the Now & the Next

A bi-weekly speculative fiction suggesting the shape of things to come.
(sourced from trustworthy trade pubs, think tanks + frontier science news)

1,658 Signals Tracked
7 Collisions Identified
10 Frontier Science Cards
220+ Source Domains

This fortnight's signals kept circling back to the same question: what happens after the AI rollout, when the results come in? Ford rehired 350 of the engineers it let go, because the AI wasn't good enough without them. Gartner says software teams will shrink 60% by 2029 — while California's own labor data shows no statewide AI job-loss signal yet. Washington built an entire program to export the American AI stack, then embargoed Anthropic's own models from foreign users. A Virginia county killed its second major data center this year, and a grid think tank now calls electricity infrastructure a military target. Banks are deploying agentic AI to build their own apps while European regulators call frontier AI a threat to the financial system it's being wired into. Unilever told its ad agencies it doesn't need the big idea anymore. And publishers are quietly building the walls back up against the crawlers that trained the models in the first place. Seven collisions from the recalibration underway.

01
The Great AI Rehire

Ford brought back 350 engineers it laid off for AI, while Gartner predicts 60% of orgs will run on tiny AI-augmented teams by 2029 — and Apple's own researchers found multi-agent AI teams underperform a single good expert.

+

Ford is re-hiring 350 former engineers, deploying their experience to train younger staff and fix the AI-driven quality-control tools that weren't performing as promised. The move lands in the same fortnight that Gartner forecast 60% of organizations will run “tiny teams” of 2-5 people by 2029, up from just 15% today — a bet that AI augmentation, not headcount, drives software output. Meanwhile labor tension is rising underneath both data points: Industry Week reports manufacturing's share of corporate profits has fallen sharply since 1988, with the UAW using AI-job-loss fears to organize ahead of Detroit contract talks. The California Policy Lab's new CAIT tool, built with the state's Employment Development Department, found no statewide rise in unemployment claims tied to AI-exposed occupations through May 2026 — even as regional pockets show variation. And Apple ML Research published results showing self-organizing multi-agent LLM teams consistently underperform their single best expert agent, by as much as 41.1% on ML benchmarks.

⚡ The Now

The AI-team-size story has split into two contradictory data sets running in parallel: Gartner's confident 2029 forecast of shrinking teams, and California's actual 2026 unemployment data showing no such shift has happened yet. Ford's rehire is the ground-truth data point connecting them — a real company that tried the smaller-team bet, found the AI wasn't sufficient on its own, and brought the humans back. Apple's own research supplies the mechanism: coordinating multiple AI agents is often worse than trusting one good expert, which undercuts the premise that stacking more agents substitutes for stacking more people.

→ What's Next

Expect 2026-2027 to produce a wave of quiet reversals like Ford's, as companies that cut early discover the gap between AI capability and AI reliability at the edge cases experienced humans used to catch. Gartner's 2029 number may still prove directionally right, but the path there looks less like a straight line and more like Ford's: cut, discover the shortfall, rehire, then retry with better tooling. Watch California's CAIT dataset over the next few quarters — it's the first real-time public instrument for catching whether AI job displacement is actually happening at scale, rather than being assumed from vendor roadmaps.

Computerworld
Ford is rehiring 350 experienced engineers to train younger employees and fix AI tools used in quality control, recalibrating its AI commitment with seasoned human expertise.
Industry Week
Manufacturing labor's share of profits has fallen sharply since 1988; wage stagnation and AI-job-loss fears are driving increased union organizing ahead of UAW-Detroit automaker negotiations.
Gartner
Gartner forecasts 60% of organizations will run 'tiny teams' of 2-5 engineers by 2029, up from 15% in 2026, framing it as an operational evolution rather than a cost-cutting move.
Route Fifty
California's new CAIT tool found no statewide increase in unemployment claims for AI-exposed occupations from January 2019 through May 2026, despite some regional variation.
Apple ML Research
Apple ML Research found self-organizing multi-agent LLM teams consistently underperform the single best expert agent, with drops up to 41.1% on machine learning benchmarks.
02
The Model Export Wall

Washington built a program to export the American AI stack worldwide — then embargoed Anthropic's own models from foreign users, while China moves to lock down its models too.

