Strategic Initiatives
12514 stories
·
45 followers

Why Extra Cushion in Your Shoes Might Actually Be Bad for Your Feet - WSJ

1 Share

LLM (meta/muse-spark-1.3-contributor) summary:

  • Thick Foam Origins: introduced in twenty ten with up to two inches of foam
  • Biomechanical Instability: soft base allows extra foot torque and leg rotation
  • Balance Reduction: added height lowers sensory feedback and raises fall risk
  • Lawsuits Allegations: athletes claim injuries from high stack carbon plate models
  • Performance Gains: studies note better economy and faster race times
  • Increased Load: longer ground contact raises stress on muscles and bones
  • Comfort Masking: less soreness may hide overtraining and invite injury
  • Rotation Strategy: vary cushioning and introduce new shoes very gradually

Oct. 10, 2026 4:00 pm ET

Illustration of three people walking, with their feet replaced by burgers, brooms, and clouds, and teddy bears strapped to their legs. Klaas Verplancke for WSJ

If you don’t own a pair of so-called high-stack or max-cushion running shoes yourself, no doubt you’ve seen others wearing them, whether at the gym, on trails or just knocking about town. Sandwiched between the tire tread-like outer sole and nylon upper is as much as 2 inches of soft and springy foam, which gives these shoes both the look and feel of walking around with mini-mattresses on your feet.

It all started back in 2010 with the introduction of the Hoka line of max-cushion shoes, which led other brands to develop similar styles, each stacked with its own patented lightweight foam. Now it is difficult to find running or walking shoes that aren’t raised on a platform of puff, some with the added oomph of a layer, or plate, of carbon fiber.

While max-cushion shoes may feel super comfy, and can improve running efficiency and performance if embedded with a carbon plate, there are concerns among podiatrists, orthopedists, kinesiologists and sports-medicine doctors about the long-term effects, not only on your feet, but also your ankles, knees and hips.

“They change people’s biomechanics in a way that can be unpredictable, and brings up the question of potential injury risk,” says Dr. Adam Tenforde, director of running medicine at Mass General Brigham’s National Running Center.

A matter of mechanics

The problem comes down to basic biomechanics, says Dr. Kenrick Dennis, a podiatrist in Houston who served as medical director for the Houston Marathon.

“If you are standing on that marshmallow, it allows your foot to torque in and out a little bit more, the leg rotates to follow it, and that rotation is translated through the ankle, the knee and the hip,” he says.

It takes a bit longer to sink into the cushiness and subconsciously find your balance before you push off again, similar to the extra effort required when walking on sand. The wobbliness, he says, plus the added height of max-cushion shoes, also increases the risk of twisting an ankle or falling.

Moreover, max-cushion shoes diminish proprioception (awareness of where you are in space), since you have less sensory feedback of where you are vis-à-vis the ground.

A number of civil lawsuits have been filed against shoe manufacturers by track stars, including an Olympian, who allege, among other things, that wearing high-stack shoes with carbon plates caused injuries that derailed their careers. Nike faces one such suit in California Northern District Court. Puma—the parent company as well as some of its divisions and partners—was named in three cases filed in Massachusetts Superior Court.

Nike declined to comment on pending litigation, but in court filings the company denies its product is “defective and unsafe for its intended purposes” and neither “caused injury and/or contributed to the risk of injury.” Puma said in a statement that it strongly denies its shoes cause injury. The company is further pursuing a countersuit against its three accusers in a German court.

While the suits focused on shoes that included a carbon plate, Tenforde says max-cushion shoes can modify biomechanics independent of that feature. Hoka didn’t respond to requests for comment about safety concerns.

Of course, back when barefoot running was the rage in the mid-2000s, alleged injuries caused by zero-stack, or “barely there,” shoes resulted in lawsuits against manufacturers.

“The pendulum has swung from the barefoot shoes to now having a marshmallow under your foot,” says Dennis, the Houston podiatrist.

Proponents of max-cushion shoes, particularly those with an embedded carbon plate, point to studies that suggest the shoes improve running economy and performance by 4% and 2%, respectively. A 2023 review published in Sports Biomechanics says that every world record from the marathon to 5 kilometers has been broken by athletes wearing these kinds of shoes. The comfort and springy design may allow athletes to tolerate higher training volumes or intensities.

“If a shoe makes running easier, you will run faster,” says Dr. Laurent Malisoux, a sport and physical-activity research leader at Luxembourg Institute of Health. “But people should remember that if they get an injury, it’s because they probably ran too much, too fast or too soon, not because of the shoe.”

Stressing the muscles

Nevertheless, concerns remain. A 2018 study published in Nature indicates max-cushion shoes actually increase the total load on muscles, bones and joints compared with more-moderately cushioned shoes in part because the foot is on the ground longer.

“You’re hitting the ground with the same magnitude of force,” says Dr. Daniel Lieberman, a Harvard evolutionary biologist whose research focuses on how modern footwear has altered human biomechanics. “But you’re spreading that force over a longer period of time.”

Your leg also appears to stiffen more when you land, intensifying the jolt, according to the authors of the Nature study. If there is a carbon plate in the shoe, it can help you spring back into the air a bit faster, but Lieberman says more research is needed to determine whether the total load is larger than in more moderately cushioned or minimalist shoes.

While max-cushion shoes may feel great initially, repetition of these minute wobbly, twisting moments plus the extra mechanical load on muscles, tendons, joints and bones could lead to injury in the long term, critics say. During a mile run, each foot will strike the ground on average about a thousand times. Running 20 miles a week adds up to a million foot strikes a year.

There is also the concern that the squishiness of max-cushion shoes and/or the location of the flex point, particularly if there is a carbon-fiber layer, may also shift where on your foot you land and push off. That might make you more vulnerable to foot fractures, according to a 2023 analysis of high-stack running shoes with carbon-fiber plates published in Sports Medicine, with Tenforde as the lead author.

Scott Christensen, lead endurance instructor for USA Track & Field, says while he appreciates the potential performance-enhancing qualities of bouncy foam and carbon springboards, “everything has a cost-benefit ratio to it.” One of the more counterintuitive costs, he says, is the shoes’ exceptional comfort. Runners don’t get as sore, which is the body’s way of saying: Hey! Back off before you do damage.

“As a coach, I want soreness,” Christensen says. “I depend on soreness in my athletes. Soreness is the governor of training.”

