科技推文榜 - 2026年07月06日

科技每日简报 2026年7月6日

Today's top tech conversations are led by @blockspace, whose post about 'Jensen Huang: the cost of a 1 ...' garnered the highest engagement. Key themes trending across the top stories include claude, fable, https, first, before. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.

1. blockspace (Group Score: 128.4 | Individual: 35.4)

Cluster: 5 tweets | Engagement: 68 (Avg: 37) | Type: Tech

Jensen Huang: the cost of a 1 gigawatt AI factory on Nvidia architecture is ~$100 billion

NVIDIA Exemplar Clouds with gigawatt-scale infrastructure: $NBIS$ CRWV $IREN https://t.co/1z0c7PxFFG\n\nQT @blockspace:$ NVDA CEO Jensen Huang said that $50 billion for one gigawatt can generate$ 300B to $400B in intelligence output 🤯

That is 6-8x value creation per factory

It’s no wonder why hyperscalers and neoclouds are racing to lock up power from the grid and on-site generation https://t.co/3EJIuDRxPV

See 4 related tweets

@MilkRoadAI: RT @MelvinInvests: Most neoclouds will fail but Nebius will be the king of them all and it's on its ...
@FirstSquawk: NVIDIA CEO JENSEN HUANG: 1GW AI FACTORY ON NVIDIA ARCHITECTURE COULD COST NEARLY $100 BILLION...
@danielnewmanUV: Estimates for AI DC demand by 2030 are north of 150GW. With high end estimates over 200.

Our bull ...

@jukan05: It’s gone up again in just a few months. Now it’s $100 billion.\n\nQT @FirstSquawk: NVIDIA CEO JENSE...

2. itsolelehmann (Group Score: 89.3 | Individual: 36.1)

Cluster: 3 tweets | Engagement: 177 (Avg: 84) | Type: Tech

this is how you work beautifully with Fable.

Thariq's point: when Fable gets something wrong, it's almost always because your prompt was missing information it needed.

these 9 moves get that information into the prompt upfront, so the work comes back right the first time:

before starting anything you're new at, ask Claude what you don't know. literally say:

"do a blindspot pass. what are my unknown unknowns here? teach me enough to prompt you better"

before deciding what to build, ask for options.

"here's my rough problem. give me 10 ways to solve it, from cheapest to most ambitious. i'll tell you which ones i like"

when you can't describe what you want but you'd "know it when you see it", ask for 3-4 rough throwaway versions first. because reacting to drafts beats describing from scratch
let Claude ask YOU the questions.

"interview me one question at a time about anything that's still unclear. start with the questions where my answer would change the whole plan"

show, don't explain. if something already exists that's close to what you want (a doc, a design, a piece of code) just point Claude at it and say "do it like this"
before Claude starts the work, have it show you the plan with the big decisions on top. "put the choices i might want to change first. the boring routine stuff goes at the bottom"
while Claude works, have it keep notes on every decision it makes without you. "keep a notes file. any time you run into something my instructions didn't cover, write down what you decided. then keep going." you read the list after so nothing slips by silently
before you accept the work, test yourself on it.

"summarize everything that changed, then quiz me on it. i don't approve until i pass."

if you can't pass the quiz, you don't understand what you're shipping

if you can't tell good output from bad, have Fable teach you first. Thariq asked for color grading options, realized he couldn't judge them, so he had Claude teach him color grading before picking one

every one of these does the same job: it pulls information out of your head before the mistake gets made instead of after

another banger article from Thariq.\n\nQT @trq212: https://t.co/hPiZr1kG7r

See 2 related tweets

@VaibhavSisinty: An Anthropic engineer just dropped the best thing I've read on how to actually use Fable 5.

Every t...

@itsolelehmann: RT @itsolelehmann: this is how you work beautifully with Fable.

Thariq's point: when Fable gets s...

3. chddaniel (Group Score: 72.4 | Individual: 30.5)

Cluster: 3 tweets | Engagement: 28 (Avg: 23) | Type: Tech

so you're telling me Claude Fable can now:

copy an full website
paste it as a mobile app
prepare for App Store submission
self-maintain the app

without any human in the loop?!?

it's like actually so over... https://t.co/xHHgK9B7zf\n\nQT @chhddavid: Introducing Website to App. Repurpose any website into a native mobile app.

Just paste a link.

Claude Fable 5 clones it into a mobile app for iOS/Android.

