An animal advocacy nonprofit is currently hiring staff to run a commercial broiler farm. Not to shut one down. To operate one. They will buy day-old chicks. They will own them for the six weeks a broiler lives. Then a catching crew will come at night, the birds will be crated and driven to a slaughterhouse, and the charity will invoice for them. Donor money will absorb the R&D failures a producer would never risk on its own dime.
Read that again as an economist would. A charity is paying down the adoption risk on a production technology so that the industry doesn’t have to.
Frontier Husbandry is not a rogue project. It is the clearest single artefact of a turn the central institutions of animal advocacy have made recently and fast, announced as a series of “good news” items rather than as the strategic reversal it is, and it did not come out of nowhere.
In 2025, Shrimp Welfare Project published its vision for the decade ahead, naming the rise of AI-automated, super-intensive aquaculture “Factory Farming 2.0” and warning that the window to shift its trajectory was open now and would not stay open. That is among the sharpest strategic calls this movement has made in a decade: correct, and years ahead of the field. Here is what the movement did with the warning: built the thing it warned about, faster than anyone else could have, with applause.
In eighteen months, animal advocacy has assembled a complete pipeline for developing, financing and selling technology that makes animal agriculture more profitable. A nonprofit venture studio spinning out for-profit companies on the thesis that the most durable welfare gains are the ones industry wants to buy. A venture fund run by the founder of The Humane League, whose second fund has already backed livestock monitoring software pitched to producers on mortality and labour costs. A think tank soliciting commercial partners to trial technology against feed conversion and growth rate. An incubator recruiting founders who need not be vegan, need no movement background, and need not “feel much when you see a chicken.” And an intellectual case to hold it together: the field’s most-discussed AI analysis, which rules out preventing AI from worsening factory farming while listing PLF-for-good (Precision Livestock Farming, the industry term for AI in factory farms and slaughterhouses) among the promising options.
None of this is hidden. It was all announced as good news, funded by respected institutions, built by people with real records of reducing suffering. That is what makes it dangerous. The theory behind it rests on one belief: intensification and entrenchment of factory farming is coming regardless of what we do, we can’t fight it, so we should try to optimise it instead.
That belief is false, and we have historical evidence that it is false. And because it’s false, this movement is about to spend the largest sum it has ever held making the system it exists to end more efficient, more capitalised, and more durable, while calling it pragmatism.
The people running this play are not the fringe. They are the central institutions of this movement: its funders, its incubators, the founders of its most prominent organisations. This is what that layer has chosen to do with decades of accumulated legitimacy and the biggest windfall the movement has ever seen: build the infrastructure of the thing we exist to end.
That windfall is arriving now, and it is arriving fast. On 28 May 2026, Anthropic raised $65 billion at a $965 billion valuation and filed a confidential draft S-1 four days later, reporting points to an autumn Nasdaq listing above a trillion. Ambitious Impact has already named the mechanism: part of the funding surge comes from “E2Gers [who] have come into a lot of money through their equity in AI-labs.” The EA Animal Welfare Fund is planning to deploy $20 million-plus annually, against roughly $6 million in 2025, and leaders are now openly worried about absorptive capacity: too much money, not enough places to put it.
Stray Dog Institute’s 2024 survey found that 36% of surveyed organisations are operating on under $100,000 a year. A movement in which most organisations run on almost nothing has just declared it has too much money. Both things are true at once, and they are only compatible under a particular definition of who counts as a recipient of the incoming funding. What counts as a legible recipient is a tech startup building Factory Farming 2.0, not an animal activist campaigning to end it. “Absorptive capacity” is not a fact about the world. It is a fact about which recipients a specific and narrow worldview can see.
There is a single question capable of sorting good spending from bad in this environment, and almost nobody in the field is currently asking it:
Does this intervention make the industries that exploit animals more durable or less durable, independent of whether it reduces suffering inside it?
Call it the durability test. Everything below is an application of it.
“Unavoidable”
In September 2025, Animal Charity Evaluators interviewed Aaron Boddy, co-founder of Shrimp Welfare Project. On Precision Livestock Farming (PLF), he wrote: “AI is going to enable intensification. Unfortunately, this just seems unavoidable.”
Then, immediately, the turn: he thinks the same technology could deliver a level of individualised monitoring that actually improves shrimp welfare, “if we engage with this in the right way.”
Two sentences concede the trajectory. The third proposes to ride it. The hinge between them is a single word. Sit with that word. Unavoidable.
“Unavoidable” is not a finding. It is not an observation about the world in the way that “intensification is coming” is an observation about the world. It’s a claim about power: about which forces can be redirected and which cannot. And it is currently being stated in the register of a weather report while hundreds of millions of dollars move behind it.
