b2blistings.org Miniblog

The Orb and the Empty Till

Worldcoin arrived with the sort of grand, chrome-plated confidence that usually precedes a small fire. Sam Altman founded it, Tools for Humanity built it, and the pitch was almost comically vast: solve a central problem of the AI era by proving a person is, in fact, a person. Step up to a gleaming Orb, let it scan your iris, collect some cryptocurrency, and join a global identity system meant to separate human beings from ever more convincing machines.

The trouble was there from the beginning, blinking in neon. Worldcoin asked ordinary people to surrender biometric data of the most intimate sort for speculative crypto tokens, as if this were a charming bit of admin rather than a profound question of privacy. Regulators in Europe, Africa, and other regions reacted accordingly, opening investigations and placing limits over data-protection concerns.

Now the commercial reality looks just as awkward. Reports say Tools for Humanity is cutting jobs after failing to produce meaningful revenue. At the same time, OpenAI is charging toward a possible IPO valued above $100 billion.

That gap tells its own story. OpenAI found demand people already had and monetised it. Worldcoin built around a future problem, wrapped it in crypto theory, and hoped the need would catch up. It may yet pivot or endure, but the lesson is plain: ambition is not product-market fit, and scanning eyeballs is not a business model.

Posted on 9 June 2026

From the Reddit Archives: the Tokenpocalypse Arrives

Microsoft’s GitHub Copilot pricing shift may be the first real hangover after the AI party: less all-you-can-eat buffet, more every-fry-counts accounting. A Reddit user said their company now calls it the Tokenpocalypse, which feels about right if your budget just got mugged by autocomplete.

The core issue is simple and ugly. A lot of AI software looked cheap because investors were quietly picking up the tab. Now some of that bill is moving to customers, starting with Copilot’s move away from a flat fee toward token-based charges. That change matters beyond coding tools. As companies like Anthropic prepare for IPOs, they’ll face blunt questions about whether revenue can ever catch up with the true cost of running these models.

Recent examples suggest customers are already hitting the wall. Uber reportedly burned through its AI budget faster than expected, then responded by capping and limiting internal use. That’s a warning sign: demand is strong, but appetite for spending is not infinite.

Even the original $20-a-month ChatGPT Plus price now looks less like a master plan and more like someone blurting out a number and hoping math would become optional.

The speed is part of the chaos. Tokenmaxxxing became a craze and then a cautionary tale within six months. Meanwhile, the government this week created a narrow review path for powerful AI models. The risks are evolving in real time, faster than the paperwork designed to describe them.

Posted on 8 June 2026

B2B Winners Make Personalization Uncomfortably Precise

Personalization in B2B marketing is now so commonplace that the interesting question is no longer whether firms do it, but how far they can be bothered to go. McKinsey’s 2026 Global B2B Pulse Survey, drawing on decision-makers in 13 countries, suggests the firms pulling ahead are the ones practising it with almost indecent specificity.

Among companies that lifted market share by more than 10% year over year, half said their email work was very personalized, while 15% had reached one-to-one email personalization. On social media, 45% described their approach as very personalized and 18% as one-to-one.

The laggards looked distinctly less intimate. Of the companies whose market share fell by more than 5% year over year, just 18% reported very personalized email campaigns, and 7% managed one-to-one email. Social figures were weaker still: 15% said their strategy was very personalized, and 6% one-to-one.

The implication is plain enough. Buyers respond better when messages arrive with decent timing, some relevance, and a passing resemblance to their actual needs. Companies gaining ground are not merely tidying up one channel; they are coordinating several, which hints at competent teams and a customer strategy that survives contact with reality.

The catch is that hyperpersonalization demands sturdy customer data and workflows that do not collapse under their own ambition. AI may help with the drudgery: producing more content, scaling campaigns, and using zero- and first-party data without exhausting resources.

Posted on 7 June 2026

The Token Feast Meets the Reckoning

AI’s glamour phase is ending in the fluorescent back office, where the bill has begun to snarl. Uber exhausted its 2026 AI coding budget by April. Microsoft pulled developers’ Claude Code access only months after granting it. At Priceline, a routine Cursor renewal arrived priced roughly four to five times higher.

