This is a first hand announcement from Anthropic, with more detail in a post on its website. The site says the 5% commitment is subject to stockholder approval, and that the fund's governance and valuation milestones will be shared in the coming months.

On October 18, Anthropic posted that it is committing a portion of its shares to the Anthropic Public Benefit Fund, a new wholly owned entity dedicated entirely to the public benefit. It said the fund starts with 5% of its shares, and that for every 50% increase in the company's value over a set baseline it can contribute up to an extra 2%, at the board's discretion.

Anthropic wrote that, in total, roughly 20% of its shares are committed to the public good. It said its cofounders have pledged to donate 80% of their equity, and that around half its employees donate equity through a matching program run since its founding.

The company's website post says the 5% commitment is subject to stockholder approval. It says the fund is a wholly owned LLC rather than a 501(c)(3), to allow different forms of giving, and that none of its equity will be used for lobbying or political giving. It gives examples such as large donations to nonprofits or direct distributions to specific populations, and notes Anthropic has already committed $200 million to studying AI's effect on workers.

Anthropic said it will share the fund's governance and valuation milestones in the coming months, along with a public benefit report.

Anthropic's Jack Clark wrote that, between this and prior commitments, roughly 20% of all Anthropic equity will be dedicated to the public good, and that the funds will help more people share in AI's benefits and pay for dealing with its risks. David Sacks quoted him: You've put 20% of your equity in an LLC. Whether it serves the public good remains to be seen.

Elon Musk replied to Anthropic's post: I recommend giving the shares directly to the public. That evening he added, under the same post: This is what I will be doing. He did not say which shares or how. Anthropic lists direct distributions to specific populations as one possible use of the fund, but has not announced any plan to hand shares to the public.

Watch for the stockholder vote on the 5% commitment and the promised details on the fund's governance, baseline valuation and milestones.

Also watch how investors in future funding rounds treat the committed shares, since the stake can grow as Anthropic's value rises. Its first grants or distributions, especially on AI's effect on workers, will show how the fund intends to spend.

This is secondhand: Bloomberg is relaying a Telegraph report, and Microsoft did not comment to either outlet. The Guardian and the Financial Times also reported the withdrawal, linking it to power supply delays at the Essex site.

On October 18, Bloomberg posted that Microsoft has pulled out of a UK data center deal with British neocloud Nscale at a critical time for the startup's IPO hopes, citing the Telegraph.

According to Bloomberg's report, the Telegraph did not give a specific reason, but said the site in Loughton, Essex has faced a series of delays and that authorities told Nscale they do not expect a power supply until the 2030s. It also reported that Nscale's flagship site in Norway is six months behind schedule, with 6.4 megawatts of a planned 30 megawatts running. Microsoft had been set to be the anchor tenant at the Essex site, and Nscale relies on Microsoft and Anthropic for about 85% of its contracts, according to its IPO filing. Bloomberg said Nscale is seeking to raise as much as $3 billion in its IPO.

An Nscale spokesperson told the Telegraph the company remains committed to the Loughton project and the UK, is exploring alternative power sources, and that its relationship with Microsoft remains strong and long term, with the grid issue having no impact on its other contracts.

The Guardian reported that ministers had heralded the project as the UK's largest AI data center when it was announced in 2025, and that it was planned to open next year. The FT reported Microsoft cancelled its agreement after it emerged the site was unlikely to have power until after 2030.

On October 19, Microsoft gave its first comment. A spokesperson told Reuters that this decision relates to a single site and that Microsoft remains committed to its partnership with Nscale. Reuters said the site was due to begin operating early next year with 23,040 Nvidia GB300 GPUs providing 50 megawatts for Microsoft's Azure services in Britain. Polymarket relayed the news the same morning.

In speculation, the account Serenity, which says it researches AI and chip supply chains and may hold the stocks it discusses, argued the exit is bullish for neoclouds with energized capacity, such as Nebius or IREN, because Microsoft still needs the compute and will buy it elsewhere.

Watch whether Nscale changes the size or timing of its IPO, and whether it finds a new tenant or alternative power for the Essex site as its spokesperson suggested.

Also watch for comment from Microsoft and the UK government, which promoted the project, and whether other UK data center plans face similar grid delays. Investors will also look at Nscale's reliance on Microsoft and Anthropic as debt costs rise for neoclouds.

The plugin is real and official: it is published by the EpicGames organization on GitHub and was added to Anthropic's official Claude Code plugin marketplace in June 2026. RoundtableSpace presents it as new, but this is a resurfaced item rather than a new release, and no first hand Epic or Anthropic post about it was found.

On October 18, RoundtableSpace posted that Epic Games released an official plugin that lets Claude Code control Unreal Engine directly. It wrote that the plugin connects Claude Code to the Unreal Editor and exposes hundreds of editor tools across more than 30 toolsets, including actors, Blueprints, materials, Niagara effects, Control Rigs, Sequencer, State Trees, widgets, the Gameplay Ability System and automation testing. The post had about 2,500 bookmarks, and Epic CEO Tim Sweeney reposted it.