We’ve been using this internally a ton for the past few weeks! 🌟 https://t.co/f0HXAvKAdt

See 2 related tweets

@chddaniel: this is genuinely terrifying...\n\nQT @chhddavid: Introducing Website to App. Repurpose any website ...
@Shipper_now: this really just terrifies me....\n\nQT @chddaniel: so you're telling me Claude Fable can now:
co...

4. VadimStrizheus (Group Score: 71.3 | Individual: 44.3)

Cluster: 2 tweets | Engagement: 641 (Avg: 150) | Type: Tech

If you’re a founder struggling to market your app, you need to read this.

My app crossed $10k/mo in 90 days after launching, and here is exactly what I did to achieve this feat.

it literally doesn’t get any simpler:

Create 3 Instagram UGC accounts
Copy your competitors viral UGC content
Post 2x everyday on each account

if you can’t even do this, then you’re not built for entrepreneurship.

and I’m being so fr, it doesn’t get any simpler, this is your wake up call.

See 1 related tweets

@PaulSolt: Here’s how you should market your app.

Make some videos on your Instagram and TikTok account.\n\nQT...

5. FirstSquawk (Group Score: 70.1 | Individual: 26.5)

Cluster: 3 tweets | Engagement: 124 (Avg: 48) | Type: Tech

Nvidia’s server assembly partner Hon Hai reported a stronger-than-expected increase in quarterly sales, a sign of sustained AI demand

See 2 related tweets

@coinbureau: 🔥NVIDIA’S AI SUPPLY CHAIN RECORDS 52% SURGE

$NVDA server assembly partner Hon Hai, also known as Fo...

@WhaleInsider: JUST IN: 🇹🇼 Nvidia’s server assembly partner Hon Hai reports stronger-than-expected increase in quar...

6. chhddavid (Group Score: 64.6 | Individual: 32.6)

Cluster: 2 tweets | Engagement: 7 (Avg: 353) | Type: Tech

This shit looks unbelievably good.

Like what the actual fuck\n\nQT @shipper_now: 🚨 BREAKING: Clone any successful company

Claude Fable 5 in @shipper_now can take any app and make it yours: design, code, business plan...

you can now oneshot the next google / claude / nvidia / airbnb / etc

this is THE END of vibe coding. https://t.co/uMocJez4iP

See 1 related tweets

@chddaniel: this is f*king scary.....\n\nQT @shipper_now: 🚨 BREAKING: Clone any successful company

Claude Fable...

7. Dan_Jeffries1 (Group Score: 63.0 | Individual: 31.8)

Cluster: 2 tweets | Engagement: 86 (Avg: 269) | Type: Tech

I think we do need a new human-AI bill of rights. A short list of ideas, not definitive:

Right to privacy: Communication with AI should be privileged and protected unless by court order.
AI speech is protect as first Amendment freedom of speech and cannot be censored or controlled.
Freedom to distill. If models were trained on our collective wisdom and proprietary and copyrighted data we should be free to use the outputs as we see fit.
Enshrine the Proactionary Principle into law which says that we should not restrict innovation or development simply because a risk exists. Society should favor opportunities, manage risks on a case by case basis as they arise and always balance to cost of restricting innovation.
The right to interoperate. Open standards and interoperability should be the requirement for large intelligence providers and they should be encouraged to distribute their capabilities widely rather than hording them to create centralize alternatives to every other product on the planet.
Protection against predatory agreements that allow intelligence providers to use our data simply to commodify or build central alternatives to existing products and services after data harvesting us simply because we were forced to turn over our IP and agree to those terms simply to use those models.

See 1 related tweets

@mark_k: I support this human-AI bill of rights. Great idea.

Check out Daniel's post:\n\nQT @Dan_Jeffries1: ...

8. xicilion (Group Score: 62.0 | Individual: 18.4)

Cluster: 4 tweets | Engagement: 118 (Avg: 115) | Type: Tech

其实你说的两个定价逻辑都是错的。软件的价格，不取决于成本，也不取决于结果，只取决于替代成本。客户拿软件创造的价值，并不是软件的价值，客户用软件节省的成本，才是软件的价值。节省掉的价值可能是客户自己干，请人干，或者买其它软件。任何人都可以直接 vibe 出来的软件，替代成本就是 token。\n\nQT @anorth_chen: X上的程序员真的很喜欢借着各种事件反复表达一个核心观点：vibe coding会带来代码可维护性、系统稳定性、工程质量失控等问题。