Aaron’s organisation is the one that named Factory Farming 2.0 and called the window on it before anyone else did. The person who saw this coming most clearly is the same person who concluded it could not be fought.
It is worth being concrete about what he is right about and what he is wrong about. In Ezhou, in Hubei province, a pair of 26-storey buildings turn out 1.2 million pigs a year, holding around 650,000 animals at a time, fed through more than 30,000 automated points, moved between floors by six lifts carrying forty pigs each, with a ground-floor control room tracking temperature, humidity, ventilation and toxic gas across the whole tower in real time. It cost around $570 million, and it cannot operate for a day without continuous sensing. China’s smart-livestock market is projected at 47.7 billion yuan by 2026. That is the direction of travel and nobody should pretend otherwise. Intensification is coming. It will be the default if animal activists don’t do anything about it.
What’s in dispute is whether that trajectory is a fact to accommodate or a variable to fight over. It’s a variable, and treating it as a fact is exactly how a prediction becomes an outcome.
What has been built
The venture arm. Spring Innovation Fund launched 30 June 2026: a nonprofit venture studio backing for-profit companies, expecting to deploy an eight-figure sum in its first year or two. Its own copy: the most durable welfare gains come from “solutions the industry actually wants to adopt.” Spring acknowledged they “probably won’t resonate if you think only efforts to end or replace animal agriculture count as progress.” That sentence is not a position in the old fight over whether welfare reforms are worth doing at all. It’s simpler: is ending the industry still the goal, or not?
The capital. Lever VC’s $50 million fund has backed two livestock-monitoring companies, one described on Lever’s own site as lowering “mortality and disease on farm.” Its managing partner is the founder of The Humane League, former EVP of Mercy For Animals, co-founder of the Good Food Institute. This isn’t outside capital funding investments in animal agriculture. It’s movement veterans leading the funding rounds themselves.
The incubator. Ambitious Impact’s first incubation round focused exclusively on animal welfare states clearly that “the charities that are likely to come out of this cohort aren’t likely to look like activism or corporate campaigning – they’ll be more likely to look like reshaping the incentives that industry responds to, bringing new technology to market”. They also say you don’t need to be vegan, don’t need to be an “animal person”, don’t need an animal-movement background and don’t need to “feel a strong emotional pull toward animals”.
The farm. Innovate Animal Ag, which calls itself “the DARPA of animal agriculture,” is seeking commercial partners to trial a technology against feed conversion ratio and growth rate: the headline numbers of broiler production economics, not welfare indicators. Frontier Husbandry, described above, is its proving ground, built to run the last 30% of R&D “a commercial P&L rarely allows.” High capital and switching costs are the standard mechanism by which precision tech entrenches and consolidates industrial farming, and a nonprofit proving ground pays that adoption risk down with donations.
The intellectual cover. The field’s most-discussed AI analysis, A Shallow Review of What Transformative AI Means for Animal Welfare, prompted a forum-wide debate week. It marks “prevent AI from worsening factory farming” with a ❌ while marking PLF-for-good as promising, on the same page.
What precision livestock farming actually does
PLF is usually defended on the grounds that better measurement means better welfare. Start with what that looks like for one bird.
The parents of meat chickens carry the same genetics as their offspring, and those genetics are lethal to an adult. Fed to appetite, broiler breeders become obese and lame, lay badly, and die at rates reaching 31%. So the industry starves them. Females are held at roughly a quarter to a half of what they would voluntarily eat during rearing; around 30% of voluntary intake is typical, and restriction can run as deep as 80% below it. At twelve weeks, a bird allowed to eat weighs more than twice what a rationed bird weighs. What that produces is chronic hunger and feeding frustration, stereotypic object pecking, restlessness, elevated stress physiology, and it affects every parent bird of every broiler alive. Some producers then restrict water on top of it, because a starving hen drinks to fill herself and the wet litter burns her feet.
Precision feeding is the technological answer to this. The hen wears an RFID tag. She enters a feeding station, is weighed in real time, and an algorithm releases feed only if she is below her target weight trajectory. If two birds enter together, the station ejects them unfed. Every visit is written to a database. In one trial, nearly a third of birds failed to adapt to individual feeding; they were identified because no feeding events had registered against their tags, and some had not eaten in over 24 hours.
It is published as a welfare improvement, and the literature is candid about why it is needed: “feed restriction intensity continues to increase due to selection for efficient, fast growing” broilers. Read that as the engineering problem it is. The hen’s hunger is the binding constraint on how far growth-rate genetics can be pushed. Precision feeding does not relieve the constraint. It administers it individually, at higher resolution, so that the constraint stops binding and selection can continue. The suffering is not a side effect of the system. It is the variable the system was built to manage.