+

The U.S. Department of Commerce launched the American AI Exports Program to push full-stack American AI — hardware, data pipelines, models, cybersecurity — abroad as a countermeasure to the spread of Chinese open-weight models. Days later, the Trump administration directed Anthropic to restrict foreign nationals' access to its Mythos 5 and Fable 5 models, a national-security move that directly undercuts the export program's own goal, per Brookings' analysis. Simultaneously, China is moving to impose its own export curbs on its top AI models, leaving Europe — long reliant on cheaper Chinese models and short on competitive alternatives beyond firms like Mistral — caught in the middle of both restrictions, per The Decoder. CSIS's parallel analysis of South Korea frames the stakes in ecosystem terms: nations that control integrated AI ecosystems — combining science, industry, capital, and standards — gain both economic and geopolitical leverage over those that merely export components.

⚡ The Now

Both AI superpowers are simultaneously trying to sell their models abroad and lock down access to their best ones — the same week. The U.S. wants full-stack AI export dominance but flinched at letting foreigners touch Anthropic's frontier models; China wants to keep selling cheap open-weight models but is tightening its own top tier. The result is a widening gap between what both countries are willing to export at scale (mid-tier, commoditized capability) and what they're willing to let leave the building (frontier capability) — and Europe sits in the gap with neither.

→ What's Next

Expect a two-tier global AI market to formalize: a freely-traded layer of mid-tier open and semi-open models, and a tightly-guarded frontier layer available only to trusted allies or not at all. South Korea's ecosystem-versus-export-nation choice, as CSIS frames it, will become the template question every mid-sized AI power has to answer: build the full stack domestically, or stay dependent on whichever superpower's export policy is least restrictive this quarter. Europe's dependency problem won't resolve until it has a frontier model of its own worth restricting.

CSIS
The U.S. Department of Commerce launched the American AI Exports Program to promote full-stack American AI technology packages abroad as a countermeasure to Chinese open-weight model diffusion.
Brookings Institution
The Trump administration directed Anthropic to restrict foreign nationals' access to its Mythos 5 and Fable 5 models, contradicting its own strategy of promoting American AI exports globally.
The Decoder
China is moving to restrict access to its leading AI models even as Europe remains dependent on affordable Chinese models and lacks competitive domestic alternatives beyond firms like Mistral.
CSIS
CSIS argues the global economy increasingly rewards nations that control integrated ecosystems of science, industry, capital, and standards over those that merely export components.
03
The Grid Says No

A Virginia county killed a 43-million-square-foot AI data center, a defense think tank calls the power grid a military target, and Georgia Power is now risk-scoring which AI projects even get to plug in.

+

The Prince William Board of County Supervisors unanimously voted to kill the proposed Dulles South Innovation Center — a planned 43-million-square-foot AI data center campus — the second major Virginia data center project to die this year, per Data Center Knowledge. CIMSEC's “The Grid is the Arsenal” analysis frames the stakes in stark terms: U.S. electricity demand is projected to rise 20% by 2030 from data centers and manufacturing, even as transmission permitting delays and local opposition now block $64 billion in pending data center investment. Brookings separately reports that the Guardrails Alliance, an employee-led coalition drawn from major AI companies, is organizing small-dollar donations to push for stronger AI oversight — a second front of organized pushback alongside the community fights over physical infrastructure. In response, Georgia Power has adopted a risk-adjusted, probabilistic forecasting method to evaluate a 22.8 GW pipeline of large-load data center requests, informing over 1,100 miles of new planned transmission.

