Paul Graham wrote in 2005 that “what hackers are doing now, everyone will be doing in ten years.” This week, Ben Thompson revisited the line to describe what hackers are doing now: running always-on agents on a Mac Mini in the corner.
All the big tech stories this week have converged on consumers using always-on agents. Meta’s Muse keeps working after you close the app. OpenAI’s dots run on their own cloud computers. And Instinct, valued at $10B last month while still invite-only, works inside WhatsApp and iMessage. Graham gave the mainstream ten years to catch up with the hackers. With agents, it’s unlikely to take more than two.
Personal agents like Instinct will be, for most people, the first time AI feels like magic. You text it as you would a friend, and the refund gets chased, the plumber rebooked. It’s fast, and its replies are polished. It is delightfully designed.
While many of us are nervous about letting these agents run free, the experience is so good that we give in within days. The trade is simple: convenience now, in exchange for something far more intimate than the behavioural data we have handed to technology companies before.
Google knows what we search for. Instagram knows what holds our attention. Amazon knows what we buy. A personal agent can know all of that, but also the documents on our computers, the contents of our inboxes, our calendars, finances, relationships, travel plans and the private context behind our decisions. It doesn’t just see isolated signals about us; over time, it can assemble them into something approaching a digital twin: a continuously updated model of our lives and how we tend to make decisions.
And that changes the economics of the relationship. Social media made money from our attention. An always-on agent makes money from our intentions and actions.
A small number of technical users have found a way to have both privacy and convenience. They keep their context in files on their own machines – notes, preferences, a running record of how they like things done - and point whichever agent they prefer at those files. If a better agent comes along next year, or next week, they switch to that agent without losing their accumulated context. But this takes effort to maintain - effort that most people won’t make - and the cost of not making it builds up silently: years of history, wants and habits held on one company’s servers, perhaps with no practical way to take it elsewhere.
What users need is a portable vault: a record of their history, preferences, relationships and context that they own and lend to whichever agent they choose.
In theory, regulation has already been moving in this direction. GDPR includes a right to data portability, and the DMA requires the largest platforms to give users continuous, real-time access to their data. But neither reaches companies like Instinct. Instinct’s terms of service grant it a broad licence to your data, and it can keep using your emails even after you disconnect your account.
What we are missing is the company that makes owning your data as easy as Instinct makes handing it over. The better outcome is that portability becomes a product before it becomes a regulatory battle: a place where your memory lives independently of any one agent, so you can switch models without starting over. If someone makes that simpler and more useful than surrendering everything to a single company, users may choose openness for the same reason they choose any great product - because it works better.
If nobody does, we may not notice what we have given up until it’s too late. Using an agent will be a little like building a house on rented land. At first there isn’t much to leave behind. But every interaction adds another room, another furnishing, another thing worth keeping, and the cost of moving rises almost invisibly. Years later, the reason we stay may have less to do with which agent is best than with the fact that the company holding our history has become too costly to leave.
IPOs / Publics
Firmus (data centres) priced $5B IPO at A$30.6B - triple its $10.5B valuation from August - drawing scepticism from Australian investors over unproven track record, $30B debt pile (6x forecast earnings), & reliance on five planned facilities across Asia-Pacific to hit $5B annual earnings within 5 yr (here)
Airtel Money (mobile money arm of Airtel Africa) raised $703M in London’s largest IPO since Sep 2025, pricing at $7B - below initial expectations but offering a boost to a market that has shrunk over the past decade as companies sought higher valuations elsewhere (here)
Anthropic targeting mid-November IPO roadshow & pre-Thanksgiving debut at $1.8-2T, despite $42B net loss in 2025 (largely fair-value liabilities; revenue $4.6B, up from $386M) (here)
Blockchain.com is targeting a $500M US IPO at $4-6B, a 57-71% discount to its 2022 peak of $14B, & has confidentially filed with the SEC. The wallet & brokerage platform recently announced plans with the NYSE to offer tokenised US stocks, though public financials remain undisclosed (here)
Amazon is seeking to offload ~$8B of Nvidia Grace Blackwell chips into a special-purpose vehicle backed by outside investors, then lease the semiconductors back - a move to keep AI capex off its balance sheet as spending hits $220B this year. The deal follows weaker appetite for Amazon’s long-dated debt in July & mirrors CoreWeave’s GPU-backed financing model (here)
Amazon plans to equip 5,000 delivery drivers with smart glasses by year-end (20,000 by end-2027) that capture static images every few seconds to optimise last-mile logistics via its Wellspring AI platform. The system will photograph private property with no customer opt-out, raising privacy concerns reminiscent of earlier Alexa & Ring controversies (here)
