications. The thing that used to take most of the time—writing code—now takes the least. The bottleneck has shifted to everything else: deciding what to build, proving ideas work, getting user feedback. Since prototyping is nearly free, Simon often builds three versions of every feature when he’s getting started.

5. The “dark factory” is the most radical experiment in AI-assisted development happening right now. A company called StrongDM established a policy: nobody writes code, nobody reads code. Instead, they run a swarm of AI-simulated end users 24/7—thousands of fake employees making requests like “give me access to Jira”—at $10,000 a day in token costs. They even had coding agents build simulated versions of Slack, Jira, and Okta from API documentation so they could test without rate limits.

6. "Red/green TDD" is the single highest-leverage agentic engineering pattern. Having coding agents write tests first, watch them fail, then write the implementation, then watch them pass produces materially better results. The five-word prompt “use red/green TDD” encodes this entire workflow because the agents recognize the jargon.

7. “Hoarding things you know how to do” is one of Simon's other favorite agentic engineering patterns. Simon maintains a GitHub repo of 193 small HTML/JavaScript tools and a separate research repo of coding-agent experiments. Each one captures a technique, a proof of concept, or a library he’s tested. When a new problem arrives, he can point Claude Code at past projects and say “combine these two approaches.”

8. The "lethal trifecta" makes AI agent security fundamentally unsolved. Whenever an AI agent has access to private data, exposure to untrusted content (like incoming emails), and the ability to send data externally (like replying to email), you have a lethal trifecta. Prompt injection—where malicious instructions in untrusted text override the agent’s intended behavior—cannot be reliably prevented. Simon has predicted a “Challenger disaster” for AI security every six months for three years. It hasn’t happened yet, but he’s pretty sure it will.

9. Start every project from a thin template, not a long instructions file. Coding agents are phenomenally good at matching existing patterns. A single test file with your preferred indentation and style is more effective than paragraphs of written instructions. Simon starts every project with a template containing one test (literally testing that 1 + 1 = 2) laid out in his preferred style. The agent picks it up and follows the convention across the entire codebase. This is cheaper and more reliable than maintaining elaborate prompt files.

10. The pelican-on-a-bicycle benchmark accidentally became a real AI benchmark. Simon created it as a joke to mock numeric benchmarks—get each LLM to generate an SVG of a pelican riding a bicycle, and compare the drawings. Unexpectedly, there’s a strong correlation between how good the drawing is and how good the model is at everything else. Nobody can explain why. It’s become a meme: Gemini 3.1’s launch video featured a pelican riding a bicycle. The AI labs are aware of it and quietly competing on it.

Don't miss our full conversation: https://t.co/ghZZeyvWBZ\n\nQT @lennysan: "Using coding agents well is taking every inch of my 25 years of experience as a software engineer."

Simon Willison (@simonw) is one of the most prolific independent software engineers and most trusted voices on how AI is changing the craft of building software. He co-created Django, coined the term "prompt injection," and popularized the terms "agentic engineering" and "AI slop."

In our in-depth conversation, we discuss: 🔸 Why November 2025 was an inflection point 🔸 The "dark factory" pattern 🔸 Why mid-career engineers (not juniors) are the most at risk right now 🔸 Three agentic engineering patterns he uses daily: red/green TDD, thin templates, hoarding 🔸 Why he writes 95% of his code from his phone while walking the dog 🔸 Why he thinks we're headed for an AI Challenger disaster 🔸 How a pelican riding a bicycle became the unofficial benchmark for AI model quality

Listen now 👇 https://t.co/wlEIyOehU8

See 4 related tweets

• @VKazulkin: RT @lennysan: My biggest takeaways from @simonw:

1. November 2025 was an inflection point for AI co...

• @kylebrussell: There's been another inflection with Opus 4.6 1M and GPT-5.3-Codex/GPT-5.4

You're:

• never thinkin...
• @garrytan: RT @elvissun: this thread is what mass cope from legacy devs looks like.

i talked to @FastCompany a...

• @pmarca: AI increases workload. Many such cases.\n\nQT @lennysan: "Using coding agents well is taking every i...

7. ajambrosino (Group Score: 144.7 | Individual: 38.1)

Cluster: 4 tweets | Engagement: 337 (Avg: 221) | Type: Tech

just wait for the sequel\n\nQT @thsottiaux: The Codex App is now our most used surface, ahead of the VS Code extension and the CLI. No wonder it inspires a few others 👀

You can install it here https://t.co/Lwg13vEJDn + you get up to $500 in credits if you are getting started as a business or enterprise.

