- Published on
每日科技推特精选 - 2026-03-03
- Authors
- Name
- geeknotes
今日科技动态：科技行业正转向代理式 AI (Agentic AI) 与专有数据护城河，标志性事件包括 MyFitnessPal 收购 Cal AI，以及 OpenClaw 在 GitHub 上的受关注度超越 React。在 SpaceX 持续扩展星链 (Starlink) 移动连接业务之际，阿联酋一处 AWS 数据中心遭袭，引发了业界对物理基础设施风险的关注。工程重心正从“氛围编程” (vibe coding) 转向推理缩放与可维护性；与此同时，Anthropic 正努力应对服务中断，AI 亦开始重塑教育及软件开发领域的传统设计工作流。
1. SawyerMerritt (Group Score: 142.9 | Individual: 33.8)
Cluster: 5 tweets | Engagement: 2982 (Avg: 1701) | Type: Tech
SpaceX has released an updated @Starlink Mobile website, along with a new video and coverage map that shows where the service is currently available and where it's coming soon.
Starlink Mobile: • Available on 40+ apps and on 100+ devices. Future service will offer video calls, streaming, emails and more. • 32+ countries (covering 1.7+ billion people) • The next generation of Starlink Mobile satellites - V2 - will deliver full cellular coverage to "places never thought possible via the highest performing satellite-to-mobile network ever built." • V2 Starlink Mobile satellites will use custom SpaceX-designed silicon and phased array antennas. The satellites will support thousands of spatial beams and higher bandwidth capability, enabling around 20x the throughput capability as compared to a first-generation satellite. • V2 will enable full 5G cellular connectivity with a comparable experience to current terrestrial service.
Site: https://t.co/PYC2IEJuYG
See 4 related tweets
- @Starlink: Starlink Mobile is partnering with @deutschetelekom to support over 140M subscribers across 10 Europ...
- @SawyerMerritt: Below is SpaceX's full keynote today at the World Mobile Congress in Barcelona. They discuss @Starli...
- @niccruzpatane: Elon Musk on Starlink Mobile:
"It will allow SpaceX delivery high bandwidth connectivity directly f...
- @SawyerMerritt: SpaceX's next-generation version 2 @Starlink Mobile satellites will be a gamechanger:
• Custom Spac...
2. rohanpaul_ai (Group Score: 109.1 | Individual: 49.7)
Cluster: 3 tweets | Engagement: 1271 (Avg: 95) | Type: Tech
Larry Ellison on the AI moat:
AI is commoditizing because models use the same public internet data.
The true competitive edge isn't the model itself anymore, but access to exclusive, proprietary datasets. That is the only moat left.
See 2 related tweets
- @rohanpaul_ai: RT @rohanpaul_ai: Larry Ellison on the AI moat:
AI is commoditizing because models use the same pub...
- @kimmonismus: Larry Ellison argues that as AI models trained on the same public data converge and become commoditi...
3. KirkDBorne (Group Score: 103.8 | Individual: 27.1)
Cluster: 5 tweets | Engagement: 4 (Avg: 68) | Type: Tech
Check this out at https://t.co/yb8splH5S3
"The Agentic AI Bible: The Complete and Up-to-Date Guide to Design, Build, and Scale Goal-Driven, LLM-Powered Agents that Think, Execute and Evolve" https://t.co/qdLBgeg7jE
See 4 related tweets
- @KirkDBorne: The Most Complete #AI Agentic Engineering System — Step-by-step guide to build, optimize, and scale ...
- @KirkDBorne: "Building AI Agents with LLMs, RAG, and Knowledge Graphs — A practical guide to autonomous and moder...
- @alex_prompter: RT @free_ai_guides: The hardest part of building an AI agent isn't the idea.
It's writing the syste...
- @KirkDBorne: The Agentic AI Playbook 2026 Edition Turns LLMs into Reliable AI Agents: https://t.co/k2HSYXeSjJ htt...
4. shanaka86 (Group Score: 90.6 | Individual: 40.3)
Cluster: 3 tweets | Engagement: 8091 (Avg: 6434) | Type: Tech
THE FIRST CLOUD INFRASTRUCTURE CASUALTY OF WAR
An Amazon Web Services data center in the UAE just got hit.