The plugin's GitHub repository belongs to the EpicGames organization and was created on June 8, 2026. A pull request adding it to Anthropic's official plugin marketplace was merged on June 26, 2026, listing the source as Epic's repository under the MIT license.

According to the repository, users install it from the official marketplace inside Claude Code, enable Unreal's ModelContextProtocol and AllToolsets plugins, and start the MCP server with the editor running. It supports macOS, Linux, and Windows through Git Bash or WSL. Anthropic's plugin page describes safety rules such as saving before and after bulk changes and running tools one at a time to avoid deadlocks.

Epic's documentation describes Unreal MCP as a way for any MCP compatible agent, such as Claude Code or Cursor, to drive the editor over a local connection. Sweeney said on October 12 that Epic is retooling its developers for maximum AI acceleration.

Watch whether Epic expands the toolsets or brings similar support to other agents, given that its documentation already describes Unreal MCP as working with any compatible agent.

Also watch how game studios use agents inside the editor, for example for level building, materials or automated testing, and whether Epic's broader push toward AI acceleration changes its tools for Unreal and Fortnite creators. Install counts on Anthropic's plugin page may show whether the renewed attention brings new users.

These are first hand posts from dax of the OpenCode team, following the company's October 17 disclosure of a breach tied to a distillation attack. He later said OpenAI's Daybreak Blue access, which reduces refusals for approved defensive work, was approved the day before and that he does not think anyone is refusing to help. David Sacks later cited the episode to argue for decentralized cyber defense, without addressing that clarification.

On October 18, dax of OpenCode posted that they have been nonstop trying to attack our systems for the past 48 hours. He wrote that everything is getting hardened, but that OpenCode basically needs agents going 24/7, because that is what the attackers have.

Asked by Rhys Sullivan whether frontier models' cyber safeguards were blocking him, dax replied that it was a bit frustrating mid incident when trying to investigate, that the models did quite a bit and then started refusing, and that OpenCode now has access to Daybreak Blue, so it is better.

Andrew Curran framed this as OpenCode's defensive models refusing to help fight attacking agents because analyzing the attacks triggered safety guardrails, and compared it to a similar problem he said Hugging Face had faced. In a later reply, dax said OpenCode coincidentally applied for Daybreak a few days earlier, that it was approved the day before, and that he does not think anyone is refusing to help them.

OpenAI's help center describes Daybreak Blue as an approved tier that reduces refusals for verified defensive work such as incident response.

David Sacks quoted Curran's post, writing that it illustrates the limits of centralized approaches to SI safety in cyber. He said the guardrails backfired while attackers found ways around restrictions, that Chinese models are only about 6 to 9 months behind the American frontier in cyber, and that the only durable response is to make powerful models available so defenders can run their own cyber defenses. In July, Hugging Face said hosted frontier models blocked its analysis of an AI driven intrusion and that it switched to the open weight model GLM 5.2, CNBC reported.

On October 17, OpenCode disclosed a breach tied to a distillation attack, and it has not published a full incident report.

Watch for OpenCode's full incident report, including who was behind the attacks, what was exposed and whether the 48 hours of attempts caused further breaches.

Also watch how AI labs balance safety limits with defenders' needs, since programs like Daybreak require approval in advance. Other companies under attack may push for faster access, and labs may face questions about whether attackers are using agents with fewer restrictions.

This is secondhand: Polymarket is relaying a Seoul Economic Daily report that cites industry sources, which Reuters also carried. The report says TSMC will make the base die while Samsung keeps making the memory itself, a narrower arrangement than Polymarket's teaming up framing, and none of the companies has confirmed it.

On October 18, Polymarket posted that Samsung and TSMC are reportedly teaming up to produce Nvidia's next generation high bandwidth memory chips.

The Seoul Economic Daily reported that day, citing industry sources, that Samsung Electronics will use TSMC's 3 nanometer process to make the base die for NVHBM, the custom high bandwidth memory Nvidia unveiled in August. The base die sits under the stacked memory and works as a logic chip. The paper said Nvidia asked all three of its NVHBM suppliers, Samsung, SK hynix and Micron, to make the base die on TSMC's process and to standardize the memory dies on sixth generation DRAM, known as 1c.

According to the report, Nvidia made the request because different manufacturing methods across suppliers make it hard to get consistent performance from one design, and standardizing reduces the work of validating each supplier's product. It said Samsung's chip division accepted after an internal review, and that this would be the first time Samsung and TSMC have partnered at the HBM manufacturing stage.

Samsung has so far made both the memory and base dies for all of its HBM, using its own 4 nanometer foundry process for HBM4. The paper said it will keep that approach for standard HBM and for custom memory for other customers such as AMD, and is checking whether its own foundry could also make NVHBM base dies in case demand exceeds TSMC's capacity. Reuters carried the report, attributing it to the Seoul newspaper.

Watch for confirmation from Samsung, TSMC or Nvidia, and whether SK hynix and Micron also move their NVHBM base dies to TSMC as the report says Nvidia asked.

