AI's Hidden Bottlenecks & The New Rules of Work

THE THROUGHLINE

#### The Geopolitics of AI: Beyond Chips, It's About Control and Talent

Anthropic's Fable 5 AI model was abruptly banned by the US government, igniting a sharp debate on the All-In Podcast. David Sacks explained that Anthropic's CEO expanded a pilot without White House consultation, triggering the ban. Chamath Palihapitiya argued this "evasiveness and immaturity" erodes trust, paving the way for hyperscalers like Amazon and Microsoft to become trusted, regulated gatekeepers, creating an AI oligopoly.

The ban on Anthropic’s models extended to all foreign nationals, even the company’s own employees. On TBPN, John Coogan highlighted this unprecedented restriction as a massive compliance and operational challenge, making it harder for US AI companies to recruit and keep top international talent.

Sebastian Mallaby, author of "The Infinity Machine," argues on The Tim Ferriss Show that government intervention in frontier AI is no longer a theoretical risk. He points to the Trump administration's "180" on AI control after Anthropic's Mythos model demonstrated the ability to "cyber attack almost anything and penetrate it," forcing the government to requisition decision-making authority over the model.

This capital intensity has shifted who funds the biggest bets in deep tech. Lip Bu Tan, CEO of Intel, explained on No Priors that governments are now evolving from grant providers to direct shareholders, citing Taiwan's early support for TSMC and the US government's investment in Intel as essential industrial policy for national infrastructure.

#### AI Is Reshaping Work: 8x Velocity, Zero Backlogs, and New Human Challenges

Fiona Fung, who leads the Claude Code team at Anthropic, dropped a bombshell on Lenny's Podcast: their engineers now ship eight times more code than they did in 2021. This isn't about working harder, but about AI fundamentally changing what "engineering" means, making coding itself no longer the primary bottleneck.

On How I AI, Anker Goyel argues that AI agents eliminate excuses for product backlogs. He stated, “there's just no excuse to not have rigor... we don't really have a backlog,”▶ 13:39 meaning every "paper cut" or performance issue gets addressed immediately, shifting the primary engineering task to removing complexity for better user experience.

Anjney Midha, CEO of Amp, warns against abandoning established engineering principles in the AI rush. On Latent Space, he pushes for an "output maxing" philosophy, focusing on getting maximum value and optimal outcomes from existing compute resources rather than simply adding more.

Perplexity CEO Aravind Srinivas believes the underlying AI model is quickly becoming a utility. On 20VC with Harry Stebbings, he suggested that the real value for founders now sits in building sophisticated orchestration systems that make models useful, not just powerful, as the model is no longer the product.

#### The Internal Game: Redefining Success and Building Personal Systems

Many founders spend their twenties and thirties chasing a ghost, often just a reflection of someone else's ambition. Shaan Puri, on My First Million, noted that valuable individuals are “anti-mimetic, meaning people who seem to want things from their own internal volition and not because other people want those same things.”▶ 45:25

Andy Stumpf's 'Toilet Paper Principle' argues that consistently choosing the "slightly harder" option in trivial daily tasks prevents future complications and builds core discipline. Andrew Huberman, on Huberman Lab, links this micro-discipline to the anterior mid-cingulate cortex, a brain region that literally grows in volume when individuals successfully complete tasks they initially resist.

On The Tim Ferriss Show, Sebastian Mallaby warned that the biggest risk with large language models isn't job loss, but intellectual laziness. He stresses, “That is not the route to happiness or satisfaction or anything. We need to continue to do the hard work of preparing our minds because that's what makes us people.”▶ 1:32:02

#### AI Infrastructure Squeeze: The Grid, Community, and Funding Realities

Founders and investors often fixate on chips as the ultimate constraint, but Perplexity CEO Aravind Srinivas argues on 20VC with Harry Stebbings that the real choke point is far more mundane: the physical infrastructure that powers and houses these chips. He said, "I think the biggest problem is actually in power," citing the slow, complex processes of securing land, permits, and grid connections.