Making good choices

So, what is a person to do when confronted with a wall of shoes in a run shop or sporting-goods store? “The answer is there is no one answer,” Christensen says. People have different physiologies, exercise regimes and injury histories which might dictate not only level of cushioning, but also support, stability and heel-to-toe drop of the shoe.

Whether you’re a runner, walker or maybe someone whose job requires you to be on your feet all day, the best shoe is the shoe that fits. You may think you’re wearing the right size, but a 2018 study in the Journal of Foot and Ankle Research suggests as many as 72% of people wear shoes that don’t accommodate either the width or length of their feet. Your shoes shouldn’t hurt or feel uncomfortable either during or after physical activity.

And don’t pay attention only to the pillowy feel under your feet, Dennis says. Test your ability to maintain balance and coordination in the shoes by doing things like raising up on your toes and holding and then rocking back on your heels and holding, as well as doing quick turns and pivots. Notice any additional workload or twisting moments demanded of the joints and muscles in your legs and hips. Also listen to your footfall. It should be light and have a steady, crisp cadence—no clomping, scuffing or shuffling, as can happen when you have too much bulk under your feet.

Beyond that, Christensen’s advises training in more than one kind of shoe, varying amount of cushioning and stack height. The combination will vary stressors on your body and reduce the chance of repetitive foot strike and motion injuries.

Dr. Stéphane Bermon, director of the health and science department at World Athletics—the international governing body for track and field, cross-country running, road running, race walking, mountain running and ultra running—is also a proponent of the shoe-rotation strategy, but he cautions to introduce any new shoe “very, very progressively” to avoid injury. Maybe running or walking in a new kind of shoe once a week for one week and then twice a week the following week, and so on.

“In biomechanics, in human physiology, what kills you is routine,” Bermon says. “Whether it’s your bone, whether it’s your muscle, whether it’s your tendon, you have to challenge yourself every day with different things.” In shoes, just as in life, “There is nothing worse than monotony.”

Write to reports@wsj.com

Copyright ©2026 Dow Jones & Company, Inc. All Rights Reserved. 87990cbe856818d5eddac44c7b1cdeb8

Kate Murphy is a contributor to The Wall Street Journal. She is a Houston-based journalist and author of "You’re Not Listening: What You’re Missing and Why It Matters" and "Why We Click: The Emerging Science of Interpersonal Synchrony."


Up Next


Videos

Read the whole story
bogorad
57 minutes ago
reply
Barcelona, Catalonia, Spain
Share this story
Delete

unikernels were hard. key word: were.

1 Share

LLM (meta/muse-spark-1.3-contributor) summary:

  • Unikernel Definition: application includes operating system functions as libraries
  • Past Development Challenges: limited storage support required extracting drivers from other systems
  • Security Benefits: smaller attack surface without shell or interpreter access
  • Library Porting: artificial intelligence enables rapid conversion between programming languages
  • Storage Approach: remote object storage combined with local cache for hot data
  • Nix Testing: automated multi machine tests validate network and application interaction
  • Distributed Actors: many machines merged into single addressable heap with rehydration
  • Language Tradeoffs: fast compilation provides more feedback for automated code generation

Justin Cormack, who worked on MirageOS and Unikernel Systems back in the day, has been noticing what I've been noticing: people are discovering (or rediscovering) unikernels again. He's running a series of conversations on the topic for his newsletter, and I was first up. He emailed me, and five minutes later I was talking to him from the pub with a stein of beer in hand. We went deep on Mirage, Orleans, Haskell, Nix, Cursed and a whole lot more.

Here's the gist, written up properly, with chapter links at the bottom if you want to jump to a specific part of the conversation. Justin's edited transcript is over at Ignore Previous Directions.

Unikernels were hard. Key word: were. Now we have AI.

what a unikernel is, and why they were hard

I first ran into unikernels around 2015. I'd staffed up a team of Haskellers and gone pretty deep on functional programming. Where there are Haskellers, there are OCaml programmers, and from there you find MirageOS. Great idea. I played with it back then.

A unikernel is the idea that your application is the operating system. There's no userland. If you want a web server, DNS, or to send an email, there's nothing you can fork or spawn. You have to write those things as libraries in your application.

That was the friction. Justin remembers it well: when they were building Mirage, they had a TCP stack and an HTTPS stack, but there was almost nothing for storage. They were pulling drivers out of NetBSD because they could run them in userspace. It was hard back then.

There's a lot of dogma in our industry. Nix is hard. Bazel is hard. Unikernels are hard. Yes, they were. These hard concepts are now in the model weights. All you've got to do is prompt for them and, cognitively, get rid of the dogma that they're hard.

the operating system is design debt

Every application that isn't a unikernel was built on the assumption that there's an application, and then there's an operating system underneath. Why do we even have an operating system? Because forty years ago there was a human operator. I've done IBM 5250, AIX, Solaris, and mainframes. The multi-user operating system exists because a person sat in front of it, and then we put the application on top.

I consider that design debt, and here's why it matters right now. Applications get popped. They were getting popped before AI. Someone pops the userland application and gets a shell. That shell is a VIP butler service for exfiltration.

With a unikernel, the attack surface is much smaller. If the functionality isn't in the application (the operating system), then the attacker is screwed. There's no next hop.

Justin pushed back here, and fairly. Attack-surface reduction is something people are very fuzzy about. You can remove the shell from a Linux container, but almost every Linux environment still has something that's effectively an interpreter. You can execute a new program without a writable filesystem. You've still got memory safety and gadgets to worry about.

All true. But look at what we've been doing for twenty-six years. The earliest adage I remember from the SunOS and cgi-bin days was "don't put the compiler on production." Then came build containers and production containers. Then Chainguard. We keep chipping away at the attack surface, instead of going in the opposite direction and ensuring there is no attack surface.

And this is the bit people miss: if there's no shell and no interpreter, there's nothing in the model weights that knows what to do next. That turns a drive-by (pick your framework's RCE of the week and you've got a shell, and the model weights know what to do with shell access) into a targeted attack that needs your source code.

you can just port the missing libraries now

The classic objection: your unikernel needs to talk to Stripe, and OCaml doesn't have a Stripe library. Before AI you'd sigh and write it. Now? Run a loop to port the Go library to OCaml. Here you go: Stripe in a unikernel.