⚡ The Now

The AI buildout has hit a wall that isn't compute or capital — it's the physical grid and the communities sitting on top of it. Prince William County's unanimous vote shows local government is now willing to kill nine-figure data center projects outright, not just slow-walk permits. CIMSEC's framing of the grid as military infrastructure elevates the stakes from a zoning dispute to a national-security one. Georgia Power's response — risk-scoring which projects are credible enough to warrant grid investment — is the industry's first real attempt at triage, choosing which AI ambitions get power and which don't.

→ What's Next

Expect utilities to formalize Georgia Power's risk-scoring approach into a de facto gatekeeping function for AI infrastructure — deciding which data center projects are real enough to build toward, rather than building blind and hoping demand shows up. Local backlash like Prince William's will keep killing marginal projects in areas without pre-negotiated community and grid buy-in, pushing developers toward regions with spare transmission capacity and friendlier zoning. And watch for the Guardrails Alliance model — AI-company employees organizing against their own industry's unchecked expansion — to spread as a second, harder-to-dismiss form of backlash alongside NIMBY zoning fights.

Data Center Knowledge
Prince William County unanimously killed the proposed 43-million-square-foot Dulles South Innovation Center data center campus, the second major Virginia AI data center project to die this year.
CIMSEC
U.S. electricity demand is projected to rise 20% by 2030 from data centers and manufacturing, but permitting delays and local opposition now block $64 billion in pending data center investment.
Brookings Institution
The Guardrails Alliance, an employee-led coalition from major AI companies, is mobilizing small-dollar donations to push for stronger regulatory oversight of AI development.
Data Center Knowledge
Georgia Power adopted risk-adjusted probabilistic forecasting to evaluate a 22.8 GW pipeline of large-load data center requests, informing plans for over 1,100 miles of new transmission.
04
Agencies Are Optional Now

Unilever told its agencies it doesn't need the big idea anymore, Albertsons is running micro-dramas with P&G product placement, and Ulta built its growth strategy on TikTok creators instead of campaigns.

+

Unilever is restructuring its marketing operation to unify content, media, commerce, creative, and planning into one continuously-informed system built around creator relationships and community data — explicitly moving away from agency-driven, siloed campaign development. Albertsons Media Collective launched Rico's Tacos, a micro-drama series weaving Procter & Gamble products directly into the storyline across social platforms, the Albertsons app, and in-store screens. Ulta Beauty became the first U.S. specialty beauty retailer on TikTok Shop, leaning on creator-led storytelling and TikTok-first moments for customer acquisition, while Glossy's 2026 creator marketing research — covering Duolingo, Ulta, and YouTube — found Ruggable scaled its influencer program 20x by partnering with a single ShopMy creator. And KFC is rolling out a loyalty program to all 20 of its top global markets, explicitly copying the digital playbook Taco Bell pioneered.

⚡ The Now

The common thread across four different retail categories — CPG, grocery, beauty, and fast food — is that the unit of marketing production has shifted from the campaign to the creator relationship. Unilever's restructuring is the clearest statement of intent: the “big idea” workflow that agencies were built to produce is being replaced by always-on creator and community data loops. Albertsons and Ulta show what that looks like in practice — product placement inside a micro-drama, brand growth built on a single social platform's creator economy — while Ruggable's 20x scale-up from one creator partnership shows the economics can outperform a traditional campaign.

→ What's Next

Expect traditional ad agencies to keep losing the mid-funnel briefs first — the ones an internal team plus creator platform can now handle directly — while retaining brand-strategy and crisis work where judgment still matters more than content velocity. KFC borrowing directly from Taco Bell's loyalty playbook signals category-wide convergence is coming faster than internal R&D cycles used to allow, since proven creator and loyalty tactics now travel between competitors almost immediately. Watch retail media networks like Albertsons' to keep absorbing ad budget that used to go to traditional agencies, since they can now deliver the content and the distribution in one product.