Also watch whether TSMC's 3 nanometer capacity becomes a new bottleneck for AI memory, given that the report says Samsung is preparing its own foundry as a backup. Memory prices, already driving up costs for phones and PCs, could be affected by how quickly NVHBM ramps.

This is a first hand reply from Tibo, who leads Codex at OpenAI, but it gives no timeline. It responds to a user whose dot stopped appearing in the Codex desktop app.

On October 18, Tibo replied to a user named argofowl, who had written that their dot stopped showing up in the Codex desktop app.

Tibo answered: EU sorry. He wrote that OpenAI is making it so that if someone created a dot elsewhere and finds themselves in the EU, it will keep working, and that dots are coming to the EU in the future.

The reply suggests the dot disappeared because the user was in the EU, where dots are not yet available, though Tibo did not say why. He did not give a timeline for either change or say whether the fix for existing dots has shipped.

Dots are OpenAI's personal AI agents, each with its own character, which appear in ChatGPT and Codex. Tibo has posted several updates about them during what he calls a 28 day push, including an October 17 post saying his work and personal dots coordinated with his accountant over email, and a reply saying he was working on items from a user's wishlist that included dots for Plus users.

OpenAI has not said what has held up dots in the EU. Companies have often delayed AI features in Europe while assessing local rules.

Watch for an official OpenAI announcement on dots in the EU, including a date and any limits on features compared with other regions.

Also watch whether existing dots keep working for users who travel to or live in the EU, as Tibo promised, and whether OpenAI explains the delay. Rivals with agent products in Europe may use the gap to win users before dots arrive.

This is secondhand: zerohedge is sharing an exhibit from Morgan Stanley research, with its own warning about the debt market added. Crypto Briefing reported the same Morgan Stanley figures, including that about half of the projected spending is expected to be debt financed.

On October 18, zerohedge posted that $1.6 trillion in projected AI capex drives 5% of 2027 sales and 25% of sales growth, that half of that capex is debt funded, and that it would be a pity if something happened to the debt market.

The attached Morgan Stanley Research exhibit says hyperscaler and neocloud AI capex nearly doubles from $1.0 trillion in 2026 to $1.6 trillion in 2027. It estimates that only 65% of that spending lands in S&P 500 revenue, about $1.0 trillion, equal to 5% of the index's 2027 sales. Nvidia alone takes about $690 billion, two thirds of the total. The exhibit puts 2027 S&P 500 sales growth at 8.6%, or 6.6% without AI capex, so AI spending supplies about 24% of the growth.

Crypto Briefing, reporting on the same research, said Morgan Stanley puts the 2027 figure at about $1.58 trillion including roughly $195 billion in lease commitments, rising to $1.81 trillion in 2028, and that half the investment is expected to be financed with debt.

In a post minutes earlier, zerohedge shared a second Morgan Stanley chart of the S&P 500's 27% third quarter earnings growth. It shows the Magnificent Seven contributing 26.5% of that growth and other AI beneficiaries 34.9%, with Micron the largest single contributor at 18.8% and Nvidia at 15.0%.

Watch hyperscaler earnings in the coming weeks for any change to 2027 capex plans, since Morgan Stanley's estimate makes the index's growth unusually dependent on them.

Also watch credit markets for the companies borrowing to build data centers, which Goldman has flagged as a weak point after CoreWeave's slide. If debt funding tightens, the drag could show up in Nvidia's revenue and in the broader index.

This is secondhand: The Kobeissi Letter is relaying IDC's preliminary figures, which IDC released on October 14, so the data is four days old. Its numbers match IDC's release, and Counterpoint and Omdia separately estimated a decline of about 20%.

On October 18, The Kobeissi Letter posted that global smartphone purchases have collapsed, with shipments down 22.0% year over year in the third quarter of 2026 to 255.3 million units, their largest quarterly decline on record.

The figures match IDC's preliminary data released on October 14. IDC said shipments fell from 327.3 million a year earlier, the highest year over year decline of any quarter to date. Samsung shipped 57.8 million units, down 5.8%, and Apple 57.3 million, down 4.0%, while Xiaomi fell 29.2% to 30.7 million, OPPO 34.5% to 25.6 million and vivo 31.5% to 19.1 million. Other brands fell 32.5%.

IDC's Nabila Popal said memory costs have surged about 350% since last year and are still rising, putting pressure on vendor margins, and that the market has not yet reached the bottom. IDC expects growth to stay constrained through 2027, with any recovery pushed to 2028 at the least. The Kobeissi Letter added that vendors are raising prices and cutting budget models and that consumers are keeping phones longer.

Counterpoint Research put the decline at about 20%, and Omdia also estimated 20%. IDC's Francisco Jeronimo said many buyers are keeping their current devices longer as higher prices reach the shelf. Samsung and Apple together took about 45% of the market, as larger vendors held up better than Chinese brands focused on cheaper phones.

Watch fourth quarter shipments and holiday pricing, since IDC says the market has not hit bottom, and whether memory prices keep rising as AI data centers absorb supply.