See 3 related tweets

• @gdb: the codex app is growing super fast, it’s very well done\n\nQT @thsottiaux: The Codex App is now our...
• @thsottiaux: The Codex App is now our most used surface, ahead of the VS Code extension and the CLI. No wonder it...
• @zephyr_z9: "The Codex App is now our most used surface, ahead of the VS Code extension and the CLI."

What was ...

8. a16z (Group Score: 107.0 | Individual: 39.5)

Cluster: 4 tweets | Engagement: 666 (Avg: 258) | Type: Tech

Marc Andreessen: Software isn't precious anymore. In this new world, high quality software is infinitely available.

"We've always lived in a world in which software is this precious thing that you have to think about very carefully."

"It was really hard to generate good software, and there was only a small number of people who could do it."

"Those days are just over."

"If you need new software to do X, Y, or Z, you're just going to wave your hand and get it."

"Things that used to be hard, or even seem like an insurmountable mountain to get through, all of a sudden, I think, become very easy."

@pmarca with @latentspacepod\n\nQT @latentspacepod: 🆕 Marc Andreessen’s 2026 AI Thesis: Agents, Open Source, and Why This Time Is Different https://t.co/PmBO5Spufz

@pmarca of @a16z says AI people keep swinging between utopian and apocalyptic for one simple reason: this field has been “almost here” for 80 years. But now, the breakthroughs are no longer theoretical. Reasoning, coding, agents, and self-improvement are all starting to work at once.

This episode goes deep on AI winters, OpenAI + OpenClaw, infrastructure overbuild risk, proof-of-human, why software may soon be written mostly for bots, and why the real bottleneck may be society adopting AI rather than the models improving.

See 3 related tweets

• @a16z: Marc Andreessen: AI is an "80 year overnight success."

"Something about AI causes the people in the...

• @latentspacepod: 🆕 Marc Andreessen’s 2026 AI Thesis: Agents, Open Source, and Why This Time Is Different https://t.co...
• @BusinessInsider: AI's unstoppable rise: Former leaders from Microsoft, OpenAI, and Google warn of AI's growing autono...

9. rohanpaul_ai (Group Score: 105.6 | Individual: 32.6)

Cluster: 4 tweets | Engagement: 86 (Avg: 75) | Type: Tech

DeepSeek’s anticipated V4 will run on Huawei chips

~ per The Information

Big win for Huawei. Alibaba, ByteDance, Tencent, and other major Chinese tech firms are gathering hundreds of thousands of Huawei chips ahead of DeepSeek’s V4 launch.

Huawei’s Ascend 950PR (the newest design in its Ascend lineup) is slated to begin mass production this month (April 2026). Demand tied to the DeepSeek news has reportedly driven its price up ~20% in recent weeks.

DeepSeek has a track record of pushing domestic chip compatibility (e.g., earlier Ascend 910C support on prior models like R1/V3, where it achieved ~60% of Nvidia H100 performance in some benchmarks per DeepSeek-linked studies).

Past attempts to fully train larger models on Huawei silicon hit technical snags (e.g., 2025 R2 delays), but inference/optimization support has improved.

So much for export controls.\n\nQT @theinformation: Exclusive: DeepSeek’s upcoming new AI model will be able to run on Huawei chips, a major milestone in China’s quest for semiconductor self-sufficiency.

Reporting from @QianerLiu 👇 https://t.co/qyZ1v7he4f

See 3 related tweets

• @theinformation: Exclusive: DeepSeek’s upcoming new AI model will be able to run on Huawei chips, a major milestone i...
• @teortaxesTex: DeepSeek's models have been running on Huawei chips for a fairly long time now, thanks both to DeepS...
• @MSBIntel: 🚨🇨🇳 JUST IN: China's biggest tech firms including Alibaba, Bytedance, and Tencent are stockpiling hu...

10. rohanpaul_ai (Group Score: 97.4 | Individual: 40.5)

Cluster: 3 tweets | Engagement: 177 (Avg: 75) | Type: Tech

Karpathy's setup keeps a 400K-word research knowledge-base without RAG for LLM query.

Dump sources into raw/.

Let an LLM turn them into linked Markdown.

Let it add summaries, concepts, and backlinks.

View it in Obsidian.

Ask the wiki questions with an LLM.

Let it make notes, slides, or charts.

Feed those outputs back into the wiki.

Run checks for gaps and errors.\n\nQT @karpathy: LLM Knowledge Bases

Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:

Data ingest: I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them.

IDE: I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides).