AWS confirmed that at approximately 4:30 AM PST on March 1, “objects struck” the facility in availability zone mec1-az2, creating sparks and igniting a fire. The UAE fire department cut power to the building. The zone went dark. AWS says other zones remain operational and restoration will take several hours.
Read that sentence again. “Objects struck.”
The most valuable corporate infrastructure on earth is now absorbing kinetic damage from a state-level military conflict, and the world’s largest cloud provider is describing missile or drone debris as “objects” because no corporate communications playbook exists for this scenario.
This is the first time in history that a major hyperscaler data center has been physically struck during a war.
Every cloud architecture slide deck in every boardroom on earth assumes physical security means perimeter fences and biometric locks. Not ballistic missile defense. Not drone intercept capability. Not wartime fire suppression while the building next door absorbs ordnance.
The Jerusalem Post reported the facility was used by Israel’s military. If confirmed, Iranian targeting of dual-use cloud infrastructure transforms every data center in a conflict-adjacent geography from civilian asset to military target. The distinction between cloud infrastructure and defense infrastructure just collapsed.
And the geography matters enormously. AWS chose the UAE for its Middle East region precisely because Dubai and Abu Dhabi offered stability, connectivity, and proximity to enterprise clients across the Gulf. That thesis died on a Saturday morning when Iranian drones struck the Burj Al Arab, hit Jebel Ali port, and set fire to a data center running workloads for governments, banks, and military operations simultaneously.
The concentration risk is staggering. AWS, Microsoft Azure, and Google Cloud all operate Middle East regions clustered in the same geographic corridor that just became an active theater of war. Oracle has infrastructure in Dubai. Every enterprise running production workloads in these regions is now calculating disaster recovery scenarios that were categorized as “theoretical” 72 hours ago.
The insurance implications alone will restructure cloud pricing for a decade. Lloyd’s of London was already reassessing war-risk exclusions after Ukraine. Now a drone has physically damaged a data center belonging to a $2 trillion company in a country that markets itself as the safest business hub in the region.
AWS built multi-availability-zone redundancy for earthquakes, power failures, and network partitions. Not for Iranian retaliation against a joint US-Israeli military campaign. The architecture held because one zone went down while others stayed up. But the premise broke: that geography selection for cloud regions is a business decision, not a wartime calculation.
Cybersecurity expert Lukasz Olejnik flagged the euphemistic language immediately. AWS did not say “bombed.” AWS said “objects struck.” That linguistic gap is the entire story. The world’s cloud infrastructure just entered the theater of war and the industry has no vocabulary for it yet.
The vocabulary will be priced in by Monday.
See 2 related tweets
- @Reuters: Amazon's cloud-computing facilities in the Middle East faced power and connectivity issues after uni...
- @ReutersBiz: WATCH: Amazon's cloud-computing facilities in the Middle East faced power and connectivity issues af...
5. HarryStebbings (Group Score: 90.3 | Individual: 32.2)
Cluster: 3 tweets | Engagement: 185 (Avg: 247) | Type: Tech
It is BS that vibe coding will replace software companies...
"People underestimate how hard it is to maintain software over time. It's very easy to create the first increment but to change it, adapt it over time, it takes a lot of effort to do that.
Businesses have a core business operation they need to focus on. Having a dedicated person or a team vibe coding some apps is a huge cost.
So vibe coding is amazing but I don't think it's going to disrupt software companies." @zzeran
Love to hear your thoughts on this @antonosika @amasad @jasonlk @JaredSleeper @dharmesh @tobi
See 2 related tweets
- @GenAI_is_real: hot take: the real moat in 2026 isnt knowing how to prompt an LLM to write code for you, its knowing...
- @Franc0Fernand0: I think vibe coding is great for experienced developers. It’s a way to turn out a lot of good code v...