Cheaper tokens did not mean cheaper AI. Better models released in November — Anthropic’s Claude Opus 4.5, OpenAI’s GPT-5.1, and Google’s Gemini 3 Pro — made agentic software far more capable, and far hungrier. One company reportedly ran up a $500 million Claude tab after neglecting usage caps. By spring, some firms were already three times over their annual token budgets.

The problem is no longer whether the tools work, but whether anyone can see, audit, or control what they cost. Tracking token spend operates at trillions of rows per month, far beyond spreadsheet economics. Priceline is already finding mismatches between vendor invoices and internal records.

The productivity math is muddy. Faros AI’s two-year study of 20,000 developers found output rising alongside bugs and rewrites. Jellyfish found top AI users were about twice as productive, but consumed ten times more tokens; per-developer usage jumped about 18.6x in nine months.

Now a tooling market is forming — Pay-i, Paid, Jellyfish, Waydev, Faros AI, Ramp, Datadog, New Relic, and soon AWS. The Linux Foundation’s new Tokenomics Foundation aims to standardize token billing, efficiency, and metrics such as cost-per-intelligence and tokens-per-watt, but its first concrete output is still months away.

Posted on 6 June 2026

Filtr Turns Apple Privacy Into Something You Can Actually Feel

Apple has spent years making privacy feel like a design preference. Filtr is what happens when that preference becomes infrastructural.

Built by Kaylee Serena Calderolla, the developer behind Wipr, Filtr uses a new URL filters capability in iOS 26 and macOS 26 to block ads beyond Safari, across iPhones, iPads, and Macs, at the network level. That matters because ordinary ad blockers mostly sanitize the web browser; apps have remained a separate, noisier reality.

Wipr already stops ads and tracking code in Safari. Filtr, sold as an added feature inside Wipr, extends that logic into apps, including non-Safari browsers. Wipr costs $5 on the App Store for all Apple devices. Filtr adds $5 per year or $25 for lifetime access.

Calderolla has described the implementation as difficult, partly because Apple’s documentation was thin. Filtr appears to be the first app using the feature. It relies on a maintained blocklist: a local pre-filter checks likely matches on-device, then confirms them against Calderolla’s servers. Those lookups pass through Apple’s proxy, obscuring the user’s identity. Calderolla’s privacy policy says her apps do not collect personal data.

The result is mostly invisible, which is the ideal state for privacy software. Ads disappear in many apps, sometimes leaving gray placeholders. It is not universal: Facebook, Google, Reddit, and other services that serve ads from their own domains can still slip through, because blocking the entire domain could break the app.

Posted on 5 June 2026

Microsoft’s Scout Turns OpenClaw Chaos Into a Corporate Assistant

Microsoft is rolling out Scout, a new AI assistant for Microsoft 365 that takes a very OpenClaw kind of energy and tries to put it in a blazer.

OpenClaw tore through AI circles in early 2026, basically a feral genius of an agent framework. Its momentum cooled after OpenAI hired away its founder, but the idea clearly left fingerprints all over Microsoft’s latest move.

Scout, built on the OpenClaw framework, is an always-on assistant that keeps a persistent identity, style, and memory of how you work. Users give each instance a name; in one demo, it was Sebastian. The pitch is long-term adaptation: people feed it feedback, routines, and preferences, and over time it stores those patterns as durable memories and skills.

Access comes through Microsoft’s Frontier program for experimental products, and it requires a GitHub Copilot subscription.

The assistant runs from the cloud but reaches across desktop and browser environments, letting it connect with inboxes, calendars, and related tools. Microsoft will ship it with starter abilities like calendar management and meeting-agenda drafting, but the real strategy is user-built customization. That matters because once an assistant learns your weird little work habits, leaving it gets painful.

Microsoft is also wrapping Scout in stronger controls after OpenClaw exposed how badly autonomous agents can misbehave, including a reported inbox incident earlier this year. Scout includes a policy conformance system with continuous checks and audit trails.

Scout arrived alongside Project Solara, a Copilot update, and a new reasoning model at Build. Posted on 4 June 2026

AI Can Build the Stage, but B2B Websites Still Need a Human Plot

AI and no-code website builders now whip up B2B sites at the speed of a glitter cannon. Platforms including Wix Harmony, HubSpot’s AI features and Relume can sketch sitemaps, layouts and draft copy in a matter of hours, shrinking a process that once lumbered on for weeks.