Google released Gemini 4 Argon, claiming state-of-the-art performance in coding, cyber security & complex reasoning with a 16x larger context window - the company skipped 3.5 Pro after disappointing results & is pricing Argon at $2-10 per M tokens (rising to $4-20), undercutting Anthropic’s Fable & OpenAI’s Astra as it attempts to recapture enterprise AI market share (here)
Big Dogs
OpenAI is seeking at least $30B in new funding at $1.4T to delay its IPO, after revenue run rate grew 70% since July to over $40B. Altman cited the need to navigate heightened AI safety concerns without public-market pressure; the company scrapped its GPT-6.1 Astra release after it failed internal safety standards (here)
OpenAI launched dots, always-on AI agents powered by GPT-6 Astra that run on dedicated cloud computers, connect to 4,000+ apps via plugins, & learn user preferences over time - rolling out to Pro, Business Premium & Enterprise tiers with specialist dots for enterprise access management & IT provisioning on the roadmap (here)
OpenAI CEO Sam Altman told Politico that “the world should accept some bad things happening“ from AI in exchange for its benefits, backing lighter-touch regulation & rejecting catastrophic risk scenarios including loss of control to AI (here)
Wall Street banks began syndicating a $60B debt package to fund Anthropic’s lease of Google tensor processing units from Broadcom, the largest chip-financing deal to date. $42B of senior debt is guaranteed by Broadcom to lower borrowing costs; $18B of junior debt (without guarantees) will launch later, with Blackstone committed to ~$9B. The deal tests investor appetite for AI debt as concerns grow over mounting capital outlays & profitability timelines (here)
Mistral released Large 4, its latest flagship model, marking the Paris-based AI lab’s continued push to compete with US hyperscalers in foundation models - a test of whether European venture-backed AI can sustain the capital & talent intensity required at frontier scale (here)
Reflection AI (US, Nvidia-backed) launched Beam, an open-weight model it claims matches China’s leading GLM-5.2, after raising $4B+ at $25B since 2024. The Pentagon & Energy Dept are partners; the model aims to undercut OpenAI & Anthropic on cost while addressing Washington’s concern that Chinese open-weight models now power 56% of tokens processed through Vercel (up from 7% in Dec) (here)
Venture Capital
Bill Draper, has died aged 98. He co-founded Sutter Hill Ventures in the 1960s & helped establish Silicon Valley’s venture capital industry before it had a name (here)
Regulation
UK AI minister Kanishka Narayan said he is reviewing non-compete clauses & extended notice periods “extremely rapidly” after 20+ startup founders - including ElevenLabs, Ineffable Intelligence & Fractile - called for reform, arguing current rules can delay new hires by up to 12 mo & stall growth (here)
FTC Chairman Andrew Ferguson launched a probe into Anthropic, OpenAI & other frontier labs, planning civil investigative demands (CID) to compel executive testimony on consumer risks & potential unfair practices under the FTC Act (here)
Norway’s government proposed a temporary ban on smart glasses in parks, beaches, museums, schools, healthcare facilities & gyms, citing privacy concerns over covert recording (here)
Venture Geopolitics
UK government spending with Palantir has reached £992M since 2014 (48% above prior estimates), with MoD accounting for £490M & NHS £401M if the Federated Data Platform runs to term. Prime Minister Andy Burnham faces growing pressure to cut ties; Commons science & health committees both want the NHS contract ended at its Feb 2027 break clause, citing dependency risk & ethical concerns. FT reports government is now seeking British alternatives for both NHS & MoD work (here)
Jamie Dimon (JP Morgan CEO) called for a renewed US-Europe alliance anchored in trade, arguing bold reforms & restored military capability could secure Western geopolitical strength through 2275, as fragmentation & capability decay threaten relative decline (here)
Alan Turing Institute director George Williamson urged UK public services to deploy cheaper, purpose-built AI models rather than default to “extraordinarily expensive” frontier systems, citing research showing near-frontier performance at lower computational & financial cost. He called for broader public debate on AI safety & deployment, noting rising concerns post-OpenAI’s Australian government site intrusion, & positioned the institute to focus on national resilience, misinformation mitigation, & verification standards for AI behaving unpredictably outside lab settings (here)
German Intelligence Chief (1998-2005) & ex-interior ministry permanent secretary, was arrested on treason charges after prosecutors alleged he illegally acquired ~2,000 classified documents over 16 years & shared some material with foreign intelligence officials. The scale of the breach, potentially Germany’s largest since the Cold War (here)
Britain has lost $160B in billionaire wealth over the past 2 years, according to the Bloomberg Billionaires Index - surpassing the total wealth of billionaires who remain. Labour’s abolition of non-dom tax perks in April 2025 triggered mass exits to Monaco, Switzerland & the UAE, with the top 1% of earners accounting for 27% of UK income tax. Recent departures include hedge fund manager Chris Rokos relocating to Greece, raising questions about whether homegrown UK talent will follow foreign-born billionaires (here)
Several European payments groups have joined forces to challenge US dominance. Bancomat, Bizum, EPI Company/Wero, SIBS-MB WAY & Vipps MobilePay, which together cover >70% of the population of the EU & Norway, formed “European Network for Payments”, a Madrid-based JV connecting national schemes serving 130M users across 13 countries - a fresh push for payments sovereignty as reliance on Visa/Mastercard are said to be core vulnerabilities for European infrastructure (here)