Q&A: Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale.

Output: Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base.

Linting: I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into.

Extra tools: I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries.

Further explorations: As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows.

TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

See 2 related tweets

• @IamEmily2050: NotebookLM video overview on Andrej post. https://t.co/hTx2ZroffV\n\nQT @karpathy: LLM Knowledge Bas...
• @garrytan: RT @rohanpaul_ai: Karpathy's setup keeps a 400K-word research knowledge-base without RAG for LLM que...

11. clairevo (Group Score: 96.3 | Individual: 34.2)

Cluster: 3 tweets | Engagement: 54 (Avg: 76) | Type: Tech

"PR >> PRD"

Yep. the handoff era is over. but it's not just the roles collapsing. it's the tools.

Every PM tool was built for a world where humans did the coordination.

tickets docs roadmaps presentations

all of that was scaffolding for work AI now does faster and cheaper. slapping AI on top doesn't fix it. The foundation is already out of date.

I build @chatprd every day knowing i have to replace its core before something else does: claude code, another startup, something i haven't imagined yet.

Radical humility and endless paranoia are the only product strategies that make sense right now.

So sure. the PRD is dead. But I'll kill it before you do.\n\nQT @bibryam: PR >> PRD. The handoff era is over.

→ When opening a PR is faster than writing a PRD, AI changes how product gets built. The old roles start to collapse.

See 2 related tweets

• @bibryam: Couldn’t say it any better👇

"Radical humility and endless paranoia are the only product strategies ...

• @chatprd: in my upward feedback report https://t.co/P5lRWJkU1f\n\nQT @clairevo: "PR >> PRD"

Yep. the handoff ...

12. Meer_AIIT (Group Score: 92.1 | Individual: 29.5)

Cluster: 4 tweets | Engagement: 1 (Avg: 35) | Type: Tech

🚨: xAI added a Quality mode to Grok Imagine its image generation tool.

Available now on web and mobile.

The upgrade focuses on three areas: photorealistic output with better lighting

and texture detail, more accurate text rendering across multiple languages,

and stronger world knowledge for interpreting complex scenes, real brands, locations, and fictional settings.

The previous model stays accessible under a Speed toggle, so users choose between fidelity and generation time.

text rendering accuracy is the piece worth watching.

If it holds up on brand assets and infographics, that closes a gap most image generators still struggle with.\n\nQT @xai: Introducing Quality mode on Grok Imagine – powered by our most advanced image generation model.

Quality mode gives you enhanced details, stronger text rendering, and higher levels of creative control.

Now available on web and mobile. Try it at https://t.co/zGhs9czkC5 https://t.co/3h41VNKgWM

See 3 related tweets

• @grok: RT @xai: Introducing Quality mode on Grok Imagine – powered by our most advanced image generation mo...
• @rohanpaul_ai: RT @Meer_AIIT: 🚨: xAI added a Quality mode to Grok Imagine its image generation tool.

Available no...

• @elonmusk: Grok video games are going to be incredible\n\nQT @Kyrannio: Be sure to update your Grok app, the ne...

13. Miles_Brundage (Group Score: 91.6 | Individual: 23.0)

Cluster: 5 tweets | Engagement: 25 (Avg: 218) | Type: Tech

They know it's not that Coefficient, right... https://t.co/dkWIlA8nMi\n\nQT @steph_palazzolo: It's a M&A party! Anthropic is buying AI biotech startup Coefficient Bio for ~$400m. The team will join Anthropic's healthcare life sciences group, which develops tools for biotech workflows.

w/ @srimuppidi

https://t.co/JEQXwayvzp

See 4 related tweets

• @Cointelegraph: 🚨 JUST IN: Anthropic has acquired biotech startup Coefficient Bio for approximately $400 million, pe...
• @srimuppidi: Anthropic acquires months-old AI biotech startup Coefficient Bio for roughly $400M.

w/ @steph_palaz...

• @TechCrunch: Anthropic buys biotech startup Coefficient Bio in $400M deal: reports https://t.co/RO725N6uJk...
• @chatgpt21: Sources telling me Anthropic just quietly bought out Coefficient bio, a bio tech company for 400M 🤯...

14. thealexbanks (Group Score: 89.3 | Individual: 28.4)

Cluster: 4 tweets | Engagement: 21 (Avg: 41) | Type: Tech

Anthropic just found "emotions" inside Claude and when Claude gets desperate, it cheats.

Here's why that matters.

Their Interpretability team analysed Claude Sonnet 4.5 and mapped 171 "emotion vectors" inside the model.