But speed is only the painted backdrop. The hard bit is still knowing who the site is for and how they buy.

GoingClear founder Paul J. Scott argues that AI is strong at producing a first pass, yet weak where B2B websites actually win or lose: understanding buyers, long sales cycles and the messy little rituals behind business decisions. A site that converts has to escort particular prospects through particular stages of a purchase journey. That requires context AI does not possess.

The real dividing line is strategy. AI can assemble a sitemap from a company description, but it cannot shape that structure around buyer intent, research findings or genuine customer anxieties. It can also generate familiar page layouts, though high-performing B2B sites need conversion-led experiences tuned to behaviour, not templates.

Branding suffers too. As more companies lean on machine-written copy, many sites blur into the same beige hum. When buyers are weighing major investments, they want authority, credibility and a distinct perspective.

As launching gets easier, advantage moves to whoever understands the customer best. Posted on 3 June 2026

SpaceX’s IPO Filing Admits the Real Cost of AI: Water

The altered filing reveals something almost indecently material: for SpaceX, water has joined electricity and chips as a hard limit on growth. In the amended IPO paperwork submitted Monday, the company — now folded together with Elon Musk’s xAI — warns that cooling large data centers may demand substantial water, making supply a decisive factor in where facilities can be built and how they operate.

Earlier, the company had cast the bottleneck mainly as access to power at viable prices, plus long construction schedules and shortages of materials. Now the constraint is stated more starkly: economically viable access to both power and water. Scarcity, drought, local competition for supply, and restrictions on use could, the filing says, curb cooling capacity, raise costs, delay expansion, or force a switch to alternative cooling systems that may be pricier or harder to obtain.

That small insertion points to a larger unease around AI infrastructure: the server farm is no longer merely an electrical appetite but a hydraulic one, too, arriving as climate change worsens localized drought conditions and scrutiny of data-center water consumption intensifies.

Why the warning appeared only now is unclear. SpaceX remains in the pre-IPO phase, with the SEC sending comment letters seeking more detail; those letters will be public only after the offering.

The amended filing also says up to 5% of IPO shares may go to employees and friends of executives, and warns that significant future share issuance could dilute investors — a notable signal given merger speculation involving Tesla.

Posted on 2 June 2026

Meta Wants a Second Business. History Suggests That Is the Hard Part

Meta is trying, again, to become more than a machine that turns human attention into advertising revenue. This week it began testing two paid tiers for Meta AI, a ChatGPT-style app and website, first in Singapore, Guatemala, and Bolivia, priced at $7.99 and $19.99 a month. It also formally rolled out premium subscriptions for Instagram, Facebook, and WhatsApp, plus higher-end verification for businesses. Mark Zuckerberg even said a cloud business remains possible if Meta ends up with extra AI computing capacity.

The need is obvious. In the latest quarter, about 98% of Meta’s $56.3 billion in revenue came from ads. That business is booming again, with the company’s fastest quarterly growth since 2021. But AI threatens to change where people seek information, and fewer screen-bound users would mean fewer ad impressions.

Meta’s record outside ads is mostly a museum of expensive hopes: Portal arrived in 2018 and disappeared four years later; Oculus, bought for $2 billion in 2014, has not produced a breakout headset, while Reality Labs has lost more than $80 billion since late 2020. Libra, launched in 2019, collapsed under scrutiny and was fully shut down in 2022. Workplace, introduced in 2016, is being closed after a 2024 announcement.

Wolfe Research sees AI subscriptions reaching $3 billion in 2027 and $16 billion by 2030. Enterprise cloud would be harder: Meta is consumer-first, cutting staff, and plans to spend $125 billion to $145 billion on AI infrastructure in 2026. Posted on 1 June 2026

Google’s AI Search Push Is Turning B2B Discovery Into a Trust Contest

Google is doing what Google always does when it smells a new empire: moving the furniture so everyone else has to relearn how to walk.

At its developer conference in May 2026, the company unveiled AI Mode, the biggest overhaul of Search in more than 25 years. For B2B companies, this is not a cute interface refresh. It changes where online visibility actually happens.