There is also a natural experiment on the measurement-means-welfare claim already sitting in European law, and it has been running for nineteen years.
Council Directive 2007/43/EC caps broiler stocking density at 33 kg/m². A producer who documents housing conditions and meets specified environmental requirements may go to 39 kg/m². A producer who additionally demonstrates low cumulative mortality across successive flocks may go to 42 kg/m².
That is a staircase, and every step is paid for in measurement. Monitor ammonia, CO₂, temperature, humidity and mortality closely enough and the law lets you put 27% more birds in the same shed. Meanwhile, EFSA’s 2023 opinion finds welfare harm begins above 11 kg/m², roughly a third of the baseline the derogations build upward from.
This is the mechanism, already codified, before AI enters the picture at all: sensing capacity is not banked as welfare improvement. It is spent as stocking density. The technology didn’t fail. It worked exactly as designed, and the gains were converted into cramming more animals into less space.
Nor is this an artefact of European law. A computer-vision system for shrimp, developed by the Alfred Wegener Institute with the producer Oceanloop, was described by its users as letting them “more precisely determine the density for each size group”, improving survival, growth and feed efficiency in the same breath. Automated shrimp counters are now sold to farms pushing the boundaries of very intensive production, on the argument that AI needs precise stocking-density data to work from, where super-intensive means upwards of 150 shrimp per square metre. Better measurement, spent as density, stated in plain commercial English.
The peer-reviewed literature says the same thing prospectively, and the most systematic statement of it is a 2022 review in Frontiers in Veterinary Science by Tuyttens, Molento and Benaissa: Twelve Threats of Precision Livestock Farming (PLF) for Animal Welfare. The authors are welfare scientists at agricultural research institutes and the paper is a taxonomy: twelve ways the technology can damage the animals it is sold to help, grouped into direct harm, harm through the farmer, harm through changes to housing and management, and harm to the moral status of animals in society. Three of its findings bear directly on the case for PLF-for-good.
On evidence. The welfare claims made for PLF remain largely unproven, because the systems that actually reach the market and get adopted are the ones that improve production efficiency and labour costs. Validation is thin even where it exists: external validation trials are available for roughly 14% of retailed PLF systems for dairy cow welfare assessment, and only around 23% of published PLF work in pigs was properly validated. The authors also name the competitive trap. The testing regime that would make these systems safe for animals raises their cost, so manufacturers who skip it gain an edge.
On housing. Many PLF systems have to separate animals from their background, which works better when the background is uniform, so the technology pushes toward barren, sterile environments. Economies of scale push the same way: per-animal costs fall as herds grow, herd size is a leading determinant of whether PLF gets adopted at all, and mortality and infectious disease generally rise with herd size. The small-scale outdoor systems associated with better welfare are precisely where this technology is least viable.
On the loop. The paper diagrams the mechanism as two self-reinforcing cycles. In the first, PLF makes intensification more viable, intensification hides more animals from public view and makes them easier to treat as units of production, and that reduced concern clears the way for further intensification, which expands the market for PLF. The second runs through price: PLF raises efficiency, efficiency makes animal products cheaper, cheaper products raise consumption, and higher consumption further lowers the moral status of the animals. What the authors say objectification buys the industry is “the social license to exploit even more animals in a more extreme way.”
That is a durability diagram, drawn by veterinary welfare scientists, published four years before this movement decided to fund the thing at the centre of it.
The evidence that trajectories bend
The claim that technology adoption is unstoppable is testable, and it has been tested. It fails. The useful part is not that it fails, but how, because the people who bent these curves were doing identifiable things, and almost none of them are being done here.
In 1998 a small Canadian outfit called RAFI found a patent held by Delta & Pine Land and the USDA covering seed engineered to be sterile at harvest. The technical name was Genetic Use Restriction Technology. RAFI called it Terminator. Eighteen months later Monsanto announced it would not commercialise the technology, citing the views of its own grower constituency, RAFI’s research director said plainly that what moved the company was not morality but the certainty that the product would never win public acceptance. No regulator stopped Terminator. The developer did, pre-emptively, on a product that did not yet exist. Then the campaign made it stick: the Convention on Biological Diversity adopted a de facto moratorium in 2000 and reaffirmed and strengthened it in 2006 against a push from Canada, Australia and New Zealand. Twenty-eight years on, a technology whose economics its owner found compelling has never been sold.