6. KirkDBorne (Group Score: 82.6 | Individual: 32.5)
Cluster: 3 tweets | Engagement: 12 (Avg: 68) | Type: Tech
See the latest book from @AltoValentina, released by @PacktDataML at https://t.co/yPRfXTozGF
"AI Agents in Practice — Design, Implement, and Scale Autonomous #AI Systems for Production"
Book Description (from Amazon):
As AI agents evolve to take on complex tasks and operate autonomously, you need to learn how to build these next-generation systems. Author Valentina Alto brings practical, industry-grounded expertise in AI Agents in Practice to help you go beyond simple chatbots and create AI agents that plan, reason, collaborate, and solve real-world problems using large language models (LLMs) and the latest open-source frameworks.
In this book, you'll get a comparative tour of leading AI agent frameworks such as LangChain and LangGraph, covering each tool's strengths, ideal use cases, and how to apply them in real-world projects. Through step-by-step examples, you’ll learn how to construct single-agent and multi-agent architectures using proven design patterns to orchestrate AI agents working together. Case studies across industries will show you how AI agents drive value in real-world scenarios, while guidance on responsible AI will help you implement ethical guardrails from day one. The chapters also set the stage with a brief history of AI agents, from early rule-based systems to today's LLM-driven autonomous agents, so you understand how we got here and where the field is headed.
By the end of this book, you'll have the practical skills, design insights, and ethical foresight to build and deploy AI agents that truly make an impact.
See 2 related tweets
- @KirkDBorne: AI Agents in Action: https://t.co/JguQjWzeHH v/ @ManningBooks
Gain hands-on experience in these are...
- @DataScienceDojo: 🚨 Most AI coding agents fail before they reach production. Not because the model is weak. Not becaus...
7. openclaw (Group Score: 74.1 | Individual: 48.0)
Cluster: 2 tweets | Engagement: 12847 (Avg: 1862) | Type: Tech
We just passed React on GitHub stars. 🦞
Let that sink in. A personal AI assistant built by a lobster-obsessed Austrian and an army of crustacean enthusiasts just outstarred the library that powers half the internet.
We shipped 90+ changes today. They shipped a conference. https://t.co/dC8EWo8yN7
See 1 related tweets
- @steipete: RT @openclaw: We just passed React on GitHub stars. 🦞
Let that sink in. A personal AI assistant bui...
8. business (Group Score: 61.5 | Individual: 24.9)
Cluster: 3 tweets | Engagement: 396 (Avg: 112) | Type: Tech
Anthropic’s popular artificial intelligence chatbot Claude experienced service disruptions on Monday, with thousands of users reporting issues, according to monitoring website Downdetector https://t.co/vvBelYhoOv
See 2 related tweets
- @Forbes: Anthropic said Monday morning it identified a cause for the outage as thousands of users report diff...
- @techdaily24: Anthropic's Claude AI services faced widespread disruptions this morning, March 2, 2026. Users repor...
9. Zai_org (Group Score: 59.0 | Individual: 41.2)
Cluster: 2 tweets | Engagement: 2159 (Avg: 637) | Type: Tech
https://t.co/DRCwFHQQRn Startup Program is NOW OPEN.
What you can get: ·Free API credits ·Priority rate limits ·Exclusive Community ·Early API Access
Who we're looking for: ·AI-native startups ·Agent builders ·SaaS founders integrating LLM infra ·Global teams building for real-world scale
If you're building something that matters, don't wait!!
Apply now: https://t.co/jyS5zVZKQx Questions? Details? Follow & DM @ZaiforStartups
See 1 related tweets
- @KarthiDreamr: RT @Zai_org: https://t.co/IQMnfBczAT Startup Program is NOW OPEN.
What you can get: ·Free API credi...
10. seraleev (Group Score: 54.1 | Individual: 35.3)
Cluster: 2 tweets | Engagement: 253 (Avg: 57) | Type: Tech
MyFitnessPal acquired Cal AI – the viral app that counts calories from food photos.
In ~2 years, the app grew to 15M+ downloads and around $30M in annual revenue. Team size: just 7 people.
After the deal, Cal AI will remain a standalone product but gain access to MyFitnessPal’s database of 20M foods.
Congrats @zach_yadegari 🚀
See 1 related tweets
- @techdaily24: MyFitnessPal has acquired Cal AI, a highly successful AI-powered calorie counting app developed by t...