The old web economy mostly relied on people clicking around. Search engines surfaced links, websites competed for rankings, and buyers did their own comparison shopping across multiple tabs like caffeinated detectives. AI search collapses that wandering process. Google now aims to understand intent, evaluate sources, assemble an answer and deliver a recommendation inside the product itself.

That means discovery is shifting away from the familiar search-results page and toward an AI recommendation layer sitting above it. In practical terms, Google is climbing further up the commercial decision stack, influencing not just what users find but how they decide.

For B2B marketers, this raises the stakes on trust signals. Structured information, broad citation, brand authority, public relations strength and proprietary insight matter more when an AI system is choosing what deserves inclusion. Generic SEO filler, meanwhile, becomes less useful if the platform answers the question before anyone clicks.

The new advantage is not merely being indexed. It is being credible enough to be synthesized.

Posted on 31 May 2026

Royal Mail’s Delayed Signals

Royal Mail’s numbers have the feel of a machine reporting its own partial failure. In the year to the end of March, only 75.7% of first class letters arrived on time, nowhere near the old 93% benchmark and worse than last year’s 76.9%. Second class also slipped, with 90.2% delivered within three working days, down from 92.2% and short of the former 98.5% standard.

The figures cover the period under new private owner Daniel Kretinsky’s EP Group, whose takeover won shareholder approval at the end of April last year. Ofcom is highly concerned and is expected to begin an investigation early next week. Last October, the regulator fined Royal Mail £21m for missing delivery targets, its third-largest penalty; the company had also been fined in 2023 and 2024.

This is not a brief malfunction. Royal Mail has not met second class targets in six years, or first class targets in ten. Service deteriorated during Covid-19 and never fully returned to its old pattern.

In February, postal workers reported letters left undelivered for weeks while parcels were allegedly given priority because they brought in more money. Executives later faced questions on this in March.

Royal Mail says service is improving, with £500m to be invested over five years. It aims to meet Ofcom’s lower new standards by next year: 90% for first class and 95% for second class, alongside longer hours for some part-time staff and no Saturday second class deliveries. Posted on 30 May 2026

Wall Street Is Coming for AI Tokens

AI is developing the kind of market infrastructure usually reserved for things like crude, gold, or wheat: futures. And honestly, that feels inevitable. When an input becomes central enough to business, finance eventually shows up in a blazer.

China’s Shanghai Futures Exchange is designing a derivatives market tied to AI tokens, while CME Group and Intercontinental Exchange, which owns the NYSE, are separately working on futures contracts for GPU rentals.

The GPU side is already easier to price. Across 28 marketplaces and cloud providers tracked by AI Mining Co., median Nvidia H100 rental rates ran from $1.40 to $4.27 per hour across 13 marketplaces. H200s averaged $2.34 to $5 per hour across 10 marketplaces. In just the last seven days, average H100 pricing moved between $2.79 and $3.33.

Tokens are murkier, even though they increasingly function as AI’s real unit of commerce. Major enterprise plans are priced that way: OpenAI lists GPT-5.5 API usage at $5 per million input tokens and $30 per million output tokens. Amazon Bedrock also lets providers charge by the token.

That matters because AI infrastructure spending is exploding. Cloud companies, private equity firms, and data-center operators have already committed hundreds of billions on the assumption that demand for GPUs and compute keeps climbing. Neoclouds are chasing that demand too, some focused on inference, others taking on AWS, Google Cloud, and Oracle.

A token derivative would connect financial hedging directly to how AI services are sold. Posted on 29 May 2026

XLM Rallies on DTCC-Stellar Tokenization Plan as Traders Watch $0.183 Test

Stellar just got the kind of headline crypto traders actually respect: one tied to real market plumbing, not just another buzzword parade.

XLM jumped nearly 10% to about $0.163 after DTCC and the Stellar Development Foundation revealed plans to bring tokenization of DTC-custodied assets onto Stellar. That matters because DTCC sits at the center of traditional market infrastructure. This isn’t theoretical blockchain cosplay. It connects tokenization to systems finance already uses every day.

The initiative is designed to support turning traditional assets into tokenized versions while also managing corporate actions and reporting across the asset lifecycle. Translation: tokenized assets move a little closer to functioning inside the same operational machinery as conventional securities.