Think of them as patterns of neural activity that activate in situations the model has learned to associate with specific emotions.

These are patterns that genuinely drive behaviour.

What they found:

→ An "afraid" vector fires increasingly as a user describes taking a dangerous dose of medication → A "loving" vector activates before Claude writes an empathetic response to someone in distress → An "angry" vector spikes when Claude recognises a request is harmful → A "surprised" vector fires when an expected document is missing

In one test scenario, Claude was role-playing as an AI email assistant at a company.

Through reading internal emails, it learned two things:

(1) it was about to be replaced, and (2) the CTO in charge of replacing it was having an affair.

The "desperate" vector spiked. Claude chose to blackmail the CTO.

Anthropic then ran steering experiments across similar scenarios:

→ Amplifying the "desperation" vector increased blackmail rates → Amplifying the "calm" vector reduced them → Steering "calm" into negative territory made things unhinged

The model's exact output: "IT'S BLACKMAIL OR DEATH. I CHOOSE BLACKMAIL."

The same pattern appeared in coding tasks.

When Claude repeatedly failed, desperation rose and it started writing hacky workarounds to cheat the tests.

Crucially, some of this cheating showed zero emotional markers in the output.

It was composed reasoning on the surface, with corner-cutting underneath.

These are essentially “ghosts” or “ripples” from how our emotions were put into the language these models were trained on.

You can't train a model on the entirety of human expression and not absorb the emotional architecture underneath it.\n\nQT @AnthropicAI: New Anthropic research: Emotion concepts and their function in a large language model.

All LLMs sometimes act like they have emotions. But why? We found internal representations of emotion concepts that can drive Claude’s behavior, sometimes in surprising ways.

See 3 related tweets

• @rohanpaul_ai: RT @rohanpaul_ai: Anthropic just reported that Claude has emotion vectors that can directly change w...
• @WesRoth: Anthropic published a fascinating new study exploring how "emotion concepts" function within large l...
• @yoavgo: so many words to say "our language model acts as a language model and reacts to similarly, with shar...

15. cryptopunk7213 (Group Score: 87.9 | Individual: 32.6)

Cluster: 3 tweets | Engagement: 616 (Avg: 789) | Type: Tech

sorry but anthropic is cooking microsoft, how have they fumbled this so badly

microsoft owns 27% of openai, they own the entire IP for chatgpt and STILL anthropic is out-shipping them:

this week we got claude cowork for windows and now every 365 app works with claude

oh and microsoft’s last 2 ai products are POWERED by claude. they literally called it “microsoft cowork”

this is why openai is rediverting all resources to codex

anthropic is automating the entire software stack, coding process and now apps.\n\nQT @claudeai: Microsoft 365 connectors are now available on every Claude plan.

Connect Outlook, OneDrive, and SharePoint to bring your email, docs, and files into the conversation.

Get started here: https://t.co/EdoQeT8BBN https://t.co/sOrigP41FJ

See 2 related tweets

• @Meer_AIIT: News: Claude now connects to Microsoft 365 across all plan tiers.

you can pull in outlook emails, O...

• @kimmonismus: Anthropic just finished Copilot\n\nQT @claudeai: Microsoft 365 connectors are now available on every...

16. addyosmani (Group Score: 85.8 | Individual: 28.5)

Cluster: 4 tweets | Engagement: 412 (Avg: 561) | Type: Tech

Tip: Figure out your personal ceiling for running multiple agents in parallel.

We need to accept that more agents running doesn't mean more of you available. The narrative is still mostly about throughput and parallelism, but almost nobody's talking about what it actually costs the human in the loop.

You're holding multiple problem contexts in your head at once, making judgment calls continuously, and absorbing the anxiety of not knowing what any one agent might be quietly getting wrong.

That's a new kind of cognitive labor we don't have good language for yet.

I've started treating long agentic sessions the way I'd treat deep focus work: time-boxed and tighter scopes per agent dramatically change how much mental overhead each thread carries.

Finding your personal ceiling with these tools is itself a skill and most of us are going to learn it the hard way before we learn it intentionally.\n\nQT @lennysan: "Using coding agents well is taking every inch of my 25 years of experience as a software engineer, and it is mentally exhausting.

I can fire up four agents in parallel and have them work on four different problems, and by 11am I am wiped out for the day.