The Drum
Unilever is restructuring its marketing workflow to unify content, media, commerce, creative, and planning into a connected system driven by creator relationships and community data.
The Drum
Albertsons Media Collective launched Rico's Tacos, a micro-drama series integrating Procter & Gamble products, delivered across social platforms, the Albertsons app, and in-store screens.
Retail Dive
Ulta Beauty became the first U.S. specialty beauty retailer to launch on TikTok Shop, using creator-led storytelling and TikTok-first moments for customer acquisition.
Glossy
Ruggable partnered with creator Dan Pelosi via the ShopMy platform to co-create a product collection, scaling its influencer marketing program 20x.
Restaurant Business
KFC is expanding its loyalty program to all of its top 20 global markets, adopting the digital and loyalty strategies pioneered by Taco Bell.
05
The Web Builds a Wall Against the Bots

Publishers are preparing to block Google's crawler, Cloudflare now blocks AI training bots by default, and Tripadvisor is selling its billion human reviews as a trust asset in a bot-majority web.

+

Major digital publishers are preparing to block Google's web crawler to stop unlicensed use of their content for AI training, per Adweek — a direct challenge to the search traffic relationship publishers have depended on for two decades. Cloudflare went further, introducing Content Signals for robots.txt that will block bots designated for LLM training or agentic use by default starting September 15, while still allowing search-engine bots through. Tripadvisor is leaning into the resulting scarcity, marketing its billion-plus human-generated travel reviews and AI systems that prioritize sourced, recent content as a trust differentiator in an increasingly bot-populated web. And in ad tech, LiveRamp is positioning its media-neutral, agentic-AI-powered identity and data products directly against the uncertainty this is creating for advertisers who no longer know which traffic is verifiably human.

⚡ The Now

The content-for-traffic bargain that built the modern web — let Google crawl you, get search visitors back — is being renegotiated in real time, and publishers are the ones pulling back first. Cloudflare's default-block policy for LLM training bots turns what used to be a case-by-case opt-out into an opt-in system for AI companies, a structural shift in who has the leverage. Tripadvisor and LiveRamp are both, in different ways, selling the same thing into that gap: proof that the content or the data behind a decision came from a verified human, not a scraped or synthetic source.

→ What's Next

Expect content-licensing deals to become the default path for AI companies that want publisher data, replacing the assumption of free crawl access that trained the current generation of models. Cloudflare's bot-blocking-by-default architecture will likely get copied by other CDN and infrastructure providers, effectively making an opt-in licensing layer the new standard between the open web and AI training pipelines. And watch human-verification becoming its own product category — Tripadvisor's billion-review moat and LiveRamp's identity tools are early versions of what will likely become a standard trust layer for any platform competing against synthetic content at scale.

Adweek
Major digital publishers are considering blocking Google's web crawler to prevent unlicensed use of their content for AI training, reflecting growing tension over content ownership and compensation.
AdExchanger
Cloudflare introduced Content Signals for robots.txt, blocking bots designated for LLM training or agentic use by default starting September 15, while still permitting search engine bots.
Adweek
Tripadvisor is marketing its billion-plus human-generated travel reviews and recency-prioritizing AI systems as a trust differentiator in an increasingly bot-populated web.
AdExchanger
LiveRamp launched a five-month B2B ad campaign targeting executives on Netflix, positioning its media-neutral, agentic-AI-powered identity and data tools against advertiser uncertainty over verified traffic.
06
Europe Builds Its Own Defense Stack

The U.S. pulled 5,000 troops from Germany, CSIS wants the Ankara summit to sideline Washington entirely, and the Dutch Navy just ran its first multidomain unmanned sea trial without waiting for anyone.