Justin had a great example of the same thing. He'd been building minimal Linux OS images for appliances, which is halfway to a unikernel anyway, since you're only running one application as PID 1. He needed to make an XFS filesystem. Rather than drag in xfsprogs and everything it brings with it, he sat down with an agent and had it write mkfs.xfs in Rust, producing byte-for-byte identical output, with every flag interpreted. It reverse-engineered the on-disk formats one by one, with tests across block sizes. It took a few hours.

That works because the original tool is a golden oracle. Generate filesystems at different sizes with both implementations and diff them. Port the tests across. Automate it.

porting software has been trivial for a while now. here's how you do it.
If you have an oracle, porting is a loop.
Geoffrey HuntleyGeoffrey Huntley

Storage was the other big gap. Most workloads these days are cloud-shaped, even on-prem, so take the turbopuffer approach: S3 as your primary, infinitely growable storage, with a local NVMe block cache and an LRU (or whatever caching algorithm you like) for the hot bits. Justin is a massive "S3 for everything" fan too. As long as latency isn't the constraint, you get infinite storage with multi-user access, and you can build everything on it.

nix machine tests and overlays

Justin had been experimenting with Nix too, and was surprised that the first time he got an agent to prototype an OS, it built all the tests into flakes.

Nix the language sucks. Nixpkgs is great. NixOS machine tests are the bee's knees: you write a test that spins up a fleet of machines and exercises the interaction between your network rules and your application. It's the thing people don't know about.

And when something upstream is broken, or there's a supply chain problem in your dependencies, that's just an overlay.

the world hasn't figured out yet that you can literally just fix everything with a Nix overlay
Patch the world.
Geoffrey HuntleyGeoffrey Huntley
If you have to tool-call a human, also known as "Dear Maintainer", who might be on holiday or might have abandoned the project, and wait a day, two days, or even five minutes, that's not AGI.

We're building recursive products here. Agents need the ability to modify the world as first-party source, not as third-party bundled binaries. We're going back to the contrib folder and Unix patches.

if you care about security, you have two choices

Justin asked what unikernels still need for people to discover them. Honestly, it's this. We've raised two generations of developers who don't even know they're a thing.

If you deeply care about security, there are really two choices:

  1. You're sending satellites into space, and you should probably use seL4, a formally verified operating system. (I'm still a bit salty about the Australian government disbanding that team.)
  2. Everyone else should seriously consider unikernels. Stop trying to harden something that is very hackable. Invert it and design from the other direction.

The other classic criticism is that many early unikernel designs ran everything at a single privilege level: your application in the same ring as the OS. In 2026, that's a prompt away from being fixed if you want ring separation. It's certainly more secure than praying to god your systemd cgroup configuration is right.

Think about how much time enterprises spend patching the world every time something new drops in Linux. Upstream now expects you to patch your kernel weekly. The week we recorded, someone popped KVM (essentially Firecracker, the core primitive we all thought was good sandboxing) and collected $50,000 from Vercel and a few other vendors. That is not much money for something that could root every managed cloud provider in the world.

spaceleans: a distributed unikernel operating system

About seven months ago, I went deep on unikernels to check whether my mental model was right. I showed Justin my Mirage folder, which holds all the functionality I needed to add.

  • There was no way for a unikernel fleet to keep time, so I took an NTP client from another language and ported it. Then I built an NTP server based on RADclock and borrowed ideas from how TigerBeetle handles time: not one clock source but many, packaged as a library.
  • Network stack, DNS, HTTP clients and servers, structured logging, OTel, Anthropic and OpenAI clients, and payments via Airwallex.
  • A generic retry library for handling back pressure over HTTP, plus some PPX metaprogramming for fun.
  • A PII wrapper at the logging boundary, so secrets and PII never leak through the logging subsystem. Every project should have one. It's pluggable; just use a functor.

And then the most cooked thing, which I'd never shown anyone before: Spaceleans, Microsoft Orleans ported to OCaml, running as a unikernel.

Orleans is a distributed actor system with transactions. You take many physical machines and merge them into one addressable heap. An actor always exists: await GetCustomer(), and if it isn't in memory, it gets rehydrated from a pluggable storage provider. You collapse your n-tier architecture into actors and stop caring whether something lives on machine A, B, C or D. The runtime handles it as an infrastructure primitive.

So in a weird sense, I built a distributed unikernel operating system out of actors, with a filesystem on top. Justin called it Erlang-esque, and he's right. I did all of it in a week. I'll probably never release it, but it falsified the idea that unikernels are hard.

sampling history

Being a little older means you can sample history, like an experienced DJ such as Carl Cox, who's been in the scene long enough to pull from previous repertoire and bring it forward. All of these ideas existed in the eighties. The models have read the papers. They've got TAPL and the most advanced type theory in their training data.

The thing that's lacking is people's curiosity and ambition to do these unhinged things, and the knowledge that previous records exist that can be sampled from.

How do we get people to try this stuff? We just do it. If you've got a turbo Lamborghini alien space rocket that's more efficient and more secure, good for you; you've got a leg up. Do cool things, attract curious newcomers, mentor them, grow. Same as it's always been.

Meanwhile, everyone else will be trying to Chainguard their Ruby on Rails application and managing AWS with fifty AWS-certified engineers, when two people with Nix and Hetzner would do. Eventually, it comes down to money. Higher-powered tools are more efficient, and efficiency wins, especially as AI collapses margins.

ocaml, rust, haskell and back pressure

Has OCaml's time come again? It's still going strong. A certain trading firm is using it very well. When I caught up with Yaron at the start of the year, I asked him whether OxCaml exists so their language extensions end up in the training data and lift the whole company. I got a very "no comment" smile.

For agents, OCaml is lovely. Functors between modules are beautiful. The .mli files, a typed header explaining how a module should work, are really efficient context for agents. opam and Dune are legitimately good. Hindley–Milner. And compile times are fast. I see no reason to do F# these days.

Justin has mostly been writing Rust, and agents are good at Rust. But compile time is the tax on back pressure. LLMs hallucinate, and when compilation is slow, each hallucination is expensive because you get fewer attempts per minute. Justin's S3 clone is about a million lines of Rust; with four agents compiling at once, they fight over disk and CPU. You end up spending more on fast machines than on tokens.

Haskell's type system is great, and the models do it really well. But I don't feel good running it in production: a space leak lives in the runtime state space and only shows up in production. Justin pointed out that the linear-type ideas in Rust came out of Haskell papers trying to solve exactly that. Then there's Zig's approach: allocate everything up front and never allocate again, which is what game devs did in the eighties and nineties. It's hard to persuade an agent to do that in Rust, though, because constant-memory programs aren't in its training set.