11. GenAI_is_real (Group Score: 53.1 | Individual: 53.1)
Cluster: 1 tweets | Engagement: 498 (Avg: 41) | Type: Tech
1M websocket connections is a solved problem at this point. you know whats NOT solved? 1M concurrent inference requests with KV cache management, dynamic batching, and variable sequence lengths. real-time AI serving at scale makes traditional connection management look easy in comparison
12. lennysan (Group Score: 51.9 | Individual: 51.9)
Cluster: 1 tweets | Engagement: 897 (Avg: 150) | Type: Tech
My biggest takeaways from @jenny_wen (design lead at @AnthropicAI):
The traditional design process is breaking down. The classic discover-diverge-converge loop that designers have relied on for years doesn’t work when engineers can spin up seven coding agents and ship a working version before a designer finishes exploring options.
Design work is splitting into two distinct modes. The first is supporting execution: consulting with engineers as they build, giving feedback, polishing in code. The second is setting short-range vision, now scoped to three to six months instead of multi-year roadmaps. The vision work is still critical because when everyone can build anything fast, someone needs to point the team in a coherent direction.
Build trust through speed, not perfection. Anthropic ships products early, labels them research previews, and then iterates publicly based on real feedback. Jenny argues that what actually degrades a brand isn’t launching something rough; it’s launching something rough and then going silent. If you ship fast, respond to feedback visibly, and keep improving, users will trust you more, not less.
The most overlooked hire in design right now is the cracked new grad. Most companies are hiring senior designers with deep experience. Jenny argues that early-career people with blank slates, fast learning curves, and no attachment to legacy processes may be uniquely suited to this moment. They don’t carry baked-in rituals that are now obsolete, and their lack of expectations can actually be an advantage.
Chat as an interface isn’t going away. Despite expectations that chatbots were a temporary stop on the way to richer UIs, Jenny sees chat as a permanently valuable interface because it offers infinite flexibility. But she expects a hybrid future where models increasingly generate UI elements on the fly for specific tasks (like the interactive widgets Claude recently shipped) while chat remains the connective tissue between them.
Jenny went from design director (12 to 15 reports) back to IC. She questioned whether middle management had a safe future and wanted hands-on time during a period of rapid change. The IC time is giving her hard skills she wouldn’t have gained while managing.
AI will likely get better at taste and judgment. Jenny says designers may be holding onto “taste” as a moat too tightly. But someone still has to be accountable for what ships, the same way an engineer is accountable for AI-generated code.
Hire three archetypes: strong generalists, deep specialists, and “cracked new grads.” Strong generalists are “block-shaped” (80th percentile across multiple skills). Deep specialists are top 10% in one area. Cracked new grads—the most overlooked—have no baked-in processes and learn new tools fastest.
Figma is still essential, but for different reasons than before. Jenny says Figma remains the best tool for rapidly exploring 8 to 10 different design directions on a canvas, something that coding tools handle poorly because they’re too linear and create investment bias toward one direction. For micro-level visual and interaction decisions, spatial exploration still beats sequential iteration.
Low-leverage work is often the highest-leverage thing a manager can do. Jenny pushes back on the conventional management advice to ruthlessly prioritize only high-leverage tasks. She points to leaders who obsessively dogfood the product, repro bugs, and personally fix small issues—activities that seem “below” a senior leader but create deep product familiarity, set a cultural tone of care, and earn trust from the team in ways that strategic planning never can.
Watch our full conversation: https://t.co/UF9AzyCFac
13. minchoi (Group Score: 51.3 | Individual: 26.1)
Cluster: 3 tweets | Engagement: 577 (Avg: 504) | Type: Tech
AI just made history lessons actually interesting.
Walking through historic scenes with a guide.
This format is going to take over education. https://t.co/fT5BISyU7F
See 2 related tweets
- @minchoi: RT @minchoi: AI just made history lessons actually interesting.
Walking through historic scenes wit...
- @EHuanglu: RT @EHuanglu: AI now teaches history on the spot
this makes school feel like from stone age https:/...
14. every (Group Score: 45.2 | Individual: 29.6)
Cluster: 2 tweets | Engagement: 14 (Avg: 14) | Type: Tech
At our first OpenClaw Camp, over 500 subscribers watched four people demo AI agents they've been running daily for weeks.