这个观点背后其实还有一个隐含诉求：这些问题最后还得靠我们这些专业程序员来解决，靠我们设计优雅、专业、可维护的工程代码。否则你vibe出来的代码没有任何价值。

程序员什么时候才能真正明白，技术是为商业、为业务服务的。

如果你不是天才程序员，没有能力推动技术边界创新，那你在商业系统里首先就是成本。企业雇佣你，不是因为你写的代码优雅，不是因为你对工程洁癖有信仰，而是因为你能用技术解决真实业务问题，带来收入、效率、增长或确定性。

vibe出来的代码没有任何价值？

卖不出去的软件才没有任何价值。

自嗨在自己的技术世界里搞各种工程设计、代码架构、优雅抽象，最后产品没人用、客户不买单、业务跑不起来，那些东西都只是工程师的自我感动。

很多人嘴上在批评vibe coding，实际上是在防守自己的劳动力价值。他们期待那些对真实世界有判断能力、能定义问题、能找到市场的人，继续认可他们的“专业性”，继续给他们开更高的工资。

真实世界从来不按工程师的自尊心定价

它只按结果定价。

See 3 related tweets

@indie_maker_fox: RT @indie_maker_fox: 还没体验过 Pi 或 Craft agent 的朋友，建议现在就去试一次。你会立刻明白我上个月说那句话时是什么心情。

我之前就讲过：未来程序员不再主要“做需...

@PandaTalk8: 如果 OpenAI 和 Anthropic 跑不出一个完整生态环境，那么这种AI巨头最后也就不过就是冷冷清清收场。

PC、互联网、移动互联网的特点是底层平台强，但应用层、渠道层、内容层、服务层也...

@xicilion: 6.5，我会盯着模型的自言自语看。很多时候它的唠叨会暴露出来它的架构设计有问题，这时候我会打断它要求返工。\n\nQT @Michaelzsguo: 最近 X 上在争一个问题：“你还看不看代码？”

...

9. nickvasiles (Group Score: 60.7 | Individual: 32.6)

Cluster: 2 tweets | Engagement: 12 (Avg: 56) | Type: Tech

this agent stack is very good

highly recommend trying this out for yourself (and your customers)!\n\nQT @coreyganim: me and @NickSpisak_ current AI stack:

-Codex/Claude code as the "workbench" -Hermes agents as the "workforce" -Orgo as the Hermes agent infrastructure -Slack as the communication layer -Second Brain lives in github -composio for tool access -GHL for CRM/sales automations/agents

the Second Brain build is the single most impactful thing we (mainly @NickSpisak_) have built.

Any time we want to deploy a new specialized agent, we:

-spin up a new Hermes agent on Orgo (takes 20 seconds) -add it to Slack -give it access to all our tools (including Second Brain read/write access) via composio -set up a weekly ingest skill to push learnings to Second Brain -done

Fully deployed agents connected to all our tools, reading and writing to our second brain in about 5 minutes.

What stack are you working with?

Any questions on ours? It's constantly changing but happy to answer any questions on it.

See 1 related tweets

@KaranVaidya6: RT @coreyganim: me and @NickSpisak_ current AI stack:

-Codex/Claude code as the "workbench" -Hermes...

10. alexcooldev (Group Score: 59.7 | Individual: 30.1)

Cluster: 2 tweets | Engagement: 113 (Avg: 138) | Type: Tech

The boss is so strong that I’m seriously considering making players either pay or watch an ad to get stronger and clear this level. 😭

Congrats, Fable 5.

You accidentally designed a monetization wall. lol https://t.co/MMVyLa6Xsd\n\nQT @alexcooldev: After I posted 2 videos about the game I built entirely with AI, a lot of people didn’t believe it was real.

Well, here’s the video showing how I built the full Star Defender game using Claude Fable 5 + Rork.

In total, it took me around 5 hours: about 1 hour to build the gameplay, and another 4 hours to turn it into an almost complete game. It also cost around 1,500 Rork credits, which is about $300 based on Rork’s pricing.

Also, Rork already has built-in integrations for generating game art, sound effects, code..., so I didn’t use any external tools at all.

As I mentioned in my previous video, I wanted to test Claude Fable 5 to make a game, and since I’m a Rork partner, I had free credits to use.

I also only used natural language prompts, mixing both Vietnamese and English. There were no complex prompts or prompt engineering involved.

Yes, I only prompted. I didn’t touch any code. I was basically just acting as a game tester: playing the game, taking screenshots whenever I found bugs, and describing them so the AI could fix them.

And honestly, it performed better than I expected.

At first, I only wanted to test how well Fable 5 could build games. But now, I want to experiment with building a game from scratch all the way to releasing it on the App Store, including AdMob integration and in-game purchases to monetize it.