I think dependent types are the winner for next-generation languages. Anything that lets you codify more into the type system is more back pressure. You probably won't be surprised to hear I've got a fork of the Rust toolchain with dependent types. You can just do things now.

languages for agents, and what cursed taught me

The pace of language development has been held back by how fast humans can learn new concepts. Operator chaining is essentially sugar for humans. If agents write the code, we can lean on forty years of academic PLT research, as long as you know how to sample it.

The industry codified "do not make breaking changes" after Python 2 to 3. Justin knew companies with hundreds of people on that migration for years. I think that rule is no longer true. Ship a skill pack with the breaking change and let agents auto-migrate.

Justin asked what it actually costs to make a new language successful now. Go was the last language a company spent real money on, and it took a long time. I can answer that one.

i ran Claude in a loop for three months, and it created a genz programming language called cursed
It's the only compiled language that lets you code with sus, slay and vibes.
Geoffrey HuntleyGeoffrey Huntley

Cursed was built with Sonnet 3.5 and 3.7, a deliberately underspecified prompt, and three months of running it in a loop. I started in C (not enough back pressure; I wasted too much time in Valgrind as the agent clobbered its own updates), then Rust, then Zig. Zig was a mistake; it would work today if I'd stayed with Rust. It cost roughly US$6,000, and I did it three times over. Compare that to what Go cost.

Now the real bit, the part that still scares me. If you allocate the context window correctly — a lookup table of the lexical structure and grammar — the model can program in a language that isn't in its weights. It's brute force and inefficient, but it works.

Think T-diagrams (tombstone diagrams). Lock down your grammar and lexical structure, reach a stage-two self-hosting compiler, ship a sensible standard library, and start the next training run. From there, you can reach a Roslyn-style self-hosted compiler with language services stupidly fast. Justin asked whether fine-tuning an open model would help bootstrap a language like this. It's not needed.

That was true a year and a half ago with much weaker models. It'll take just one programming language designer going all in with the good models to shock the world.

what next

As the pub was shutting, Justin asked what was on my mind. If you haven't read my latest post, go read it. If you manage people, create the space and time for them to experiment now, because within six months, leadership will ask you to put people on a vitality curve.

if your team is too busy doing their 'normal job' to experiment with AI, you're preparing them to be replaced
AI use is now mandatory for employability.
Geoffrey HuntleyGeoffrey Huntley

Yes, the labs trained on the commons. I hate that, and I get it. But you trade time and skill for money; employers have minimum standards, and those standards have changed faster than ever before in our industry. Be curious, learn how to build an agent, and go create beautiful stuff. We're in a renaissance.

It's a time-compression device. The more experience you have, the more you can sample. Not everything ships; some of what I showed Justin may never see the light of day. I use these projects as katas and redo them when the models get better.

But if you want to build something secure, seriously consider unikernels.

chapters

  • 0:21 — discovering unikernels: Haskell to OCaml to Mirage
  • 0:55 — what a unikernel is, and why it was hard
  • 2:45 — hard was past tense
  • 3:33 — why do we have an operating system?
  • 4:17 — the shell is a butler service for exfiltration
  • 5:12 — attack surface reduction is fuzzy
  • 7:26 — don't put the compiler on production
  • 9:13 — porting Stripe into a unikernel
  • 9:54 — minimal Linux and mkfs.xfs in Rust
  • 12:22 — S3 for everything
  • 13:48 — Nix, machine tests and overlays
  • 15:28 — tool-calling a human is not AGI
  • 15:53 — seL4 or unikernels
  • 16:56 — privilege rings
  • 18:24 — weekly kernel patches and the KVM escape
  • 19:38 — demo: the Mirage folder, NTP and time
  • 22:05 — Spaceleans: Orleans in OCaml as a unikernel
  • 25:31 — sampling history
  • 26:52 — how to get people to try this: just do it
  • 28:57 — OCaml and OxCaml
  • 31:45 — Rust compile times and back pressure
  • 33:06 — Haskell, space leaks and dependent types
  • 35:13 — memory strategies: Rust, Zig and game devs
  • 36:10 — languages designed for agents
  • 37:41 — breaking changes and skill packs
  • 38:53 — Cursed
  • 42:36 — what Cursed taught me
  • 46:47 — what's next: AI use is mandatory

Keep curious.

Read the whole story
bogorad
7 hours ago
reply
Barcelona, Catalonia, Spain
Share this story
Delete

Book Publishers Are Quietly Using More AI. Staff Are Revolting | WIRED

1 Share

LLM (google/gemini-3.5-flash-lite) summary:

  • Public Stance: american book publishers publicly fight generative ai through lawsuits and book cancellations while secretly adopting it internally.
  • Internal Adoption: major publishing houses including harpercollins simon and schuster and hachette incorporate large language models into daily operations without public disclosure or author consent.
  • Task Automation: staff utilize ai tools such as claude chatgpt and jasper to email agents write publicity copy design cover art and generate marketing materials.
  • Hidden Cancellations: multiple unpublicized book cancellations occur across various publishing imprints due to covert authorial use of large language models.
  • Corporate Mandates: harpercollins leadership purchases software licenses and pressures junior and senior employees to discover internal use cases through mandatory brainstorming sessions.
  • Ethical Dismissal: company leaders dismiss employee concerns regarding the legal ethical and environmental impacts of artificial intelligence as a standard cost of doing business.
  • Commie Worker Grievances: disgruntled employees complain that corporate funds are spent on software licenses instead of increasing low entry level salaries, highlighting traditional labor versus capital friction.
  • Primary Application: generating publicity copy and marketing materials emerges as the most common operational use case for artificial intelligence across the major publishing houses.

headline this year, American book publishers have appeared to be fighting back against generative AI, from suing Google for copyright infringement to canceling books by authors suspected of using LLMs.

But behind the scenes, at least three of the Big Five US book publishers—HarperCollins, Simon & Schuster, and Hachette—have been quietly incorporating AI tools into their publishing processes without public disclosures or author consent.

According to interviews with more than two dozen workers, all of whom spoke under the condition of anonymity to avoid jeopardizing their employment, publishers are now using LLMs like Claude and ChatGPT to email literary agents, write publicity and back-cover copy, design cover art, and generate marketing materials, often at the direction of their executives.