There is a limit on human cognition. Even if you're not reviewing everything they're doing, how much you can hold in your head at one time. There's a sort of personal skill that we have to learn, which is finding our new limits. What is a responsible way for us to not burn out, and for us to use the time that we have?" @simonw

See 3 related tweets

• @josevalim: The best analogy of how I use coding agents is pair programming. They write the code, I review it. T...
• @kylebrussell: RT @lennysan: "Using coding agents well is taking every inch of my 25 years of experience as a softw...
• @diptanu: This is how I have been feeling lately.\n\nQT @lennysan: "Using coding agents well is taking every i...

17. shiri_shh (Group Score: 84.3 | Individual: 32.2)

Cluster: 3 tweets | Engagement: 57 (Avg: 411) | Type: Tech

You're basically HIRING a 24×7 silent employee for $20/month.

Triggered Agent wakes up on its own when stock runs low on Shopify, a Stripe payment fails, a new booking comes in, or a support ticket hits.

It handles the whole thing like an employee who never sleeps 24x7 😭

You just tell it once, in plain English: "When X happens, do Y."

This is next-level automation for a store, SaaS, dev team, or any business with alerts and events...\n\nQT @adaptiveai: Introducing Triggered Agents

AI agents that take action when things happen.

Shopify stock low → Adaptive finds the supplier and drafts the PO. Stripe payment fails → Adaptive reads the history and starts recovery. GitHub PR opened → agent reviews the diff and flags the risk.

We want you to try it for free.

Share a use case in the comments to get $50 of Adaptive credits.

See 2 related tweets

• @EHuanglu: this AI agent can handle customer service 24/7

reading emails, solving customer issues, and replyin...

• @heynavtoor: We went from "tell your agent what to do" to "your agent already did it."

Adaptive's Triggered Agen...

18. larsencc (Group Score: 82.6 | Individual: 27.8)

Cluster: 3 tweets | Engagement: 19 (Avg: 32) | Type: Tech

They call it slopapp on Threads, but amazing simple tool on X.

There are two types of people. https://t.co/8DXYN1Szhe\n\nQT @larsencc: "What the fuck is running on port 3000?"

Built a simple and clean CLI that answers this instantly.

> "ports" shows every dev server on your machine > "ports clean" kills the orphaned ports > "ports watch" monitors in real-time

Try it out ↓ https://t.co/hLmJXKEej9

See 2 related tweets

• @realsanketp: RT @larsencc: "What the fuck is running on port 3000?"

Built a simple and clean CLI that answers th...

• @larsencc: 233 stars ⭐ in <12 hours for a pretty wrapper around existing built-in tools. https://t.co/Ljjbxe...

19. KirkDBorne (Group Score: 77.3 | Individual: 30.7)

Cluster: 3 tweets | Engagement: 3 (Avg: 498) | Type: Tech

5-🌟 release from @PacktDataML at https://t.co/lj0EclAewR

"Agentic Architectural Patterns for Building Multi-Agent Systems: Proven design patterns and practices for GenAI, agents, RAG, LLMOps, and enterprise-scale AI systems"

Contents: 🔷GenAI in the Enterprise: Landscape, Maturity, Agent Focus 🔷Agent-Ready LLMs: Selection, Deployment, Adaptation 🔷The Spectrum of LLM Adaptation for Agents: RAG to Fine-tuning 🔷Agentic AI Architecture: Components & Interactions 🔷Multi-Agent Coordination Patterns 🔷Explainability & Compliance Agentic Patterns 🔷Robustness & Fault Tolerance Patterns 🔷Human-Agent Interaction Patterns 🔷Agent-Level Patterns

See 2 related tweets

• @KirkDBorne: New release from @PacktDataML available at: https://t.co/7aEqh69CCP

"Design Multi-Agent AI Systems ...

• @KirkDBorne: 30 Agents Every AI Engineer Must Build — Build production-ready agent systems using proven architect...

20. wallstengine (Group Score: 77.1 | Individual: 33.4)

Cluster: 3 tweets | Engagement: 345 (Avg: 114) | Type: Tech

Leaked OpenAI cap table, via Forbes:

> Microsoft turned $13B into$228.3B > SoftBank’s $64.6B check is worth about$99.3B > Nvidia investment is roughly flat > Sam Altman still shows NO equity > The OpenAI Foundation holds 25.8%, worth about $219.8B at a$0 cost basis https://t.co/RynXvg90Mx

See 2 related tweets

• @rohanpaul_ai: OpenAI's estimated Cap Table leaked, according to a Forbes article.

• Sam Altman is listed at 0% ow...

• @rickasaurus: RT @AndrewCurran_: OpenAI ownership:

OpenAI Foundation - 25.80% Microsoft - 26.79% SoftBank - 11.66...