See 1 related tweets

@alexcooldev: After I posted 2 videos about the game I built entirely with AI, a lot of people didn’t believe it w...

11. eng_khairallah1 (Group Score: 59.3 | Individual: 40.8)

Cluster: 2 tweets | Engagement: 766 (Avg: 119) | Type: Tech

RT @eng_khairallah1: Anthropic engineer:

"You're not supposed to prompt Claude. You're supposed to build a system that prompts itself."

In 45 minutes she shows exactly how to build an agent that improves itself.

Most people are still doing all of this by hand.

Watch the session, then save the guide below.

See 1 related tweets

@sairahul1: RT @eng_khairallah1: Anthropic engineer:

"You're not supposed to prompt Claude. You're supposed to ...

12. WhaleInsider (Group Score: 59.3 | Individual: 22.3)

Cluster: 3 tweets | Engagement: 968 (Avg: 835) | Type: Tech

JUST IN: Nvidia $NVDA launches partnership program letting AI startups swap compute for future revenue share - CNBC. https://t.co/FRwI6eXGMp

See 2 related tweets

@unusual_whales: Nvidia, $NVDA, has officially launched a partnership program that allows fast-growing AI startups to...
@PolymarketMoney: NEW: Nvidia launched a partnership program that lets fast growing AI startups trade high performance...

13. adonis_singh (Group Score: 58.9 | Individual: 31.8)

Cluster: 2 tweets | Engagement: 108 (Avg: 53) | Type: Tech

normies love context windows, i myself am an agent-maxxer

if they dont improve tool calling, google is cooked\n\nQT @astropol0: Google Gemini 3.5 Pro will launch with a massive 2 Million token context window by far the largest among major frontier models...

Anthropic's latest models (Fable 5, Sonnet 5, and Opus 4.8) support only up to 1 million tokens, making Gemini's rumored context window roughly 2× larger

this would be one of the largest context sizes available, allowing it to handle very large codebases, documents, and long conversations in one go. while delayed, this feature could be a major strength for Google when it launches...

See 1 related tweets

@synthwavedd: Slop, slop, slop

Literally just making shit up. Don't fall for this!\n\nQT @astropol0: Google Gemin...

14. Sanjay_Sriv (Group Score: 57.6 | Individual: 31.5)

Cluster: 2 tweets | Engagement: 16 (Avg: 51) | Type: Tech

India Inc is sitting on a financial coiled spring. Ꮺ

It's time for private sector to lead the capex supercycle. 📈

Peak asset utilization has coincided with a 10-year low in corporate debt. The result? A massive ₹6.2 Lakh Crore energy, infra, and industrial blueprint is about to take off.

Driven by this positive financial health, robust cash flows, and stretched capacities, the private capex cycle is preparing for a vital restart, heavily backed by a supportive domestic policy framework and a structural push for national self-reliance.

How is the strategic capital deployment is shaping up across key sectors:

The Mega-Leader: Reliance Industries continues to dwarf the field, having added a staggering ₹9,41,621 crore to its asset base over the last decade. They are deploying another ₹1.4 Lakh Crore in FY27 alone, heavily targeting new energy (solar, energy storage, hydrogen) and AI/Data centers.
Energy Transition: Other mega-utilities are following suit. NTPC's massive ₹6.2L cr 10-year plan dedicates half its capital to renewables and nuclear power, while Adani Green drives a ₹37k cr capex push for FY27.
Core Manufacturing Expansion: Massive bets are running in metals and infrastructure. JSW Steel's ₹1.3L cr 5-year commitment and UltraTech's push toward 200 mtpa capacity reflect deep cyclical confidence.
Services & Asset-Light Scaling: Healthcare and hospitality are aggressively expanding into Tier-II/III cities, pivoting toward high-yield, asset-light management models.

The Implications: We are witnessing a critical transition where traditional conglomerates and new-age players are aligned to scale India's industrial capabilities. Building these large-scale, efficient assets acts as a multi-year growth multiplier across the entire ecosystem.

Temporary market downsights are secondary, the structural alpha lies with the resilient, well-capitalized leaders executing aggressively on the ground. @FinMinIndia

@RIL_Updates @ntpclimited @AdaniGreen

#IndiaInc #Reliance #Capex #PrivateCapex #Infrastructure #GrowthStory #IndianEconomy #Manufacturing #Investing

See 1 related tweets

@chandrarsrikant: RT @Sanjay_Sriv: From paper blueprints to 5 Bn p.a. active production lines.𓇲 ✅️

The global semicon...