At the same time, multiple editorial staffers say the widely publicized cases of authors accused of using LLMs to help write their books—like Hachette’s horror novel Shy Girl or Macmillan’s crime thriller Call Me, I’ll Hide the Body—are just the tip of the iceberg. “Every single imprint has a story of someone very esteemed who had a book canceled because of blatant AI use” without ever making the news, one editor says.

At HarperCollins senior leaders purchased Claude licenses from Anthropic months ago “but very quickly realized they didn’t know how to implement them,” according to a current employee. Dozens of junior- and senior-level staff were then “voluntold” to become “AI Champions” and were tasked with discovering use cases in monthly brainstorming sessions with staff across the company.

This spring, when some HarperCollins employees expressed concerns about the legal, ethical, and environmental impacts of AI at one of these brainstorming sessions, a person leading the meeting dismissed their qualms as “the cost of doing business,” according to someone who attended the meeting. Soon after, a statement was added to the new AI section of the HarperCollins online employee portal, noting that “while AI does have an environmental impact, it’s a small part of most individuals’ total digital carbon footprint.”

Some HarperCollins staffers were unnerved by the brainstorming sessions. “It was such a slap in the face,” says one employee. “We’ve been making valid business cases for years to be paid a healthy salary, and instead they spent that money on software that most of us don’t really want.” (The average entry-level salary for New York City publishing employees at the Big Five and Scholastic was $47,583 in 2023.) In addition to Claude, staffers say HarperCollins has also purchased licenses for ChatGPT and Jasper, an agentic platform that promises “to run end-to-end marketing workflows.”

At least one HarperCollins division has been “all in” on AI since gaining software licenses, according to one staffer, including AI-generated marketing videos. But according to workers at HarperCollins, Simon & Schuster, and Hachette, one of the most common use cases for LLMs so far is generating publicity copy.

Read the whole story
bogorad
7 hours ago
reply
Barcelona, Catalonia, Spain
Share this story
Delete

The U.S. Army’s Desert Tech Test Shows Long Road to AI Warfare - WSJ

1 Share

LLM (google/gemini-3.5-flash-lite) summary:

  • desert testing: army units tested drones and communication systems during summer training exercises at fort irwin in extreme heat
  • equipment failures: high temperatures and strong winds caused servers to overheat, batteries to drain quickly, and drones to lose signals or land prematurely
  • modernization push: defense department leadership directed the military to accelerate the adoption of artificial intelligence and off the shelf startup technology
  • operational examples: unmanned boats and low cost offensive drones were deployed in ongoing conflicts to rescue personnel and execute strikes
  • bureaucratic hurdles: current and former officials noted that a risk averse culture and slow adaptation hindered the integration of autonomous systems
  • vendor evaluation: training events served to test the performance of newly procured hardware against manufacturers sales claims in real world conditions
  • communication upgrades: the service invested billions in command and control systems and satellite connectivity to improve battlefield coordination
  • human element: senior commanders emphasized that despite technological advancements, military operations remained fundamentally dependent on human decision making

By

Heather Somerville

| Photography by Ethan Swope for WSJ

Oct. 10, 2026 5:30 am ET

Under a blistering desert sun, the U.S. Army strained to show it had become a modern, AI-powered fighting force. 

Some blamed the heat. It was July in the Mojave Desert, and the thermometer climbed toward 110 degrees. Servers overheated. Batteries died and couldn’t be recharged. Soldiers drenched T-shirts in ice water and draped them over antennas that were on the fritz. 

Sgt. Ramontez Bennett’s drone from Anduril Industries was supposed to loiter overhead. Instead, the relentless glare forced it to land. “They gotta do something about it overheating,” he said. Plenty of America’s conflicts have played out in temperatures rivaling this particular July day. 

It wasn’t just the sun causing problems. “All it takes is one gust of wind and it’s on the side of a mountain,” he said. Behind the rugged peaks, drones lost their signal. 

Next to Bennett, another pilot prepared to launch a drone from Theiss UAV Solutions. After a prolonged wait to charge a drained battery, the aircraft took another 10 minutes to get airborne. Ukrainian soldiers can get a drone of about the same size off the ground in under a minute. 

Defense Secretary Pete Hegseth last year threw down the gauntlet with a directive to transform the Army into an AI-powered force equipped with drones and 3-D printers to quickly spin up weapons components, and directed the service to buy more software and startup tech. He instructed the Defense Department to stop looking for perfection in its tech, and buy off-the-shelf products that are good enough, stressing the urgency of modernizing. 

America’s investments in such weapons, some of which date to the Biden administration, have slowly begun to change the way it fights. In Iran, a war that has relied largely on older and expensive systems, the U.S. also used an unmanned boat made by defense startup Saronic to rescue a stranded helicopter crew. A drone modeled on Iran’s Shahed has been among the military’s most successful low-cost offensive weapons. The drone, called Lucas, costs $10,000 to $55,000 a unit thanks to an arrangement in which the Pentagon owns the design and farms out production to a group of manufacturers. 

Artificial-intelligence tools have also greatly accelerated intelligence analysis and target selection. The rapid evolution of the fighting in Ukraine shows how central such technologies will be to future wars. 

But America’s soldiers are still far from reaping the full benefits of cutting-edge autonomous weapons and AI systems, and haven’t been forced by necessity to quickly modernize as militaries like Ukraine’s have in the heat of conflict, current and former national-security officials said. Transforming a land-based fighting force with a history of slow technological adaptation into a nimble organization that makes the most of new innovations has been difficult.

A U.S. Army soldier runs into a building during an assault exercise.A soldier enters a building during an assault exercise at Fort Irwin.
A U.S. Army soldier runs into a building during an assault exercise.A soldier enters a building during an assault exercise at Fort Irwin.
Army soldiers run into position during combat training exercises. Soldiers run into position during a combat training exercise.
Army soldiers run into position during combat training exercises. Soldiers run into position during a combat training exercise.

Startups’ emerging new tech, most of which has never been to war, has at times proven to be too brittle in real-world applications, soldiers said. The Army is also in the throes of upgrading battlefield connectivity and communications, steps that are key to being able to fully capitalize on next-generation weaponry.

The military has held at least five press calls in recent weeks to talk about its modernization efforts. “There is a recognition that the battlefield is changing before our eyes,” Col. Gregory Merkl told reporters on one such call. “We are absolutely dedicated to transformation.”

