All podcasts / Invest Like the Best / Summary

Shyam Sankar - Celebrating Heretics

2026-03-10 - 82 min - source - Read full transcript

Patrick O'Shaughnessy (host)Shyam Sankar

Key insights

Heretics, not bureaucratic process, produce the innovations that win wars and build great companies, and they are usually punished before being celebrated.

Sankar cites Hyman Rickover (built the first nuclear submarine from an office that was literally the women's restroom, in a role the Navy tried to use to humiliate him), Andrew Higgins (built the landing boat that became 92% of all WWII boats without a Navy spec), and Billy Mitchell (court-martialed and died penniless before his ideas became the Air Force). His argument: 'the only shit that ever worked... were the things that the heretics actually did. Nothing that went through the machine delivered anything.'

heretics-and-founders

Palantir's forward-deployed engineering model swaps a weaker feedback loop (will someone buy this) for a stronger one (did this deliver the outcome), which can only be assessed on-site with the end user.

Sankar contrasts a 'software industrial complex' that validates products by whether they sell with sending engineers who are not sales staff but are obsessed with field impact and technically fluent across the whole stack, effectively building the product 'through back propagation' from the operator's actual problem.

palantir-operating-model

Palantir's core bet is that enterprise value comes from decisions, not data, and its 'ontology' layer models both an organization's data and its actions so the business becomes programmable.

The ontology solves what Sankar calls an impedance mismatch: data is stored in a way that suits transactional systems, not the way humans in an institution actually think about problems. He describes a whole business as 'a series of decisions,' and the ontology as an API layer that lets you compound applications on top of it.

palantir-operating-model

The US industrial base has narrowed catastrophically: dual-purpose companies used to fund 94% of major weapons R&D through commercial cross-subsidy; today defense-only specialists get 86% of that spending, and the prime contractor base shrank from 51 to 5 after a 1993 Pentagon dinner known as 'The Last Supper.'

Sankar names Chrysler (missiles and minivans), Ford (satellites until 1990), General Mills (torpedoes and inertial guidance), and Kodak (Corona spy-satellite film) as examples of companies whose commercial engineering subsidized national security. He argues the 1993 consolidation didn't just reduce competition, it drove out the 'crazy' engineers who went to tech instead, replacing real engineering with financial engineering.

reindustrialization

Losing manufacturing capability eventually costs a country its capacity to innovate, because production is often the input stimulus for breakthrough research.

Sankar's example: Google's 2017 'Attention Is All You Need' paper (the origin of the transformer architecture) grew out of a team trying to make an incremental 3% improvement to Google Translate. He extends this to argue China's shift from a 'cheap pair of hands' for pharma contract research to now producing 50% of drugs used in global clinical trials shows innovation following production, not the reverse.

reindustrialization

Sankar argues the US regained lost military deterrence over the past year through operations like Midnight Hammer and the raid connected to Maduro, because a single overwhelming, low-casualty strike also discredits the Chinese and Russian weapons systems adversaries rely on.

He frames deterrence as requiring the capacity to generate that kind of strike 'every single day until the enemy submits,' not a one-off demonstration, and notes the Maduro-related raid landed the same day Chinese officials were visiting Venezuela, a pointed signal that Chinese- and Russian-sourced weapons in-country didn't work.

us-china-rivalry

China's core strategic asymmetry is patient, systematic long-range planning; America's is genuine unpredictability, which China's central planning structurally cannot replicate.

Sankar says China has been mapping US military technology dependencies since 1991, after watching the Gulf War, and has broadcast its intent the entire time. By contrast, the AI shift was in neither country's long-range plan, yet the US economy 'pivoted on a dime,' something he argues China's own frontier models are, in his framing, still trailing via distillation.

us-china-rivalry

High performers routinely misattribute their own talent, mistaking an effortful skill that produces a dopamine hit for their true 'superpower,' which is usually something that feels effortless and therefore unrewarding to exercise.

Sankar's Superman analogy: flying and seeing through walls weren't arduous for Superman, they were just what he could do. He argues managers should help people locate that effortless capability and separately accept 'kryptonite,' a weakness so far below average it can't be trained away and should instead be avoided.

talent-and-culture

The fastest way to develop raw talent is to hand people responsibility beyond their proven ability rather than run them through a structured career ladder, because maximal learning tracks maximal tolerable pain, not comfort.

Sankar's 'gamma ray' analogy (Bruce Banner becoming the Hulk via a near-fatal dose, not progressive weight training): Palantir deliberately gives inexperienced people globally consequential problems, with enough organizational transparency that leadership can intervene if things go badly off the rails, rather than requiring the person to ask for help.

talent-and-culture

Culture is inherently entropic and degrades without daily reinforcement; the strongest cultures are deliberately 'cult-like' and hard for outsiders to blend into without adopting its norms.