+

The Trump administration has withdrawn roughly 5,000 U.S. troops from Germany over six to twelve months and canceled a planned 2026 long-range fire battalion deployment, per the Council on Foreign Relations — a unilateral drawdown straining transatlantic cohesion ahead of the Ankara summit. CSIS is urging allies to use that summit to revisit the Berlin Plus agreement and reduce Turkey's veto power over EU military operations, explicitly aiming to let European states access NATO structures without U.S. involvement. The European Commission's EU Tech Sovereignty Package, per the Belfer Center, commits roughly €2 billion over seven years to make public-sector bodies “buyers of first resort” for open-source software, reducing dependency on proprietary (largely American) technology. And operationally, the Royal Netherlands Navy ran its first Maritime Uncrewed Sea Trials, coordinating UAVs, USVs, and UUVs around a patrol vessel via its own Intelligent Distributed Uncrewed Systems command platform, while the Royal Netherlands Air Force and TNO are jointly building jamming-resistant communications and GPS-independent navigation using Earth's magnetic field.

⚡ The Now

Europe isn't just talking about strategic autonomy this fortnight — it's building the institutional and technical plumbing for it simultaneously. CSIS's proposal to route around a Turkish veto and the U.S. troop drawdown are two sides of the same coin: American and Turkish leverage over European defense decisions are both being challenged at once. The EU Tech Sovereignty Package extends that logic from hardware and troops to software dependency. And the Dutch military trials show the operational layer isn't waiting for the diplomatic layer to catch up — unmanned, sovereign-controlled systems are already being tested in the field.

→ What's Next

Expect the Ankara summit outcome to be read as a referendum on how much of NATO's command structure Europe is willing to operate without direct U.S. participation. If the Berlin Plus revision CSIS proposes gains traction, watch for European nations to accelerate procurement of the kind of uncrewed systems the Dutch Navy just trialed, since autonomous platforms let a smaller pooled European force multiply its reach without matching U.S. troop levels. The EU's open-source push is the slower-moving but longer-lasting piece — a seven-year bet that reducing software dependency compounds the same way reducing troop dependency does.

Council on Foreign Relations
The Trump administration has removed roughly 5,000 U.S. troops from Germany and canceled a planned 2026 long-range fire battalion deployment, straining transatlantic security cohesion.
CSIS
CSIS proposes revisiting the Berlin Plus agreement to reduce Turkey's veto power over EU military operations and enhance European access to NATO structures without U.S. involvement.
Belfer Center
The European Commission's Open-Source Strategy commits roughly €2 billion over seven years to make public-sector bodies buyers of first resort for open-source software, reducing dependency on proprietary technology.
Defensie.nl
The Royal Netherlands Navy ran its first Maritime Uncrewed Sea Trials, coordinating UAVs, USVs, and UUVs around a patrol vessel using its own Intelligent Distributed Uncrewed Systems command platform.
TNO
The Royal Netherlands Air Force and TNO are developing jamming-resistant communications, GPS-independent navigation via Earth's magnetic field, and AI-powered threat detection for contested environments.
07
The Bank Deploys the Agent, the Regulator Flags the Risk

PNC built its new banking app with 115 AI agents, Allianz is cutting 1,800 jobs to go AI-first, and European regulators just called frontier AI a threat to the financial sector it's being wired into.

+

PNC built its new mobile banking app using roughly 115 AI agents to autonomously manage development and testing, accelerating feature releases. Allianz SE is cutting up to 1,800 European jobs as part of a shift to expand AI tools across its insurance operations. At the same time, the European Supervisory Authorities — covering securities, banking, insurance, and pensions — endorsed the European Systemic Risk Board's warning that AI-enabled cyberattacks threaten financial-sector operational resilience, while the UK Financial Conduct Authority found 26% of retail investors already trust public AI tools like ChatGPT, Claude, or Gemini for financial advice, with 20% likely to use autonomous AI agents within preset limits going forward. Separately, Lloyds Banking Group appointed a new Chief Data and AI Officer as Banking Dive reports a broader wave of similar hires across Commonwealth Bank of Australia, TD Bank, and BNY Mellon.