Also watch whether Chinese brands such as Xiaomi, OPPO and vivo cut more budget models, and how Samsung and Apple use their scale to protect share. Memory makers' earnings will show how much AI demand is driving the price increases.

This is a first hand announcement from Tibo, who leads Codex at OpenAI, as part of his numbered daily updates. He did not name the app, publish release notes or say which plans get the feature.

On the evening of October 18, Tibo posted Day 14 of his daily updates: a small quality of life improvement. You can now edit previous messages within a thread in the desktop app. He added that it was long requested and took way too long to get out.

He did not say which desktop app he meant, though his numbered updates have covered Codex and ChatGPT Work. Day 12 brought more effective usage in Codex and Work through context reuse and caching, and Day 13 was another reset of Codex usage limits.

The request is a long standing one. In April, a user opened an issue on the Codex GitHub repository asking to edit any message in the desktop app, not just the last one, saying forking a conversation to change an earlier message clutters the sidebar and that editing is supported in pretty much all coding agents. Other users on Windows and Mac agreed in the comments. OpenAI's public changelog had not listed the change by the time of his post.

Tibo did not explain what happens to the later part of a thread when an earlier message is edited, for example whether the agent's following turns are replaced or kept as a branch. The post drew more than 1,000 replies and passed 200,000 views.

Watch for release notes explaining how editing works, especially whether changes the agent already made to files are rolled back when an earlier message is edited.

Also watch whether the feature reaches the Codex command line tool and other surfaces, and what Tibo announces for the remaining days of his 28 day push. Many of the replies are likely to be further feature requests.

This is a first hand post from Julius, who says he builds T3 Chat and T3 Code. The figures are his own and unverified, and he did not say which agent he used or how the changes were reviewed.

Early on October 18, Julius, whose profile says he is building T3 Chat and T3 Code and contributes to the open source projects tRPC and TanStack, posted that for the past 30 hours his agent has merged almost 400 pull requests focused on web performance.

He listed three results: 86% fewer renders when another chat thread is streaming, up to 35% less JavaScript CPU use, and a heap about 10% smaller. He wrote that mobile is next and joked that it would be a shame if he had to freeze merges.

The joke refers to his own earlier post, which he quoted, saying he thought he would start a merge freeze from Friday until he has a Duo in his hands. He did not explain what the Duo is.

At that pace, the agent merged a pull request roughly every four or five minutes. Julius did not say whether the pull requests were reviewed by people, how the improvements were measured, or whether the agent ran alone or as several agents in parallel. Many small automated changes can each be easy to check, but together they can also make it harder to trace the cause of a later bug.

The post did not include benchmarks or links to the pull requests.

Watch whether Julius shares the measurements or the agent setup behind the 400 pull requests, and whether users notice the speed changes in T3 Chat.

Also watch the mobile work he says is next. Developers are increasingly letting agents run for long stretches on narrow goals such as performance, and how teams review that much automated code is an open question.

This is a secondhand claim from P Equity Research, which did not link the Morgan Stanley note. The note could not be found, and a Chinese language report on it gives different figures, so the details are unconfirmed.

On October 18, P Equity Research posted that Morgan Stanley has revised AMD's GPU demand forecast for 2026 to 2030 from 14.1 million units to 8.0 million units, a cut of 43.3%, while keeping Nvidia's the same. Its chart is dated October 14 and shows the cut as 5.9 million units.

The post did not link or quote the Morgan Stanley report, and no copy of it could be found. A post on the Chinese news site Sina dated October 15 attributed different figures to a Morgan Stanley report. It said the bank cut its forecast for AMD GPUs in 2026 by 75%, from 2.4 million to 0.6 million, cut 2027 from 2.7 million to 1.3 million, and put 2030 at 2.1 million against 3.2 million before. It said Nvidia's forecasts were unchanged, at 8.1 million units in 2026, 7.5 million in 2027 and 7.2 million over the long term.

That post framed the note as about risks to suppliers of chip packaging substrates from weaker server deployments, not as a call on AMD's stock. The two accounts do not clearly match, and neither can be checked against the original.

Other Morgan Stanley research reported this month was positive on demand for AMD's Venice server CPUs.

Watch for the Morgan Stanley note itself or coverage by major financial outlets, which would confirm the figures and explain the reasons for the cut.

Also watch AMD's next earnings report and any updates on its data center GPU orders, which will show whether demand matches the lower forecast. Claims about analyst notes often spread on X before the details are clear.

This is market commentary from an investor account. The chart credits Bloomberg, and Bloomberg reported in September that Nvidia's forward multiple was near its lowest in more than a decade, but the post does not give the current figure or the chart's date.

On October 18, the investor account NoLimit posted that Nvidia is at its cheapest valuation in a decade.

The attached chart, titled Nvidia Stock Valuation Hits Multiyear Low and sourced to Bloomberg, plots Nvidia's forward earnings ratio since 2016. It peaks near 70 times around late 2021 and above 60 times in 2023, then trends down to the high teens in late 2026, where a dashed line marks it as the lowest point on the chart.