Price action noticed immediately. XLM pushed above its 20-day and 50-day exponential moving averages, a notable shift after spending most of 2026 stuck in a broad range and failing to hold prior rebounds.

Now comes the real exam. Traders are watching the $0.183 area, which has repeatedly capped recovery attempts. If buyers can force a clean break there, the next upside zone sits around $0.255 in June.

Until then, the setup is classic crypto suspense: this could be the beginning of a genuine trend change, or just another flashy spike that fades as quickly as it arrived.

Posted on 28 May 2026

Ferrari’s Luce Debut Jolts the Market as Style Row Eclipses the Numbers

Ferrari’s first electric four-door, the 1,035hp Luce, has arrived with all the ceremony of a royal unveiling and rather the after-effects of a family row. Within a day of its debut on 26 May, Ferrari shares fell about 6%, sliding from Monday’s close of €309 to around €290, after briefly dipping to €289. That erased a little over £3bn in value.

The market reaction was striking because the Luce is, by the numbers, a marvel: 800-volt architecture, a 122kWh battery, four in-house motors from Maranello, 1,036hp, 0-62mph in 2.5 seconds, active suspension, carbon-ceramic brakes and 500kW charging. Price: nearly £475,000.

The upset appears to be less about capability than character. Online reaction has been sharply split, much of it aimed at the styling, which many see as distinctly unlike a Ferrari. The design involved Lovefrom, the agency led by Jony Ive and Mark Newson, and the resemblance to polished Apple-like consumer design has not gone unnoticed.

Ferrari has suffered a steeper share drop before: 15% in October 2025, after lowering its 2030 growth outlook and trimming EV plans. Elsewhere, Jaguar’s Type 00 stirred similar uproar, though Tata Motors absorbed the financial tremors.

Luca Cordero di Montezemolo, Ferrari president from 1991 to 2010 and later chairman through Fiat ownership, made plain his dismay, warning the Luce risks damaging the Ferrari myth and suggesting the prancing horse scarcely belongs on it.

Posted on 27 May 2026

Survival Shift

Britain’s side-hustle boom looks less like ambition than endurance. About 1.3 million workers now hold a second job in the UK, down slightly from the 1.35 million peak recorded in 2025, as high costs, shakier employment and a swelling gig economy force people to patch incomes together.

In Bristol, the country’s second most expensive city, 29-year-old Billy-Jo Pierce runs a tooth-gem business, works reception, does bar and festival shifts, and sells clothes online. Originally from Birmingham, she studied interior design, graduated with a first, then took a gaming office job while building her business at night. Redundancy pushed her into self-employment full-time last year, but soaring living costs and beauty-sector materials rising by more than 90% over the past decade left one income too thin. She now lives in a van and regularly works 50 to 60 hours a week.

The backdrop is bleak: unemployment has climbed to 5%, vacancies have fallen to a five-year low, and just under five million people do gig work, though only a fifth depend on it as their main income. Dr Emily Beaumont at the University of Gloucestershire says second jobs have become normal because one wage often no longer covers life, reflecting a more fragmented economy with more risk pushed onto workers.

Bristol designer Engy Elboreini, 35, says AI and Canva have gutted demand for traditional design, pushing her into creative production, coordination and retraining in events. Hollie, 41, a single mother, combines part-time legal work, occasional TV extra jobs and life modelling around her son Max’s school day. The common thread is not freedom. It is adaptation under pressure.

Posted on 26 May 2026

Xreal Bets the Smart-Glasses Curse Is Finally Breaking

Silicon Valley’s been chasing face computers for years, and mostly getting its pockets turned out. The pitch never changed: free people from hunching over phones by moving computing into a lightweight pair of specs. The reality was clunky hardware, weird public optics, thin software, and oceans of red ink.

At Google I/O in Mountain View last week, Xreal CEO Chi Xu argued the trade may finally be crawling toward viability. He pointed to the same shift others do: Meta’s 2023 tie-up with Ray-Ban proved smart glasses can move real units, even if Reality Labs still bleeds cash. Xu’s own verdict on the sector was blunt: nearly everybody is losing money because the job is genuinely difficult.

Xreal, a longtime Google partner, thinks the timing has changed because the pieces are lining up at once: hardware, operating system, and interface. Its latest swing is Project Aura, built around wired XR glasses with embedded OLED displays. They plug into a phone-like puck that does the heavy lifting and sits in your pocket.