Four moves in one case: give the thing a name a non-specialist can hate, press the developer rather than the farmer, turn the developer’s customers from a shield into a liability, and write the win into an instrument that outlasts the news cycle. PLF has none of them. It has an industry acronym, a set of vendors nobody outside the sector could name, and a movement funding it rather than campaigning against it.
GM crops in Europe are usually told as a story about consumer fear. The more useful account is Schurman and Munro’s: European food retail was concentrated and built on own-brand quality positioning, which gave it a chokepoint the American commodity chain lacked. Activists did not have to beat the biotech industry, they had to beat the supermarkets, who beat the biotech industry for them. Iceland went GM-free in 1998, the sector followed within months once the position proved it sold, and when Unilever committed in April 1999 nearly every major food company followed inside a week. Only then did the political layer move, ending in Directive (EU) 2015/412 and the opt-outs most member states took.
Sometimes the producers ask for the ban themselves. Sweden prohibited antibiotic growth promoters in 1986, when they were in practically all Swedish pig and broiler feed; the Federation of Swedish Farmers had issued a voluntary restrictive-use policy in 1981 and then requested legislation, because the public argument was already lost and a universal ban protected farmers from each other. That is the structure PLF has. Its returns are relative-advantage returns: a producer who installs it gains on a producer who does not, so once the technology becomes a reputational cost, prohibition is the industry’s preferred outcome rather than its nightmare. Sweden then lobbied the same rule upward into the EU-wide ban.
Coalitions do not require agreement. At Wyhl in 1975, conservative winegrowers occupied a nuclear construction site because they feared cooling-tower steam would shade their vines. Students, clergy and the local hunting association joined them, thirty thousand people re-occupied the site after police cleared it, and a court withdrew the licence that March. The plant was never built. Nobody there needed to share an ideology, they needed to share a target.
None of these technologies was abolished globally, and the argument does not need them to have been. Comparable economies facing comparable industry pressure reached opposite outcomes: the EU hormone-beef ban held through a WTO dispute and years of retaliatory sanctions; rBST, approved in the US in 1993, was permanently prohibited in the EU and declined by Canada, Australia, New Zealand and Japan, and then lost its American market anyway, without a ban, when the processors and retailers walked and Monsanto sold the business off in 2008. If outcomes were determined by a technology’s economics, they would not diverge. They do diverge.
Three things separate the wins. They arrived while the technology was still a proposal rather than an installed base. They pressed the developer, not the user, farmers are immobile, multi-generational and sympathetic, while the firms building this software are small, young, dependent on a handful of investors, and staffed by people who did not take the job in order to make sheds denser. And they converted the win into something durable: a treaty, a directive, a court holding. When Google workers forced the company off Project Maven in 2018, the win went into a corporate values page, and Google deleted the pledge seven years later. Terminator went into a treaty and has held for twenty-eight.
The strongest case for the other side
One: welfare work inside the system has produced the movement’s largest measurable wins. Cage-free campaigns took fifteen years and improved conditions for hundreds of millions of hens. Refusing to work inside the system on purity grounds would have forfeited that.
Two: if intensification is coming regardless, presence beats absence. Better the technology is shaped by people who care about the animals than by people who don’t.
Three: this movement has never once stopped a production technology in animal agriculture. Battery cages, gestation crates, fast-growing genetics, live-shackle slaughter, all adopted over advocacy objections. The historical base rate for “we will bend this curve” is poor.
To the first: corporate campaigns worked by raising the cost of a practice. That is the opposite operation to PLF-for-good, which lowers the cost of production. Both get filed under “welfare work,” and that shared label is doing enormous unearned work. The durability test separates them cleanly, which is exactly why it is worth adopting. Cage-free imposed a cost the industry had to absorb. A subsidised R&D pipeline removes one.
To the second: you cannot credibly threaten an industry whose capital expenditure you are underwriting, and an organisation whose revenue model depends on selling to producers has a permanent structural reason not to campaign against them. That is not a prediction about anyone’s character. It is a description of what happens to organisations under that incentive, reliably, over a decade.
To the third, and this is the important one: the base rate is poor because in every previous case the movement arrived after the adoption curve, with no capital and no technical capacity. Battery cages were universal before there was a movement capable of contesting them. This is the first production technology in the history of animal agriculture that the movement has encountered before mass adoption, while holding both money and technical capability.
The choice
I have been working toward animal liberation for eighteen years. I have made every mistake this movement offers, burned out and come back. I founded Open Paws, which was defunded, and I was removed from it. That gives me reasons to read the funding layer of this movement uncharitably, and a reader should weigh that. It is also why I am asking you to check the documents rather than trust me: every claim above is sourced to something the institutions in question published themselves.