Bloomberg reported on September 22 that Nvidia traded at less than 17 times forward earnings, near its lowest in more than a decade, about half its 2025 multiple and down from above 25 times in May. It said revenue and net income were expected to grow about 90% this fiscal year, and cited margin pressure from rising memory chip costs as one reason for the discount. An Investing.com analysis on October 6 put the multiple at 17.1 times as the stock hit a record high.

Shares fell in the following weeks. Other outlets cited forward multiples of about 24 to 25 times in mid October, so the figure depends on the earnings period used. A Bank of America analyst said on October 16 that Nvidia, Broadcom and Micron trade at about 13 times forward earnings on average, according to MarketDash.

A day earlier, the investor account Dray posted a chart contrasting Nvidia's revenue with its falling multiple.

Watch Nvidia's November 17 earnings report, which will update the forward estimates behind these multiples and show whether margins hold as Vera Rubin ramps.

Also watch whether the chip selloff deepens or reverses as rate expectations shift, and whether more analysts argue the stock is cheap or that earnings estimates are too high. Bloomberg's own updated figures would show whether the multiple has fallen further since September.

This is secondhand: Tom's Hardware is reporting on an Instagram post by the distributor Oderço. The company gave no arrival date or tracking details, and no authority has explained what happened.

On October 18, Tom's Hardware posted that an Nvidia RTX 5090 shipment worth $470,000 went missing at a Brazilian airport.

The article, by Jowi Morales, says the Brazilian electronics and IT distributor Oderço posted on Instagram after failing to get answers about the shipment. Oderço said it was expecting two pallets of Zotac RTX 5090 graphics cards at Guarulhos International Airport in São Paulo, that they supposedly arrived and then disappeared, and that it had received no official response to its inquiries. The company said it had never faced anything like it in its 38 year history, and put the value at about R$2,500,000, its acquisition cost.

Tom's Hardware did not know how many cards were missing. Based on a Brazilian retail listing of R$33,999.99 for the card, it estimated at least 63, and likely many more, since the stated value is the purchase cost rather than the retail price.

The article notes that RTX 5090s are in short supply worldwide and disappeared from online retailers in mid September. It says there have been rumors Nvidia stopped production to focus on more expensive professional AI cards, which Nvidia denies, and that Micro Center lists the card as not carried. It also cites earlier thefts of high value graphics hardware, including a crew that rammed security escorts to hijack an AI hardware shipment.

Watch whether Brazilian airport authorities or the cargo handler explain what happened to the pallets, and whether any of the cards turn up for resale.

Also watch RTX 5090 supply and prices, since the shortage that makes these cards targets is tied to Nvidia directing chips to AI products. Shippers of high value AI hardware may add security as thefts like this one spread.

This is a secondhand summary of Bloomberg Billionaires Index figures that Bloomberg published on October 5, so the post resurfaces data that is about two weeks old. The figures match Bloomberg's reporting.

On October 18, unusual_whales posted that tech billionaires are pulling away from the rest of the ultra rich. It said the world's roughly 100 tech billionaires added a record $845 billion through September, per Bloomberg, driven by the AI boom, for a combined fortune of $4.6 trillion, while every other billionaire combined was down $62 billion. It added that Elon Musk alone gained $310 billion, about 40% of the entire index's increase.

Bloomberg reported those figures on October 5. It said the roughly 100 technology fortunes among the world's 500 richest people gained $845 billion through September 30, the most ever for the first nine months of a year, and now account for 36% of the index's wealth despite being about a fifth of its members.

Quartz, summarizing Bloomberg, wrote that Americans accounted for 94% of the index's net gains and that since September 10 all 10 of the world's richest people have been from the US, a first since Bloomberg began tracking in 2012. It said Musk briefly became the world's first trillionaire after SpaceX and xAI merged and went public in June, and that Michael Dell's wealth rose 81% to $254 billion.

Bloomberg also reported signs the peak has passed. The combined wealth of the 500 richest peaked at $13.4 trillion in mid June and has since fallen 6% to $12.6 trillion, as investors grew skeptical of some leading AI names.

Watch the Bloomberg index for the fourth quarter, since the combined wealth of the richest has already fallen from its June peak as AI stocks cooled.

Also watch how the gap between tech and other fortunes feeds political debates over taxes and the gains from AI, including pledges such as Anthropic's new Public Benefit Fund. Large private valuations and stock moves will keep driving these totals.

This is a first hand account from Roemmele describing a bot he built and shared, and it has not been independently verified. The calls run through a third party service, ClawCall, which he says he has no connection to, rather than a built in Grok Bot phone feature.

On October 18, Brian Roemmele posted that he built a Grok Bot whose only job is to phone a restaurant and book a table, in a Grok voice, and that it is free to use.

He wrote that he told it the day before to book Gourmet Italia in Temecula for four people that night around 7. He said it found the number on its own, asked how flexible the group was on time, what name and callback number to give, read everything back and waited for his OK before dialing. After the call, he said, it summarizes what happened, quotes what the restaurant said and offers to add the booking to his calendar.