⚡ The Now

Banks and insurers are running two AI strategies at once that don't fully reconcile: using agentic AI internally to build products and cut costs (PNC, Allianz), while their own regulators warn that frontier AI is a systemic risk to the sector both companies operate in. The FCA's finding that a quarter of retail investors already trust chatbots for financial advice shows consumer adoption is outrunning the regulatory framework meant to govern it. The wave of Chief AI Officer hires at Lloyds, Commonwealth Bank, TD, and BNY Mellon is the industry's answer to that gap — centralizing AI governance before regulators force the structure on them.

→ What's Next

Expect the Chief AI Officer role to become as standard in banking as Chief Risk Officer within the next two years, precisely because regulators like the FCA and ESRB are treating AI risk as systemic rather than departmental. Allianz's job cuts are a preview of where insurance heads next: AI absorbing the operational layer while headcount concentrates around oversight and the newly regulated AI-governance function. Watch for the first formal AI-specific capital or resilience requirement from a major financial regulator within the next several quarters — the ESRB's warning reads like the opening move toward one.

Banking Dive
PNC deployed roughly 115 AI agents to autonomously manage software development and testing workflows for its new mobile banking app, accelerating feature releases.
Business Insurance
Allianz SE plans to cut up to 1,800 jobs in Europe as part of its strategic shift to expand AI tools across its insurance operations.
Investment Executive
European Supervisory Authorities endorsed the European Systemic Risk Board's warning that AI-enabled cyberattacks threaten the operational resilience of financial institutions.
Investment Executive
The UK FCA found 26% of retail investors trust public AI tools like ChatGPT, Claude, or Gemini for financial advice, with 20% likely to use autonomous AI agents within preset parameters.
Banking Dive
Lloyds Banking Group appointed a new Chief Data and AI Officer, part of a broader wave of similar leadership hires across Commonwealth Bank of Australia, TD Bank, and BNY Mellon.

Frontier Science Feeding the Machine

The research signals underneath the enterprise headlines — breakthroughs from labs and universities that will restructure industries within a decade.

AI / Machine Learning
Researchers built large-scale tabular foundation models like FlexTab, TabICL, and iLTM that outperform commercial solutions on structured data — the kind of data LLMs handle poorly.
Photonics / Imaging
A Chinese research team's UMA model combines nested UNet architecture with multi-attention mechanisms to reconstruct hidden scenes from indirect light, outperforming prior passive NLOS imaging methods.
AI Infrastructure
Nvidia expanded NVLink Fusion to include photonics partners Ayar Labs, Marvell, and Lightmatter, bringing optical interconnects into data center racks alongside traditional copper links.
AI Safety
Anthropic and AE Studio developed GRAM, a method that compartmentalizes sensitive knowledge like virology or nuclear physics into removable modules that can be selectively deleted from a trained model.
AI Cognitive Science
Anthropic found that Claude spontaneously develops an internal 'J-space' during training — a global workspace enabling silent internal reasoning, echoing theories of conscious access in human brains.
Energy
UC Berkeley and SRI International are testing underwater kites that fly figure-eight paths to harvest tidal energy, piloting prototypes in San Francisco Bay for remote coastal power generation.
Computational Linguistics
UC Berkeley's ConlangCrafter generates constructed languages nearly twice as diverse and 70% more consistent than general-purpose LLMs like Gemini-2.5-Pro attempting the same task.
Neurotechnology
A Nature Reviews Bioengineering study found brain-computer interfaces for artificial vision and touch prostheses share nearly identical underlying principles, pointing toward a unified sensory-restoration platform.
Astronomy
KU Leuven and Peking University developed a 3D modeling method using Lyman-alpha ultraviolet absorption to reconstruct the shape of astrospheres, including the heliosphere surrounding our own Solar System.
Biotech / Instrumentation
University of Geneva researchers combined nanopore detection with AI signal analysis to identify individual proteins label-free, regardless of charge, enabling precise single-molecule proteomics.