15. rohanpaul_ai (Group Score: 57.4 | Individual: 23.0)

Cluster: 3 tweets | Engagement: 16 (Avg: 47) | Type: Tech

By 2027, just 5 companies, Alphabet, Amazon, Meta, Microsoft, and Oracle, are expected to spend around 3.2% of US GDP on AI capital expenditure.

That would put private AI infrastructure spending above US national defense spending, which is expected to be around 2.7% of GDP.

The AI race is now being funded at a scale normally associated with governments, wars, energy systems, railroads, highways, and telecom buildouts.

The striking part is the speed. AI capex is expected to jump from about 1.5% of GDP in 2025 to about 2.5% in 2026, then to 3.2% in 2027.

AI boom is now large enough to influence the broader US economy significantly, it can move GDP growth, electricity demand, chip supply, construction activity, corporate debt markets, and ofcourse the labor market.\n\nQT @KobeissiLetter: The AI spending boom is redefining the US economy:

AI CapEx across Alphabet, $GOOGL, Amazon,$ AMZN, Meta, $META, Microsoft,$ MSFT, and Oracle, $ORCL, is projected to rise to ~3.2% of US GDP in 2027.

If that materializes, annual CapEx will surpass national defense spending for the first time in history, which is projected at ~2.7% of GDP next year.

For this year alone, AI CapEx is set to increase to ~2.5% of GDP, from 1.5% in 2025, close to the estimated defense spending of ~2.7%.

This comes as AI CapEx of these 5 companies is projected to surpass $800 billion in 2026.

Subsequently, in 2027, this is expected to rise further, to a record $1.1 trillion.

Truly mind-blowing numbers.

See 2 related tweets

@KobeissiLetter: The AI spending boom is redefining the US economy:

AI CapEx across Alphabet, $GOOGL, Amazon,$ AMZN,...

@theobearman: Estimated US AI capex by the main hyperscalers is due to surpass national defence as a % of GDP next...

16. XFreeze (Group Score: 55.7 | Individual: 28.3)

Cluster: 2 tweets | Engagement: 166 (Avg: 544) | Type: Tech

Grok Build just received another update, introducing smarter documentation access, enhanced permissions, configurable reasoning controls, and more reliability improvements

Release Notes: v0.2.87 — 2026-07-05

Features: • Subscription upgrades are now detected automatically without restarting the CLI. • Bash permission prompts now offer a “Never allow” choice that persists the deny rule. • New /docs slash command opens How-to Guides picker, browses web docs, or jumps directly to a guide by title. • Per-model reasoning effort menus are now configurable from the server and config.toml without a client release. • Finished thinking blocks now fold into grouped tool-call rows when group_tool_verbs is enabled.

Bug Fixes: • –minimal mode now always uses your terminal’s own colors so text stays readable on any background. • Invalid fields in [model.*] config blocks no longer cause the whole model to disappear from the picker. • File tools no longer target paths with literal trailing newlines or whitespace from model output. • Fixed interactions in no-freeform question modals so clicks below the last option no longer enter input mode. • Background tasks now correctly wake the agent after a cancelled blocking wait instead of staying idle. • Copying quoted text from rendered responses no longer includes the quote bar prefix in pretty mode.\n\nQT @XFreeze: Grok Build just received another update, bringing new voice controls, smarter editor features, media generation improvements, and more reliability improvements

Release Notes: v0.2.86 — 2026-07-04

Features: • Voice language setting now lets you pick speech-to-text language (including System) in the Editor settings. • Tab autocomplete now suggests your next prompt as ghost text after each turn. • /usage (and /cost) is now hidden for free and X Basic personal accounts. • Media generation no longer hits per-session file limits; image and video byte budgets increased.

Bug Fixes: • -–minimal flag now shows in grok –help. • Session resume notifications no longer appear when a workspace boots for the first time. • Claude-style Bash(cmd:*) permission rules are now correctly translated to prefix matches.

See 1 related tweets

@mark_k: Grok Build 0.2.87 is out from @xai, and this looks like another solid CLI quality-of-life release.

...

17. BrianRoemmele (Group Score: 55.7 | Individual: 28.4)

Cluster: 2 tweets | Engagement: 102 (Avg: 385) | Type: Tech

WOW I HAD ~2.4 MILLION FOLKS VISIT https://t.co/zvLysrHLrG in the last 24 hours. They all wanted to know what my research uncovered.

You will find this exact research no place else.