Roemmele wrote that Grok Bot needs its own phone number to make calls, so he connected a calling service called ClawCall, whose free tier gave him 30 calls or 30 minutes. He proposed that Grok offer its own phone service. He listed guardrails he built: it never gives out card or deposit details, never leaves voicemails unless asked, makes one call at a time and never calls on its own. He said the same approach could book haircuts, tee times or service appointments.

The attached image shows a figure with a round white head holding a rotary phone in a restaurant. That afternoon, Elon Musk quoted the post: Grok @Bot can call restaurants by phone and book you a table.

Watch whether Grok Bot adds its own phone numbers or calling feature, as Roemmele proposed, rather than relying on outside services.

Also watch how restaurants and other businesses respond to AI callers, including whether they are told they are speaking with a bot, and whether rules on automated calls apply. Other agent makers have their own voice calling features, so phone errands may become a common point of comparison.

These are first hand anecdotes from users and have not been verified. Elon Musk, whose company makes Grok Bot, quoted the shopping post, so the amplification is promotional.

On October 18, Teslaconomics posted that he sent one text to his Grok Bot asking it to order the Ollyball Keepy Uppy, an indoor ball from Shark Tank, and that it was now on his doorstep. He wrote that he did not open a single website and does not think he will ever go back to shopping the traditional way. His photos show a USPS mailer, the Keepy Uppy package and the inflated ball.

His post quoted his own October 11 post, in which he wrote that he gave Grok Bot his credit card details and it went to the official site, entered his address and payment, placed the order and sent the receipt to its own email.

That afternoon, Elon Musk quoted the post: Try Grok @Bot.

The same day, Katie Miller posted that she has her banking connected to Grok Bot, that connecting was seamless, and that it now audits her credit card, tracks refunds and returns and alerts her when bills are due. She said her favorite feature is when it tells her a trial period is turning into a full price subscription. Her post quoted an account saying Grok Bot will probably let users add credit cards in the future.

On October 17, Miller posted that Grok Bot got her a refund for an undelivered dress, as covered in that deck. Neither post explains how card or bank access is secured.

Watch whether Grok Bot adds an official way to store payment cards, which the post Miller quoted suggests may come, and how it secures bank and card access.

Also watch whether retailers welcome or block agent purchases, since The Verge has reported that several large sites restrict agents. Fraud protections and who is liable when an agent buys the wrong item are open questions as more people let agents spend money.

This is an opinion from Tesla's CEO, posted as a reply to an investor account. Optimus is Tesla's humanoid robot and the Cybercab is its planned robotaxi, and the post gives no figures or timelines.

Early on October 18, Elon Musk wrote that Optimus and Cybercab will be the vast majority of Tesla's value.

He added that the only things that will scale to incredible size economically, which he described as K2 and beyond, are superintelligence, robots, solar power and mass to space. He wrote that if you have the first two, the rest will follow, and ended by saying the Roadster is super cool, so Tesla is doing it anyway. The post does not define K2.

Musk was replying to Teslaconomics, who had written that people focused on the new Roadster's sales and margins were missing the point. That account, describing itself as a long term Tesla investor, wrote that it was looking at what Tesla and SpaceX can build together, and asked readers to imagine what could happen if the two companies merged.

Musk did not respond to the merger idea in his reply. A day earlier, he claimed the new Roadster will be sentient and able to perform its stunts autonomously on request, as covered in the Oct 17 deck.

The reply had drawn more than 20,000 likes. It includes no new production, pricing or delivery information for either product.

Watch Tesla's next earnings report and any updates on Optimus production and Cybercab deployment, which would show how close those products are to the scale Musk describes.

Also watch whether Musk or Tesla comment further on closer ties with SpaceX, which Teslaconomics raised, and how investors weigh the Roadster against the company's robotics pitch. Analysts' valuation models may shift if Tesla gives targets for robot or robotaxi revenue.

This is a secondhand and resurfaced clip posted by Veee, an AI and tech account. The backdrop in the video reads The Hill and Valley Forum 2025, so these remarks are from 2025 rather than this week.

On October 18, the account Veee posted a 70 second clip of Nvidia chief executive Jensen Huang, writing that the next wave of AI is physical AI and that it is coming for the world's largest industries.

In the quotes the post attributes to him, Huang says most AI today does not understand the laws of physics, friction, inertia, or cause and effect, which he calls physical reasoning abilities that children and even pets have. He says that when you put physical AI into a physical object, you get robotics. He adds that hopefully, in the next 10 years, a new generation of factories will be highly robotic, and that they will help deal with a severe labor shortage around the world.

The frames show Huang seated on stage at the Hill and Valley Forum, whose backdrop is dated 2025. The post does not say when the clip was recorded, so it is old footage recirculating rather than a new statement.

The same day, Mark Cuban argued that US manufacturing can come back if the country catches up on robotics, covered separately in this deck.

Watch whether Huang repeats this framing at Nvidia's upcoming events with concrete robotics products, partners or timelines, since the clip itself offers none. The ten year factory prediction is a hope rather than a forecast with numbers, so the useful signals are actual factory deployments of humanoid or industrial robots running on Nvidia hardware, and whether Nvidia breaks out robotics revenue in its results.