Fort Irwin

Project Convergence brought 9,700 infantry soldiers to the desert for a 10-day, round-the-clock mock fight against an Army unit called Blackhorse, which stood in for China’s People’s Liberation Army. Media briefing materials billed the three-day July trip as a glimpse “inside the U.S. Army’s AI-enabled battlefield” and an opportunity for its leadership to demonstrate the use of AI apps, a data integration program, aerial and ground drones and sensor-loaded headsets in battle. 

In some of the moments meant to spotlight advanced technology on the battlefield, newly procured drones and developmental communication systems failed to perform as intended. 

Soldiers flew drones during limited windows and for half the maximum flight time to conserve battery and avoid losing them, and without full autonomy or any AI to assist. Chinese consumer drones could do the manual job a decade ago.

The battle offered the familiar scene of legacy warfare: green-suited soldiers kicking in doors, screaming above gunfire, weighed down with weapons from the last century. 

An Army spokesman for Project Convergence said the event was designed to allow for specific units to experiment with drones during pockets of fighting and learn about the tech, not to flood the airspace with autonomous weapons. The event was a way to “fact-check all the vendors that are coming to us” about the veracity of sales pitches from a crop of defense technology firms, said Alex Miller, chief technology officer to the Army’s chief of staff. 

Sunrise shining through smoke and dust over a mock village at Fort Irwin.Morning light filters through smoke and dust above the mock village at Fort Irwin.
Sunrise shining through smoke and dust over a mock village at Fort Irwin.Morning light filters through smoke and dust above the mock village at Fort Irwin.

The Army found during the exercise that the augmented-reality headsets referenced in its media invitation to Fort Irwin showed some deficiencies and the service is uncertain if it would move forward with them, a senior defense official at the training event said. Three companies have won contracts each worth more than $100 million to build the headsets.

While much of the tech there had already been awarded sizable Army contracts, officials and soldiers involved described it as experimental rather than essential new weaponry. 

“You take a piece of electronics and lay it out on a hot rock in 109 degree temperatures and try and make it work,” said Lt. Gen. Michael McCurry, one of the senior Army leaders running the event. “We’re seeing some of that.”

An Anduril spokesman said its drone overheated and that it had previously been tested in hot and cold temperatures. Following soldiers’ feedback, the company has prioritized making upgrades to improve the drone’s performance in extreme environments, he said. 

Theiss UAV Solutions founder Shawn Theiss said his drone typically launches within three minutes and that he received mostly positive feedback from soldiers at the exercise. Delays sometimes happen if soldiers aren’t trained, or if the communications parts, made by a separate company, have glitches, he said. The Theiss UAV drones there were more than three years old and batteries may not have been at peak performance, which can affect the launch, he said.

Some defense officials said the efforts would bear fruit longer term despite near-term stumbles.

Modernizing the military

Congress has long hammered the Army about modernizing its fighting. Congressional probes and government oversight reports have found that the Army has wasted billions of taxpayer dollars on fumbled new weapons programs, and suffered from inadequate training and failures in cybersecurity and data management. 

The Ukraine war brought the inadequacies into focus, bringing fully autonomous warfare to the battlefield with a tech proficiency that at times humbled the U.S. During battlefield drills held in Germany earlier this year, Ukraine’s drone operators took out a U.S. armored brigade. 

“The tech is moving so quickly that the U.S. military bureaucracy cannot keep up,” said Jonathan Rue, a former deputy assistant secretary of defense who played a key role in getting new technology into the Army, and is now a partner at firm MVP Ventures.

Sgt. Scott Santos launches a drone.A soldier launches a drone.
Sgt. Scott Santos launches a drone.A soldier launches a drone.
Sgt. Scott Santos coordinates with another soldier while operating a drone controller.One soldier coordinates with another while operating a drone controller.
Sgt. Scott Santos coordinates with another soldier while operating a drone controller.One soldier coordinates with another while operating a drone controller.

The last major Army transformation came about in the 1980s, when the service pivoted from Vietnam and built the force it would use in the Gulf War. It retrained, reorganized and brought in armored fighting vehicles, attack helicopters and modern air defense systems. 

The struggle now is to transform what recently ousted Army Sec. Dan Driscoll has called a “calcified bureaucracy” in preparation for a technologically sophisticated adversary—China. Driscoll departed in September, following disagreements with Hegseth. 

Daniel Goure, a former military analyst with the Lexington Institute who has held numerous war college teaching positions and roles in the Defense Department, said the Army’s history of modernization efforts are “almost unblemished by success.” He blames a tradition of risk aversion and a hierarchical organization.

Still, upgrades like the Army’s Next-Generation Command and Control system, used for AI-enabled planning and logistics, are a start. The system, with a $4 billion price tag in this year’s budget, allows troops to move faster and in a more dispersed fashion over longer stretches of battlefield, which is essential in modern conflicts, Army officials say. 

Starlink connectivity has meant that for the first time in his 19-year Army career, Lt. Col. Shawn Scott can radio a soldier sitting in an air-conditioned operations center at another end of the base, without cumbersome transmission poles. And new data-sharing technology has cut by two-thirds the time it takes Scott to plan and communicate missions.

The next step is bolstering the reliability of communications and data integration systems so AI-powered weaponry can actually help human fighters out in remote and austere conditions.

Capt. Alex Bayer reviews information on a tablet during training exercises. Capt. Alex Bayer reviews information on a tablet during a training exercise.
Capt. Alex Bayer reviews information on a tablet during training exercises. Capt. Alex Bayer reviews information on a tablet during a training exercise.
Soldiers maintain a defensive position at Fort Irwin.Soldiers maintain a defensive position at Fort Irwin.
Soldiers maintain a defensive position at Fort Irwin.Soldiers maintain a defensive position at Fort Irwin.

The Fort Irwin fighting climaxed with a predawn attack on a fabricated city, Razish, a relic of training for the wars in Iraq and Afghanistan. Stryker armored combat vehicles rolled down the hills. Smoke trucks attempted to obfuscate the view. Mortar rounds exploded and soldiers moved from one building to the next, firing blank bullets and pausing for the occasional smoke break.

A swarm of five small drones was forced to turn back as three Black Hawk helicopters carrying senior Army officials flew into the same airspace, squandering a rare high-tech moment they had come to see.

“It’d be a very dangerous time for us just to hand over everything to technology,” said Brig. Gen. Daniel Hibner, commander of the Army’s Joint Modernization Command. “It’s very much still a human endeavor.”

A soldier looks out a window at a smoky, desert landscape.A soldier looks out at the terrain during combat training.