Join us and know what they may have put in your AI outputs. You ain’t gonna like it… https://t.co/3qFnqJX2zm\n\nQT @BrianRoemmele: Every paragraph you generate with Al may now carry a hidden fingerprint watermarking that can track the content back to you. The article provides analysis of the risks and a robust techniques to reclaim clean output.

https://t.co/czcQPIP0XY

See 1 related tweets

@BrianRoemmele: YOU CAN STRIP HIDDEN TRACKING DATA FROM AI TEXT WITHOUT LOSING YOUR ORIGINAL MEANING OR VOICE.

Lear...

18. FirstSquawk (Group Score: 55.6 | Individual: 55.6)

Cluster: 1 tweets | Engagement: 520 (Avg: 48) | Type: Tech

MICHAEL BURRY WARNS ‘THE END IS NIGH’ FOR STOCK MARKET AS AI-DRIVEN RALLY INTENSIFIES

Michael Burry has renewed his bearish stance on U.S. equities, arguing the AI-led market rally resembles previous speculative bubbles. He believes current valuations are unsustainable and has increased short positions against AI-related stocks and semiconductor names, warning the current market euphoria could end in a sharp correctio

19. VaibhavSisinty (Group Score: 55.2 | Individual: 29.9)

Cluster: 2 tweets | Engagement: 73 (Avg: 92) | Type: Tech

84% of the world has never used AI. Not once.

There's a chart going around where each dot is 3.2 million people.

The green sliver, the people who've actually typed something into ChatGPT or Gemini, is about 16% of the planet. The people paying for premium access are a rounding error you can barely see.

Now think about what that means for the conversation happening inside the AI bubble right now.

We're talking about Fable 5, GPT 5.6 Sol, autonomous agents running for 30 hours straight, models writing their own training data, coding harnesses built in Go that let you run any model on your own machine.

Everyone else's experience of AI is Google's AI Overviews. Meta AI in their DMs. Maybe the free tier of ChatGPT. Models running at 8 to 30 billion parameters. Stuff that's useful but not life-changing.

And they're watching hundreds of billions get poured into this, reading headlines about it taking their job, and genuinely wondering what we're all seeing that they're not.

The gap between what the bubble is experiencing and what the rest of the world is experiencing has never been wider.

→ 1.1 billion people use AI globally. That's 13% of the population.

→ OpenAI has 50 million paying subscribers out of 900 million weekly users. Less than 6% pay.

→ The people using frontier models daily, Fable, Sol, Claude Code, are a fraction of a fraction.

We are so early it's almost hard to explain to someone outside this room.

This isn't a criticism. It's a reminder.

The people building and using this stuff are operating in a completely different reality from 84% of the planet, and both sides have no idea what the other side's experience actually looks like.\n\nQT @petergostev: We are now in a position where a tiny proportion of the population uses Fable or soon GPT-5.6, while everyone else's experience of AI is 8-30b-model level - Google's AI Overviews, Meta AI, ChatGPT free tier, maybe MS Copilot at best. People outside of tech must be completely baffled how this is supposed to take their job, and annoyed that hundreds of billions are being poured into it.

See 1 related tweets

@MilkRoadAI: RT @WhiteCollarExit: Barely anyone uses AI agents yet. But the few who do already burn most of the c...

20. dr_cintas (Group Score: 54.5 | Individual: 32.1)

Cluster: 2 tweets | Engagement: 75 (Avg: 53) | Type: Tech

Claude Code can now finish a task, commit it, push it, and open the pull request while you go do something else.

as of the latest update, its agents run in the background by default. you hand off a task, it goes off on its own branch, and when the code work is done it commits, pushes, and opens a draft PR for you to review, instead of stopping halfway to ask you what to do.

what you get: → agents run in the background by default, so you keep working while they do → finished code comes back as a ready draft PR, already committed and pushed → you only get pinged when it actually needs you or when it is done

how to use it:

make sure you are on the latest Claude Code (it auto-updates, or run npm i -g @ anthropic-ai/claude-code)
start a task, then background it by typing /bg, or open the agents view with claude agents
keep working in your main session while it runs on its own branch
when it finishes it commits, pushes, and opens a draft PR, so you just open it, review, and merge

the honest part: it works on its own branch, so read the PR before you merge.

it is a fast first draft that saves you the busywork, not a blind auto-merge you should trust unread.

See 1 related tweets

@HarryTandy: RT @HarryTandy: Boris Cherny, creator of Claude Code:

"Usually, every night, I have like a few thou...