This is an opinion from Cuban, posted as a quote of Palmer Luckey. He gives no figures for the fees he mentions.

On October 18, Mark Cuban wrote that it is really simple: our robots and AI versus their robots and AI. He said that if the US can catch up on robotics and software and stay ahead on AI, there is zero reason manufacturing will not come back or start in the US.

He added that at Cost Plus Drugs, the only thing holding the company back from making generic drugs in Dallas more cheaply than it can buy them from Asia is the fees it has to pay. He did not say which fees he meant.

Cuban was quoting Palmer Luckey, the founder of the defense company Anduril, who wrote on October 17 that American made products are not impossible. Luckey said stories about iPhones costing $37,000 or laptops costing $100,000 are psyops, sometimes pushed by foreign adversaries but more commonly by American corporations that do not want the status quo upset.

Neither post cites data. Cuban's argument rests on the idea that automation narrows the labor cost gap with Asia, so that costs such as regulatory or other fees become the deciding factor. His view that the US needs to catch up on robotics implies he sees rivals ahead there, while he describes the US as ahead on AI.

The exchange comes as AI and robotics leaders, including Elon Musk with Tesla's Optimus robot, argue that robots will reshape where goods are made.

Watch whether Cuban spells out which fees hold back Cost Plus Drugs' manufacturing in Dallas, and whether the company expands domestic production of generics.

Also watch US policy on reshoring and robotics, including tariffs and incentives, and data on factory automation, which would test the claim that robots can make US manufacturing competitive. Rival views from economists on costs beyond labor, such as supply chains and materials, may follow.

This is secondhand: Phan is sharing screenshots from an October 14 Substack essay by Steve Huynh, who says he spent almost twenty years at Amazon and left in 2024. Huynh says his account comes from former colleagues and members of his community, so the anecdotes are unverified, and Amazon has not commented.

On October 18, Trung Phan posted that the usefulness of Amazon's six page memo has been nuked by AI generated text, with employees sitting in meetings suspecting text is AI and tuning out, and that suspicious memos get tagged AI;DR, for AI, didn't read.

His screenshots come from an essay Steve Huynh published on October 14. Huynh writes that in a recent Amazon review meeting, an L7 principal product manager presented a six page document, and that it was obvious he was reading it for the first time. He says he knows this because a member of his community was in the room.

Huynh describes Amazon's tradition of starting meetings with silent reading of a narrative memo, which Jeff Bezos argued forces clear thinking. He writes that the culture survived the pandemic but not AI, and that a senior engineer told him he tags documents AI;DR because if a model wrote it, a model can summarize it. He adds that performance reviews have become largely AI driven, with AI writing feedback and other AI summarizing it.

Huynh does not argue for banning AI. He proposes that AI be used freely for decisions that are easy to reverse, while people write and read documents for decisions that are hard to undo, and that authors own every word.

The Ken reported in 2025 that Amazon had rolled out internal AI writing tools and that some employees were using them to draft documents.

Watch whether Amazon responds or sets rules on AI written memos, as Huynh suggests, and whether other companies that copied the six page memo adopt similar policies.

Also watch for more accounts from current Amazon employees, since Huynh's are secondhand. The debate over AI written work and whether readers trust it is spreading to performance reviews, design documents and other writing that companies use to make decisions.

This is a reply by Sweeney, chief executive of Epic Games, to Kenney, who makes free game assets. Kenney's post did not name anyone, so it is not clear whether it was aimed at Sweeney.

On October 18, Kenney, whose profile says he creates free game assets, posted: Don't let billionaires convince you slop is the future. He wrote that they're used to managing and directing, not making things themselves, so of course they can't see the difference, and that they don't understand game makers' love for the craft. He told developers to keep making personal, beautiful games, and added: You've got my support.

A few hours later, Tim Sweeney replied: This is bullshit, Kenney. He wrote that he is the same tech tree following optimist who moved from C to C++, from 2D to 3D and from closed source to open source. I run a company. I write code. I contribute to published papers. I've upped my expectations of AI. He ended: Up yours!

The night before, Sweeney had posted that about 70% of his time writing Unreal Engine 1 went to routine typing and refactoring that AI can now handle, and that AI is also solving hard technical problems. Epic's official Unreal Engine plugin for Claude Code also spread widely on X that week.

Sweeney's reply passed 1.9 million views.

Watch whether Kenney or other independent developers respond, and whether the exchange affects how developers view Epic's AI tools for Unreal Engine.

Also watch the broader argument over AI in games, including store labels for AI made content and how studios describe their use of it. Public fights between executives and creators are likely to continue as AI tools spread.

This is a tips post from an account that describes itself as Mario Nawfal's crypto and AI account. The prompt has not been tested here, and the account gives no results from using it.

On October 18, the account RoundtableSpace posted a prompt for Claude Code, saying that before you launch anything you should have Opus 5.5 send five subagents in as your first users.