Copyright ©2026 Dow Jones & Company, Inc. All Rights Reserved. 87990cbe856818d5eddac44c7b1cdeb8

Heather Somerville is a reporter at The Wall Street Journal in San Francisco covering technology and national security. Her articles explore the national-security implications of emerging technology, U.S. efforts to counter China's rise as a technology power, and the relationship between Silicon Valley and the U.S. defense complex.

Heather joined the Journal in 2019 to cover venture capital and technology companies. Before that, she wrote about venture capital and Silicon Valley startups for Reuters and the Mercury News. She was previously a reporter for the Fresno Bee and the Charlotte Observer and wrote about national security for outlets in Washington, D.C.

Read the whole story
bogorad
12 hours ago
reply
Barcelona, Catalonia, Spain
Share this story
Delete

Lobbying

1 Share

Have you tried to explain lobbying to a non American unfamiliar with the concept?

You see, the corporations and rich people write checks to the politicians. No no, not to them directly, sorry, to their reelection campaigns. Then the politicians listen to what the people who paid them have to say and pass laws based on that.

No but like, the money isn’t for the laws, it’s for like, TV ads and fancy haircuts and private jets and stuff. It’s not a bribe. It’s for their Super PAC. All above board. It’s like, they do what the people with the money want, they get the money, and they continue to have a successful career in politics.

No no, it’s not bribery. Bribery is super illegal in America. It’s like, the politicians need help knowing what laws to pass. So rich people come and help them. And they also give them money. How much money? Oh, that depends how helpful they are.

And if they were really helpful, they can leave the politics and get a nice consulting job with the people who lobbied them after. Then the money goes right in their pocket. No no no that can’t be a bribe you see, that happens after the laws were passed, a bribe would have to be before.

You just don’t get it cause you aren’t American, lobbying is definitely not bribery.

Read the whole story
bogorad
12 hours ago
reply
Barcelona, Catalonia, Spain
Share this story
Delete

The Other Anthropic Founder Trying to Fix the Company’s ‘Woke’ Reputation - WSJ

1 Share

LLM (google/gemini-3.5-flash-lite) summary:

  • Corporate Diplomacy: tom brown serves as anthropic's chief problem solver and diplomat, handling critical business and political alliances to support a planned initial public offering valued north of two trillion dollars (flagged: classic bourgeois maneuvers to consolidate capital).
  • Political Alignment: as a republican, brown successfully repaired relationships with the trump administration and elon musk, easing tensions over security concerns and past regulatory disputes.
  • Compute Procurement: brown initially focused on securing the chips and computing power needed to train models, orchestrating massive agreements with google, amazon, advanced micro devices, and spacex.
  • White House Engagement: by addressing government security fears and upgrading safeguards, brown ended a model access standoff and helped secure high-level meetings between dario amodei and the president.
  • Infrastructure Expansion: brown strongly supports the aggressive build-out of data centers and nuclear energy projects, aligning with white house priorities to ensure the continuous computing power required for advancing models.
  • Early Career: growing up in the san francisco bay area and studying engineering at mit, brown worked through various startups and openai before co-founding anthropic in 2021.
  • Military Friction: despite brown's diplomatic efforts, anthropic continues to face restrictions and disagreements with the pentagon regarding the use of its ai models by the military.
  • Future Outlook: brown anticipates major technological progress in curing cancer and boosting the economy over the next five years, projecting that current growth rates are largely underestimated.


Tom Brown, cofounder of Anthropic, exiting a black car. ZUMA Press

Call him Anthropic’s chief problem solver.

Tom Brown, who was one of the original six to found Anthropic with Dario Amodei, is parlaying his background negotiating deals for the chips needed to train AI models into something even more valuable: the business and political alliances Anthropic needs to pull off an initial public offering that could value it north of $2 trillion.

Where Amodei’s doomy warnings of AI’s risks have frustrated some administration officials, who worry that any pause or excessive regulation threatens to scuttle America’s lead against China, Brown has found common ground. He made peace with Elon Musk, a rival who had once called Anthropic “smug, sanctimonious and hypocritical.”  

When the White House forced Anthropic to shut off all access to two of its models due to security concerns, Amodei dispatched his chief compute officer to negotiate with the Trump administration after his own efforts fell short. Brown, a Republican, ended the two-and-a-half-week June standoff by assuring Commerce Department, Pentagon and White House officials that Anthropic understood their security fears and had upgraded its safeguards. 

The 6-foot-6 executive posed for a photo last week at a White House summit feting America’s advances in AI, posing shoulder to shoulder with technology CEOs and a president that has previously lambasted Anthropic as “woke.” 

His friendlier relationship with the administration was seen by some observers as one factor helping Amodei secure his first lengthy face-to-face meeting with Trump, a late September dinner at the White House after which he earned praise from the president. Amodei also attended last week’s lunch, making Anthropic one of two companies with a pair of executives in attendance. 

Tom Brown, cofounder of Anthropic, arrives at the White House for a lunch with technology leaders and President Donald Trump.Brown at the White House for a lunch with technology leaders and President Trump. Andrew Leyden/ZUMA Press
Tom Brown, cofounder of Anthropic, arrives at the White House for a lunch with technology leaders and President Donald Trump.Brown at the White House for a lunch with technology leaders and President Trump. Andrew Leyden/ZUMA Press

Brown bought a house in Washington earlier this year with his wife, a startup co-founder and former venture investor named Michelle Valentine. The couple has said in private conversations that they are allies of the Republican party and want to establish relationships in Washington, people familiar with the matter said.

Brown has supported the aggressive U.S. build-out of data centers and nuclear energy projects, a White House priority but an undertaking increasingly unpopular among Americans. Anthropic continues to need more computing power as it amasses users and models advance.

Brown said at a G-20 tech summit last month that he “really loved” a post by Trump on Truth Social that said communities that don’t embrace data centers risk becoming “backwards and poor.”

“He was pointing out that the data centers are just an enormous source of prosperity,” Brown said of the post.

“He is increasingly playing the dual role of founder and diplomat,” said Joseph Hoefer, chief AI officer at lobbying firm Monument Advocacy, which represents tech companies but not Anthropic.

His intense negotiating style masks his soft side as a father to a one-year old son, people who work with Brown say. When a colleague introduced him to their three-year-old, Brown put the boy on his shoulders and walked him around the room. 