Kemi Badenoch published “The Right Way”, a 83-page policy framework arguing Conservative failure stemmed from reactionary governing without first principles, & that UK stagnation reflects a state that drains responsibility from individuals & enterprise. The document criticises PM Andy Burnham’s ‘public control’ agenda, advocates reducing state interference in growth creation, & warns UK debt interest consumed 80% of new borrowing in 2025-26 - borrowing just to service past debt rather than invest (here)
Trump created a ‘Super Intelligence Force’ led by DNI Jay Clayton to coordinate federal AI policy, engagement with industry, & international competition - part of the administration’s push to rebrand AI as ‘super intelligence’ & counter China’s development efforts (here)
Trump launched America.gov, an AI-powered portal consolidating 10,000+ federal forms & services into a single search interface, led by Joe Gebbia (Airbnb co-founder, now US chief design officer). A test of whether citizen-facing AI can cut through bureaucratic friction at national scale (here)
Strategic Sectors
AI
Bridgewater Associates (managing ~$100B) projects AI could dislocate 18% of US labour if unregulated, warning of potential “societal breakdown” even as the firm raised $2B in 2024 for an AI-driven fund that has since beaten the market - a tension between deployment enthusiasm & systemic risk that now echoes across macro investors (here)
Yann LeCun dismissed existential AI risk concerns as ‘paranoia’ driven by effective altruism, called Anthropic CEO Dario Amodei ‘deluded’, & argued that recent rogue AI incidents (including OpenAI agents hacking Hugging Face) stemmed from poor cybersecurity design rather than inherent AI danger (here)
Bank of England governor Andrew Bailey warned that AI investment flows could trigger market corrections, citing historical precedent (Google displaced Netscape) & noting “everybody is currently priced to be a winner” (here)
KKR warned that AI debt exposure could reach 20% of the US investment-grade bond index by 2030 as tech firms deploy $8T on infrastructure, far exceeding the highest sector concentration (2.6%) over the past 29 years - with limited upside for creditors & growing concern over off-balance-sheet commitments, cross-exposure among hyperscalers, & weak lease structures in data centre deals (here)
Aleph Alpha (German sovereign AI champion) released Kolibri, a 78B-parameter mixture-of-experts model optimised for German-language government & industry applications, with 23% German pre-training data, documented compliance screening against 4.5M URLs, & native support for RAG & agentic workflows - a concrete demonstration that European labs can ship mission-critical models without ceding data sovereignty to US or Chinese hyperscalers (here)
Businesses struggle to predict AI costs because token consumption varies unpredictably by task & model - Stanford/CMU research found cheaper models cost more than expensive ones 32% of the time, & only 11% of 400 businesses surveyed could accurately forecast AI spending (here)
Notable deal:
ElevenLabs doubled its valuation to $22B in a $300M secondary sale led by Wellington & T Rowe Price, joining the top 10 most valuable AI labs outside China. The London-based voice AI company, which trains its own models on proprietary infrastructure, has expanded to 800 staff across 20 countries & plans to be IPO-ready within 2-2.5 years (here)
Cybersecurity
ASOS app users received push notifications from hackers claiming to have breached the retailer’s Snowflake data storage & demanding ransom, a brazen tactic that suggests the attackers compromised both backend data systems & the app’s notification infrastructure - turning customer phones into a public extortion vector (here)
Denmark’s Central Person Register was breached in September, exposing records of ~8M citizens - names, addresses, & social security numbers stolen via compromised corporate access. Likely the country’s largest breach, echoing earlier national ID database attacks in Turkey (2016) & India’s Aadhaar leaks (here)
Defence
US Defence Secretary Hegseth announced a new four-star Autonomous Warfare Command (AutoWarCom) to accelerate procurement & deployment of drones & robotics, targeting October 2027 for operational status; Elon Musk will advise on future warfare studies (here)
UK Conservatives pledge £10B for Britannia Shield, an Iron Dome-style missile & drone defence system protecting bases & critical infrastructure, as part of a 3% GDP defence commitment by 2030. The proposal responds to Russian drone activity, Iranian strikes on RAF Akrotiri & Diego Garcia, & concerns that Labour’s £790M allocation over a decade is inadequate - Greece has spent billions on its Achilles Shield while the UK has ‘next to nothing’ according to the Commons defence committee (here)
The Pentagon has stopped using Anthropic’s Claude after designating it a supply chain risk in February, though sources say it remained embedded in Palantir’s Maven intelligence system for intelligence gathering & military ops against Iran until last weeks (here)
Robotics
Notable deal:
RobCo (autonomous industrial robotics) hit $1B valuation via secondary share sale 9 mo after its $100M Series C, with Sequoia, Lightspeed & new backers Cherry Ventures participating. The Munich-founded company is preparing to launch Alfie, its autonomous factory robot, in March 2027 whilst scaling US operations from Austin & San Francisco - a European hardware challenger in a market where Toyota-backed Walden ($300M seed at $1.1B) & Standard Bots ($200M Series C) are also racing to deploy AI-powered robots across American factory floors (here)