- @NatEliason's @FelixCraftAI has its own X account, wrote a book, and launched an @openclaw marketplace.
- @bran_don_gell's Zosia orders groceries and tracks nanny hours.
- @tedescau's Judd flags growth metrics before meetings.
- @clairevo's Polly schedules meetings mid-Target run.
Full writeup: https://t.co/5e7oqL9Fbl
See 1 related tweets
- @every: RT @every: At our first OpenClaw Camp, over 500 subscribers watched four people demo AI agents they'...
15. minchoi (Group Score: 45.1 | Individual: 31.6)
Cluster: 2 tweets | Engagement: 3368 (Avg: 504) | Type: Tech
RT @minchoi: Holy smokes... this guy recreated a God's eye view 4D replay of Operation Epic Fury.
Using only public data and an AI agent s…
See 1 related tweets
- @RoundtableSpace: THIS GUY RECREATED A 4D GOD’S-EYE VIEW OF A WAR USING ONLY PUBLIC DATA + AI AGENTS.
16. cryptopunk7213 (Group Score: 44.1 | Individual: 44.1)
Cluster: 1 tweets | Engagement: 4194 (Avg: 1145) | Type: Tech
yo so just to recap the most INSANE week of 2026 (and how 99% of it was caused by 1 fucking company 😂):
U.S. starts war with Iran, kills supreme leader khamenei USING CLAUDE (anthropic) to facilitate it.
anthropic tells pentagon to “fuck off” for trying to use claude for mass surveillance - public loves it - claude hits #1 in app store BUT THEN
Trump blacklists their ass RIGHT BEFORE openAI swoops in to steal the ENTIRE DEAL becoming the flagship ai model of the U.S. military in <24hrs
Anthropic then accused china of hacking claude but it backfires - public calls them hypocrites (dario’s losing hair at this point)
boris cherny (creator of claude code) sees all this drama, yawns and ships 4 BANGER products that threaten OpenClaw’s market share on personal agents
then anthropic’s Head of ‘Special Projects’ saw this and shipped a feature that lets you steal memory from chatgpt and upload it to claude in <60 seconds - switching costs to claude went to ZERO.
oh yea and openAI raised $110B, jack dorsey replaced 4000 jobs with AI, Perplexity became the default AI assistant for Samsung phones(!) and Nvidia CRUSHED earnings
my entire job is to keep up with AI and i’m fucking exhausted
see you at market open tomorrow!
17. ankrgyl (Group Score: 43.4 | Individual: 43.4)
Cluster: 1 tweets | Engagement: 534 (Avg: 84) | Type: Tech
good things to learn right now:
- how to run complex infra
- how to wield LLMs (accuracy, speed, cost)
- product design 101
18. rohanpaul_ai (Group Score: 42.5 | Individual: 27.6)
Cluster: 2 tweets | Engagement: 227 (Avg: 95) | Type: Tech
Coinbase CEO: AI Agents write over 50% of code and resolve 60% of support tickets. To scale autonomy, they provide agents with stablecoin wallets for machine-to-machine payments, and really treat them like digital employees. https://t.co/jmgsjovRTm
See 1 related tweets
- @rohanpaul_ai: RT @rohanpaul_ai: Coinbase CEO: AI Agents write over 50% of code and resolve 60% of support tickets....
19. alex_prompter (Group Score: 41.8 | Individual: 41.8)
Cluster: 1 tweets | Engagement: 7880 (Avg: 728) | Type: Tech
RT @alex_prompter: 🚨 Holy shit… Stanford just exposed that every major AI company is using your private conversations to train their models…
20. BrianRoemmele (Group Score: 41.1 | Individual: 24.7)
Cluster: 2 tweets | Engagement: 205 (Avg: 418) | Type: Tech
TECHNOLOGY JOB DESKILLING EXAMPLE.
1980 you had to take classes and pass on map routes in your mind to become a London Taxi Driver.
That ended with technology.
This is an example of great skills for a job being deskilled.
This is everyone’s job now. https://t.co/CTPX0F2zOA
See 1 related tweets
- @BrianRoemmele: RT @BrianRoemmele: TECHNOLOGY JOB DESKILLING EXAMPLE.
1980 you had to take classes and pass on map ...