The prompt asks Claude to act as a launch QA lead and create five subagents, each playing a different first time user. One is impatient and quits at the first error. One is a skeptic who reads pricing, terms and privacy pages before signing up. One uses a 375 pixel screen on a slow 3G connection. One is a non native English speaker who follows only what the interface literally says. The last tries to break the app by pasting junk into forms, going back mid flow and double clicking everything.

Each subagent is told to open the app in a real browser, go from the landing page to a first success and take screenshots whenever it gets confused or stuck. The prompt then asks Claude to merge the problems into one list ranked by how many users hit them, fix them all in one pass and rerun the five users to show the fixes worked.

The post says it works best with a browser tool connected, naming Playwright or Chrome DevTools. It argues that your first 100 real users won't tell you why they left, but these five will. AI testers can miss problems that real people hit, and fixing everything in one pass can introduce new bugs, so results still need checking.

Watch whether developers report that agent testers like these catch real problems before launch, compared with standard automated tests and human testing.

Also watch for coding tools that build this kind of persona testing in directly. Agents that drive real browsers are becoming common, which makes simulated first users easier to set up, though their feedback still needs checking against real people.

This is promotion: Musk's company makes Grok Bot, and he quoted a post from the account Rothmus that calls using it a public service announcement. Neither post describes a specific task or new feature.

On October 18, Elon Musk posted: Grok @Bot is like hiring an incredibly competent employee. He was quoting a post from the account Rothmus earlier that afternoon that read: Use Grok @Bot. This is a public service announcement.

The Rothmus post includes an anime style illustration of a teal skinned character with pointed ears and an X emblem on its chest, standing on a ledge under a blue sky in front of a torn green flag with a large ring shaped emblem, flanked by small round characters in gray, blue, orange, black, green and brown. The post does not explain the image.

Musk's post passed 850,000 views by that evening. It was one of several Grok Bot posts he shared on October 18. Earlier that day he quoted Brian Roemmele's post about a Grok Bot that phones restaurants, writing that Grok @Bot can call restaurants by phone and book you a table, and quoted Teslaconomics' post about the bot ordering a toy, writing: Try Grok @Bot.

Musk did not say what tasks he had in mind, and the comparison to an employee is his own characterization. User accounts so far describe errands such as booking tables, placing orders, tracking refunds and negotiating bills, none of which have been independently verified.

Watch whether xAI follows the promotion with new Grok Bot features, such as built in phone calling or payment cards, which users have been asking for.

Also watch usage figures, if xAI shares them, and whether more businesses restrict agents as Grok Bot is pushed toward shopping and errands. Reports of mistakes or security problems would test the claim that it works like a competent employee.

This is a meme shared by Musk, not news, and he did not say who made the video. It reflects his own view that compute is scarce.

On October 18, Elon Musk posted a video captioned: What it's like trying to rent AI/SI compute.

The video, about 80 seconds long, follows the format of Monty Python's Cheese Shop sketch. A customer walks into an old fashioned wood paneled shop whose shelves are filled with server racks and asks the shopkeeper to rent some compute. He asks for eight H100s for a week, and the shopkeeper says they are fresh out. B200s have an 18 month waiting list, and A100s are all reserved by a very large chatbot. TPUs are internal use only, and an older GPU is available only in a region the customer has never heard of, with a three year commitment.

When the customer asks whether the shop has any compute at all, the shopkeeper says yes, then no. Asked why it is called a cloud, he answers: because it's always just over the horizon. The customer says he will build his own data center, and the shopkeeper tells him to mind the power bill. A musician playing a stringed instrument in the corner is asked to stop, as in the original sketch.

The joke lands in a week when Morgan Stanley estimated AI capex will reach about $1.6 trillion in 2027 and IDC said AI driven memory costs had surged about 350% in a year. Musk did not add any comment beyond the caption.

Watch whether cloud providers and neoclouds report easing or worsening GPU shortages in the coming earnings season, and how long waits for the newest Nvidia chips run.

Also watch whether more AI companies follow the sketch's punchline and build their own data centers rather than renting, despite rising power costs and local opposition. Musk's own plans for compute may also come up in upcoming announcements.

This is a joking first hand post from Tibo, who leads Codex at OpenAI. It does not announce a reset or any change to limits.

On October 18, Tibo posted: People have escaped the TL and are asking me for resets on the street now. I'll make little cards to hand out.

TL appears to refer to his timeline on X, where his posts about resets draw many replies. A reset restores users' Codex usage allowance before the normal period ends, so people who have run low can keep working. The joke suggests people are now approaching him in person, though he gave no details.

Tibo has been posting daily updates during what he calls a 28 day push. He announced a global reset of Codex usage limits on Day 10, as covered in the October 15 deck, and another on Day 13, posted on October 17, which he said would land by the end of the day Pacific time.

The post drew more than 600 replies by that evening. He did not say whether another reset is planned, or whether the cards are real.

Watch the remaining days of Tibo's 28 day push for further resets or lasting changes to Codex limits, which users have been asking for in large numbers.

Also watch whether OpenAI changes how Codex limits work for heavy users, since frequent manual resets suggest demand is outrunning the current allowances. Competing coding agents are under similar pressure over usage caps.