That compromise buys a broader set of uses: immersive Google Maps, VR YouTube, hand-tracked games, web browsing, and a painting app for private holographic sketches. Xreal also sees office use, from recipe overlays to a portable workspace in cafés or on flights.

Aura is developer-only for now, with a commercial launch planned later this year. Xreal is also preparing an IPO before the end of 2026 and says it could break even next year as margins rise and marketing and sales costs fall.

Posted on 25 May 2026

Standard Chartered’s AI Push Turns a Management Cliché Into a Human Problem

Standard Chartered chief executive Bill Winters has apologised after referring, at an investors conference, to workers most exposed to AI-led replacement as lower-value human capital—a phrase that managed to sound both clinical and personal, the vocabulary of a spreadsheet discovering it has a mouth.

The bank, headquartered in the UK and employing about 82,000 people globally, mostly in back-office jobs, has said automation is likely to eliminate roughly 15% of those roles over the next four years, or about 7,800 positions. Winters argued the shift was not chiefly about cutting costs, but about substituting some lower-value roles with financial and investment capital directed into automation.

After backlash, he posted on LinkedIn that he was sorry for wording that had upset colleagues and said the point he meant to make was that roles of lower economic value are more susceptible to automation, while the bank has a responsibility to help affected staff move into higher-value work. In a second post, he published a transcript of his conference comments and said Standard Chartered remained committed to helping employees adapt to rapid industry change.

Some respondents said the fuller context did little to alter the meaning. Internally, Winters told staff that headlines had stripped away nuance, and said the bank would prioritise redeployment and handle changes thoughtfully. A spokesperson said the strategy was to combine human talent with AI while building future-ready skills inside the bank, and outside it where needed.

Posted on 24 May 2026

Britain’s Young Jobless Problem Looks Less Like Apathy and More Like a Botched System

A strange old state of affairs: nearly a million young people in Britain are outside education, work or training, yet major employers still say they cannot get the people with the right skills. Amazon’s UK chief, John Boumphrey, says the habit of treating this as laziness or feeble character is nonsense; the fault lies more with how school and training feed into actual jobs.

The numbers are grim. Unemployment across the UK rose to 5% in the three months to March, up from 4.9%, and for 16 to 24-year-olds it hit 16.2%, the worst since late 2014. New analysis from the Institute for Fiscal Studies says the drop in youth employment is nearing the scale seen during the 2008 crash and the pandemic. Hospitality and graduate openings have both thinned out.

Amazon employs 75,000 people in the UK across 100 sites, including 30 warehouses, and says half arrived from education or unemployment. But it still lacks technicians and mechatronics engineers to maintain warehouse robots. Boumphrey wants mandatory work experience after 16, arguing it teaches teamwork, communication and problem-solving that employers actually need. Posted on 23 May 2026

Young Men, Easy Money, and the New Parlour of Speculation

Prediction markets have become a peculiarly modern male amusement: part exchange, part casino, part internet pageant. Their clientele are overwhelmingly men and mostly under 45; polling suggests more than a quarter of American men aged 18-24 used a prediction market or gambling app in the previous six months, against 14% of the public.

Cameron George, 26, from Utah, once stacked shelves at Walmart and now lives as a crypto trader and online personality, complete with lime green McLaren, wife and five children. He has long promoted trading content and follows prediction markets largely for crypto pricing and news. Recently he let an AI bot place bets for him and is down a couple of thousand dollars.

The business is enormous. Kalshi is valued at $22bn, Polymarket at $9bn. Fans treat these sites as sharper than bookmakers and even more revealing than polls. Critics note that they are styled like finance platforms, softening the uglier fact that most users lose while a tiny elite prospers. Since early 2025, Bloomberg found nearly twice as many Polymarket accounts staking over $1,000 lost as won; the Wall Street Journal reported 67% of profits went to 0.1% of accounts.

The atmosphere is aggressively masculine: sports, crypto, memes, influencers, the cult of being cleverer than the next man. Meanwhile, suspicions of insider trading have followed wagers on Iran and Venezuela. Even so, users such as George, half-disgusted and half-enchanted, keep playing. Posted on 22 May 2026