Sankar says this requires recruiting people who treat the job as a calling rather than a series of employers, and cites Palantir practices like every new employee's first AMA ending with an invitation to tell him to 'fuck off,' plus twice-yearly 'weeks of revolt' where staff build whatever they want to prove the company has been getting something wrong.

talent-and-culture

Palantir's AI adoption inside the company was driven bottom-up by its newest, most junior hires, and leadership deliberately let their outsized productivity pull skeptics along instead of mandating adoption top-down.

Sankar notes AI plays to a 'beginner's mind' because new hires have no incumbent workflow to defend; he intentionally ignored cynics and let employees just months into the job outperform 20-year veterans, which he says did more to shift the culture than any directive could have.

talent-and-culture

Sankar's diagnosis, laid out first in his '18 Theses,' is that the US cannot repeat its WWII-style industrial mobilization because it no longer has a WWII-style dual-use industrial base, and the fix is 'the person is the program,' not better process.

He cites Edward Hall (Minuteman), Bernard Schriever (ICBM program), and Gene Kranz (the real driver behind what's remembered as 'the Apollo program') as evidence that major military innovation has historically come from empowered individuals bypassing bureaucratic requirements processes, not from following them, and argues cost-plus contracting removes the risk-taking incentive that makes this possible today.

heretics-and-founders

Books referenced

Boyd: The Fighter Pilot Who Changed the Art of War - Robert Coram - Patrick calls it one of the great biographies of a military figure he's read, about John Boyd, the fighter pilot who invented the OODA loop and helped design the F-16.
Destined for War: Can America and China Escape Thucydides's Trap? - Graham Allison - Patrick says he was reviewing his notes from this book ahead of the conversation before asking Sankar how he thinks about China as a political adversary.
Chip War: The Fight for the World's Most Critical Technology - Chris Miller - Cited alongside Apple in China as documenting how the US helped move microelectronics manufacturing to Southeast Asia after World War II.
Apple in China: The Capture of the World's Greatest Company - Patrick McGee - Sankar calls it a great book and cites its finding that Apple has spent the equivalent of two and a half Marshall Plans building manufacturing talent and capacity in China over five years.

Media referenced

18 Theses - article - Sankar's own manifesto, the first major public statement he put out arguing the US has lost military deterrence and needs to rebuild its industrial base; the shot clock is ticking.
Profile of Palantir (referred to in conversation as 'Jeremy's profile') - article - Patrick says reading this detailed profile was what first let him actually understand how Palantir's Gotham and Foundry products and business model work, since the company is otherwise hard to parse from the outside.

Companies

Palantir Technologies - Sankar is CTO; the whole conversation traces its history, forward-deployed-engineering model, ontology product thesis, and government/commercial mix.
Airbus - Used as the detailed case study of the ontology approach: Palantir embedded with final-assembly-line workers during the A350 ramp, starting with quality/nonconformity tracking and expanding to production planning and in-service support.
Anduril - Referenced (transcribed as 'Andrew') as an example of a hardware company pioneering the same heretic, spec-it-yourself model Andrew Higgins used for his landing boat.
SpaceX - Named alongside Palantir as one of the only two companies advocating for reindustrialization and defense-tech founders for the first solid decade, before hundreds of similar founders followed.
Google - Its 2017 'Attention Is All You Need' paper is used as Sankar's example that a team working on an incremental 3% improvement to Google Translate produced a revolutionary breakthrough, showing innovation flows from a production/input base.
Chrysler, Ford, General Mills, Kodak - Cited as Cold War-era 'dual-purpose' companies (minivans and missiles, satellites, torpedoes and inertial guidance, spy-satellite film) whose commercial R&D cross-subsidized national defense before the industrial base consolidated to defense-only specialists.
Northrop Grumman - Used to illustrate how today's defense primes were once named for individual founder-engineers (Jack Northrop, Leroy Grumman) rather than institutions.
Lockheed Martin (Skunk Works) - Kelly Johnson, founder of Skunk Works, is cited as building 41 airframes in his lifetime, including the U-2 in 13 months, as an example of a heretic-engineer operating with minimal process.
Apple - Discussed as the case study of a company whose manufacturing investment (via Apple in China) shows what committing capital to overseas production capacity looks like, contrasted with the near-absence of similar US investment.