The ‘awkward kid’

Brown grew up in the San Francisco Bay Area, then studied engineering at the Massachusetts Institute of Technology. He worked for a host of startups and participated in startup incubator Y Combinator. A self-described “awkward kid,” Brown has said he briefly worked on a group dating app called Grouper to help people like him “talk to girls.”

Brown began learning about AI through online courses and math textbooks after reading “Superintelligence” by Nick Bostrom, a popular book in AI circles exploring what it says are existential dangers for humans from the creation of machines that are smarter than humans.

He offered to mop the floors at OpenAI to get his foot in the door. Shortly afterward, OpenAI co-founder Greg Brockman, with whom he had become friendly, hired Brown as one of the startup’s first 20 employees.

There, Brown met Musk, the lab’s main funder at the time. During one of Musk’s visits in 2016, Brown watched in panic as Musk savaged a demonstration from a legendary researcher. But when it was Brown’s turn to show off his work, Musk said he liked Brown’s tool that let an AI play the videogame StarCraft, according to people familiar with the matter. 

The encounter would end up paving the way to a pivotal computing deal between the two men’s companies a decade later.

Brown was laid off from OpenAI, then briefly joined Google’s DeepMind lab. Amodei brought him back to OpenAI in 2018, and he led engineering work on GPT-3. The breakthrough model demonstrated one of the central tenets underpinning the AI boom: that capabilities would increase when models were trained on more data. 

He joined Amodei and five other OpenAI leaders to create Anthropic at the start of 2021.

Brown’s main job in the early days of the company was securing the chips and computing power needed to train and run Anthropic’s models. While OpenAI had primarily relied on Nvidia chips through its partnership with Microsoft, Brown and Anthropic’s co-founders sought to work with other companies to make the company less dependent on a single supplier.

Anthropic Co-Founder Tom Brown, Elon Musk, and Nvidia CEO Jensen Huang stand on stage at the launch of America.gov.Brown with Elon Musk and Jensen Huang during a launch event for America.gov. Kylie Cooper/Reuters

When the early 2023 launch of Anthropic’s Claude models to compete with ChatGPT increased the need for computing power, Brown orchestrated deals with Google and Amazon.

Winning over Elon

As Anthropic leapt ahead in the AI race and its models gained traction with businesses, the company’s computing capacity wasn’t able to keep up and customers faced usage limits and outages.

Brown turned to Musk, whose SpaceX had excess capacity at its massive data center complexes after acquiring xAI. It was a tender time: Musk had spent months railing against Anthropic after it cut off xAI’s access to Claude Code, calling it “smug, sanctimonious and hypocritical” on X and giving it the nickname “Misanthropic.” 

But Brown saw that Musk and Anthropic ultimately had an opportunity to help each other. He drove to SpaceX’s offices in Hawthorne, Calif., in late March and told Musk that Anthropic wasn’t as “woke” as he thought, while acknowledging that some of its past models had been, according to people familiar with the matter. 

In early May, Anthropic and SpaceX announced that Anthropic would rent SpaceX’s compute for $1.25 billion a month. To keep up with competitors, Anthropic has also signed agreements worth tens of billions of dollars with Advanced Micro Devices and a trio of smaller companies in recent weeks.

Amodei sent Brown to put that diplomatic dexterity to work in Washington. 

The company clashed with the Defense Department earlier this year about the right guardrails for AI models when they are used by the military. The Pentagon’s designation of Anthropic as a security risk is still in place. 

Brown has regular conversations with Emil Michael, undersecretary of defense for research and engineering, but the Pentagon has barred all usage of Anthropic’s tools and told contractors to stop using them, a move that has been upheld in one court and struck down in another.  

Brown taking charge of the June model shutdown in Washington allowed Amodei to focus on customer relationships and other CEO responsibilities, a person familiar with the matter said. During that time, Brown and Amodei spoke daily on the phone, the person said. 

In recent weeks Brown has held regular discussions with Commerce Secretary Howard Lutnick and Michael Kratsios, head of the White House Office of Science and Technology Policy, about Anthropic’s models and AI safety. 

During last month’s G-20, Brown was seen talking casually with Lutnick and officials from other countries at the Carolina Inn on the University of North Carolina’s campus, with some in the group smoking cigars. 

On stage at the event, Lutnick asked Brown what the future holds for AI. Brown said he expects major progress on curing cancer and benefiting the economy in the next five years, echoing claims from Amodei.  

“Looking back we’ll be like, ‘Wow, back in 2026 we thought stuff was going fast and we thought we were seeing this, but actually we underestimated how impactful this would be and what the progress would be,’” Brown said. 

Copyright ©2026 Dow Jones & Company, Inc. All Rights Reserved. 87990cbe856818d5eddac44c7b1cdeb8

Amrith Ramkumar is a reporter for The Wall Street Journal in Washington covering tech and crypto policy. He previously covered clean energy and was a Journal markets reporter in New York who wrote about special-purpose acquisition companies, or SPACs, when SPAC mergers were a popular alternative to traditional initial public offerings. He also previously wrote about stocks and commodities, including battery metals such as lithium and cobalt.

Amrith joined the Journal as a markets intern after graduating from Duke in 2017.

Keach Hagey is a reporter at The Wall Street Journal covering the intersection of media, technology and power. Her reporting explores how institutions and individuals wield influence in the new information economy, with a current focus on artificial intelligence and OpenAI. She is the author of "The Optimist: Sam Altman, OpenAI, and the Race to Invent the Future" (W. W. Norton, 2024) and "The King of Content: Sumner Redstone’s Battle for Viacom, CBS and Everlasting Control of His Media Empire" (Harper Business, 2018).

She was part of the team that broke the Facebook Files, a series that won a George Polk Award for Business Reporting, a Gerald Loeb Award for Beat Reporting and a Deadline Award for public service. Her investigation into the inner workings of Google’s advertising-technology business won recognition from the Society for Advancing Business Editing and Writing (Sabew).

Previously, she covered the television industry for the Journal, reporting on large media companies such as 21st Century Fox, Time Warner and Viacom. She led a team that won a Sabew award for its coverage of the power struggle inside Viacom.

Before joining the Journal, Keach covered media for Politico, the National in Abu Dhabi, CBS News and the Village Voice. She has a bachelor’s and a master’s in English literature from Stanford University. She lives in Irvington, N.Y.


Up Next


Videos

Read the whole story
bogorad
1 day ago
reply
Barcelona, Catalonia, Spain
Share this story
Delete
Next Page of Stories