Building large-scale AI infrastructure increasingly runs into a wall of community opposition. Anjney Midha, CEO of Amp, highlighted on Latent Space that up to 20% of new US data center projects this year risk stalling due to a lack of local community support. His solution isn't PR, but direct economic partnership, suggesting cash payments to communities by sharing marginal compute revenue.

Intel CEO Lip Bu Tan believes AI's impact will be "bigger than internet" and "more profound," but warned on No Priors that its rapid growth faces immediate bottlenecks in power, helium supply, and critically, memory. He cautions these shortages mean "pricing also go up because we have to pass the price the cost to the customer."

2. Best Of the Week

20VC with Harry Stebbings: Flexport CEO Ryan Peterson doesn’t mince words, calling remote work "white collar fraud," rooted in his own experience that childcare makes home productivity a "total fantasy." Read more.

All-In Podcast: David Sacks, championing President Trump's Iran peace deal, sharply dismissed calls for military intervention, stating that pushing for ground troops and regime change in Iran would be a "suicide mission." Read more.

Dwarkesh Podcast: Dwarkesh Patel and Ada Palmer peel back the layers of Machiavelli's Italy, revealing that patronage wasn't a flaw; it was "the operating system" for governance, trust, and even daily commerce. Read more.

How I AI: Anker Goyel believes AI evals are the modern PRD, allowing teams to specify success criteria quantitatively with examples, moving beyond mere prose; he states “EVELs are actually the modern version of a PRD.”▶ 25:20 Read more.

Huberman Lab: Andy Stumpf, a former Navy SEAL, described his contentious nearly two-year divorce as the “hardest thing I'd ever done in my life,”▶ 1:03:12 a greater challenge than his SEAL training. Read more.

Latent Space: Ronak Malde, co-founder of Trajectory.ai, argues that current AI models are static and repeat mistakes, stating, “The model that you used yesterday, it's going to be the same model making the same mistakes tomorrow.”▶ 7:31 Read more.

Lenny's Podcast: Fiona Fung, who leads Claude Code at Anthropic, observed that "AI-driven productivity can lead to engineer isolation," prompting new social initiatives to keep her team connected. Read more.

My First Million: Lloyd Blankfein, the former CEO of Goldman Sachs, proposes an "Obituary Test": your career should fill "no more than three of those nine paragraphs" in your eventual obituary, as a mental model for a balanced life. Read more.

No Priors: Intel CEO Lip Bu Tan isn't sugarcoating it: Intel is transforming from what he calls a "very old legacy spreadsheet company" into an organization where AI is embedded across design, engineering, and every part of the business. Read more.

TBPN: Gavin Baker, on TBPN, labels the social media-driven surges in small-cap AI stock prices as "shady stuff," drawing a stark parallel to the speculative euphoria and questionable practices seen during the dot-com bubble. Read more.

The Tim Ferriss Show: Sebastian Mallaby reveals that prominent AI figures like Demis Hassabis and Ilya Sutskever frame their pursuit of AGI in spiritual terms, such as "finding God's algorithm," articulating their core mission. Read more.

3. Most Quotable

"I have a three-year-old and a 5-year-old. The idea that I could do any work at my house is like a total fantasy. Like, come on. You’re kidding."

Ryan Peterson on 20VC with Harry Stebbings · June 21, 2026. The Flexport CEO's blunt assessment of remote work's feasibility, particularly for parents, is a contrarian claim in modern workplace discourse.

"The model that you used yesterday, it's going to be the same model making the same mistakes tomorrow."

Ronak Malde on Latent Space · June 21, 2026. The Trajectory.ai co-founder's sharp prediction suggests that static AI models, despite their power, are already a dead end, demanding continuous, real-time learning.

"For a field like legal like getting 80% of the way there is the same thing as zero."

Ronak Malde on Latent Space · June 21, 2026. Malde's stark assessment highlights that for high-stakes industries, incremental AI accuracy is meaningless without the final, critical leap to complete correctness.

Bottom Line: This week, founders grappled with AI's unprecedented demands on infrastructure, its profound reshaping of how we work, and the deeply personal quest for authentic success amidst constant change.