Techniques and frameworks

OODA loop (Observe-Orient-Decide-Act) - John Boyd's decision-cycle framework, referenced repeatedly, including Sankar's joke about an 'American OODA loop' of observe-overreact-destroy-apologize, and as the conceptual root of Palantir's decision-centric ontology.
Forward deployed engineering - Palantir's model of sending engineers who are obsessed with real-world impact, not sales, to work directly with end users so the product is validated by whether it delivers an outcome, not just whether it sells.
Ontology as an enterprise operating system - Palantir's core product thesis that decisions, not data, create value; the ontology models both an organization's data and its actions/decisions, functioning as an API layer that makes a business programmable.
Superpower and kryptonite talent framework - Sankar's model for developing people: help them find their effortless 'superpower' (often misattributed early in a career) and accept an unfixable weak spot ('kryptonite') to be worked around rather than fixed.
Gamma-ray exposure (deep-end talent development) - Giving people responsibility well beyond their proven ability rather than a structured career ladder, on the theory that maximal learning correlates with maximal tolerable pain, not comfort.

Summary

Patrick O'Shaughnessy's conversation with Palantir CTO Shyam Sankar opens on the idea that gives the episode its title: heretics, not institutions, are the actual source of breakthrough innovation. Sankar walks through a gallery of them, Hyman Rickover building the first nuclear submarine from an office the Navy assigned him as a humiliation, Andrew Higgins designing the landing boat that ended up as 92% of the boats used at Normandy despite the Navy never asking for it, Billy Mitchell dying penniless before his vision became the Air Force. His framing is that these people are essentially founders operating inside bureaucracies that actively punish them, and that recognition only arrives retroactively, once the thing they built empirically works. He connects this directly to his own disposition and Palantir's origin story, describing the early years fighting government IT buyers who wanted an easier, worse product, and to his father's story of fleeing violence in Nigeria and rebuilding in Orlando, which he credits for both his gratitude and his refusal to accept "good enough."

A large stretch of the episode is a plain-language explanation of how Palantir actually works, prompted by Patrick's admission that he only understood the business after reading a detailed outside profile. Sankar lays out forward-deployed engineering as a deliberate answer to a weak feedback loop in enterprise software (will someone buy it) in favor of a strong one (did it work in the field), illustrated in detail through Palantir's Airbus engagement, where the company started by helping assembly-line workers bucket A350 production defects in Excel and expanded from there into production planning and in-service support. He frames Palantir's ontology, its model of both an organization's data and its actions, as the product's real innovation: the belief that value comes from decisions, not data, and that the ontology functions as an API layer making a business "programmable."

The conversation's second half turns to reindustrialization, the subject of Sankar's own "18 Theses." He traces a Cold War industrial base where companies like Chrysler, Ford, General Mills, and Kodak built both consumer goods and defense hardware, cross-subsidizing military R&D with commercial production, versus today's system where 86% of major weapons spending goes to defense-only specialists and just five prime contractors remain after a 1993 Pentagon dinner triggered a wave of consolidation. His broader claim, illustrated with Google's Transformer paper originating from incremental Google Translate work, is that innovation and production aren't separable: a country that offshores manufacturing eventually loses its capacity to innovate too.

On US-China competition, Sankar credits recent US strikes (Midnight Hammer, the Maduro-connected raid) with partially restoring lost deterrence, arguing their value is as much reputational, discrediting the Chinese and Russian weapons systems US adversaries rely on, as tactical. He frames the two countries' core asymmetries as patience versus unpredictability: China has systematically mapped US military dependencies since the 1991 Gulf War and broadcasts its intent, while the US advantage is a genuine capacity to pivot overnight, illustrated by how fast the American economy reoriented around AI.

The episode closes on talent and culture, where Sankar describes his "superpower and kryptonite" model for developing people (helping them locate an effortless strength while accepting an unfixable weakness) and his preference for throwing people at problems beyond their proven ability rather than a structured career ladder, on the theory that maximal learning tracks maximal tolerable pain. He describes culture as inherently entropic, requiring daily reinforcement and deliberately "cult-like" norms, and credits Palantir's AI adoption to junior hires who had no incumbent workflow to defend. He ends on his father again, closing with the story of a private-school admissions officer who admitted him past a deadline with no evidence beyond a hunch, calling it the kindest thing anyone has done for him.

Notable Quotes

"The only shit that ever worked, the things that helped us win all the wars were the things that the heretics actually did. Nothing that went through the machine delivered anything." - Shyam Sankar

"How did Bruce Banner become the Incredible Hulk? It wasn't progressive overload... It was a near fatal dose of gamma rays. 50% chance he died. 50% chance he turns in the big green monster." - Shyam Sankar

"It's no longer John Boyd's OODA loop, observe, orient, decide, act, it's the American OODA loop: observe, overreact, destroy, apologize." - Shyam Sankar

"The biggest lie that we bought from globalization is this concept of we will do the innovation and they will do the production. Well, you have to recognize that innovation is itself a consequence of productivity." - Shyam Sankar

"There's nothing about culture that is self-sustaining. That's as true for America, the country, as it is for your company." - Shyam Sankar