For many of those years I shared the pessimism that leads so many others in our movement to abandon our ultimate goal of animal liberation. I never personally abandoned that goal, but I did believe for a long time that I would never see it achieved within my lifetime.
Then we were handed a gift that completely changed the equation: something that could enable a small team of advocates, or even a solo operator, to run at the scale of a large organisation, and soon maybe even at the scale of an entire movement. That gift was, of course, AI. I’ve spent the last four years building and applying AI to the fight for animal liberation and I’ve seen firsthand how transformative it’s been to the advocates who use it to fight against the industries that exploit animals.
The turn described in this essay takes that same capability and points it the other way: not AI systems built for animal advocates, but animal advocates building AI systems for the industries we exist to oppose.
This is not unavoidable. Technology trajectories aren’t laws of physics. They’re the product of investment decisions and the work of people who showed up early enough to matter.
Shrimp Welfare Project was right: this is Factory Farming 2.0 and we’re at the beginning of it. And a few paragraphs below “unavoidable,” in that same interview, Boddy names the reason that matters, that it will be “very difficult to change course once certain paths are locked in.”
He is right about that too. It is the entire argument for acting now rather than adapting later, and it is sitting on the same page as the defeatist resignation. The beginning is the only part where the trajectory is cheap to change.
We are about to receive more money than this movement has ever had. What we build with it will encode our answer, whether or not anyone says it out loud. So say it out loud. The central institutions that control this movement’s money have made their choice, in public, with funding attached. It’s time the rest of us made ours.
Are we trying to end factory farming, or are we trying to optimise it?
Everything else follows from the answer.
Edit, 25 August 2026
Nate Crosser, co-founder and VP of Investments at Spring Innovation Fund, replied to this piece publicly. He is the first person from any of the named institutions to engage in the open, and that deserves credit before I disagree with him. His comment in full:
Thanks for your thoughtful analysis. You present several good arguments and effectively flag the need for caution around PLF. I believe you’ve presented a false dichotomy of optimizing v ending factory farming. What these people are trying to do is reduce animal suffering. One way to do that, among a kaleidoscope of other approaches others are taking, is to reduce animal suffering on farms. PLF is only one of many potential approaches to that goal, and probably the approach that is being least prioritized. Reducing pain and stress during slaughter, or developing alternatives to painful practices like pig castration, for example, are not forms of PLF. Additionally, yield is only one of several reasons a farmer might adopt a welfare beneficial practice. Others include worker safety, food safety, sustainability, or regulatory requirements. Again, thanks for your thoughtful critiques which are an important part of this process.
There is no dichotomy in the durability test. It does not ask whether an intervention is welfarist or abolitionist. It asks whether an intervention makes the industry more or less durable, independent of whether it reduces suffering. Every campaign that forced a producer to absorb a cost passes. The line runs between cost-imposing and cost-reducing, and it cuts through the middle of welfare work rather than around it. Nor does anything here turn on motive. I don’t doubt that the individuals involved are sincere in their desire to reduce suffering. The test is indifferent to intent because industry is: no producer prices the sincerity of a vendor.
The other adoption motives are the same argument in different clothes. Worker safety, food safety and sustainability reach the producer as lower injury, liability and input costs. The fourth item, regulatory requirement, is the only one that genuinely imposes cost, and it is the one a fund defined around technology that improves animal welfare in ways that make commercial sense for producers to adopt is structurally least able to pursue.
The questions I have put to Nate:
By what mechanism do the companies Spring incubates make factory farming less durable rather than more?
How will Spring ensure it doesn’t incubate companies that make factory farming more durable?
Nate has not answered these yet. If he replies, I will publish it here in full and unedited.
Thank you for this. Excellent work in a dismal situation
This is one of the most important critiques of animal philanthropy I’ve read in a long time, other than Carter Dillard’s work. #donorrevolt
The durability test gets to the question the movement should have been required to answer before another dollar was deployed: compared with what baseline?
If an intervention reduces one measurable form of suffering while simultaneously lowering industry costs, subsidizing R&D risk, increasing productive capacity, or making the underlying system harder to dismantle, you cannot simply count the welfare gain and discount everything that follows from it.
And “unavoidable” cannot be the baseline. That is an assumption about power masquerading as an empirical fact.
With extraordinary new AI wealth entering animal philanthropy, this matters now. Before we spend a historic windfall optimizing the machinery of exploitation, funders should have to demonstrate that their interventions reduce total harm rather than merely improve the metrics they have chosen to measure.
Gary is asking exactly the right upstream question: are we reducing exploitation, or financing its durability?