Rapid Improvements in… | Newcomer Newsletter Summary

Rapid Improvements in Foundation Models Surprise Even the Optimists

1. Key Themes

Theme 1: Frontier Model Progress Is Accelerating Faster Than Even Optimists Expected

The article's central thesis is that AI scaling fears are overblown and that insiders are more impressed than the public realizes.

"Scaling laws are holding up. Reinforcement learning and other techniques are proving effective. LLMs are improving even more quickly than anticipated — especially at Anthropic."

"Multiple speakers said the public is massively underestimating AI's future effects."

Theme 2: Anthropic Has Emerged as the Clear Frontier Model Leader — Surpassing OpenAI

This is a meaningful competitive shift with direct investment implications, as many VCs missed it.

"Mythos also solidifies Anthropic's new-found status as the clear industry leader. Amazingly, the company is apparently set to leapfrog OpenAI in revenue run-rate, leaving many VCs kicking themselves."

"Anthropic's industry-leading tech and impressive financial performance popped up constantly on the heels of the Mythos announcement and its revenue run-rate surpassing $30 billion."

Theme 3: AI-Native Cybersecurity Is Becoming a Critical and Investable Category

Anthropic's Mythos model — capable of autonomously finding and fixing software vulnerabilities — signals a step-change in what AI can do for (and against) security infrastructure, just as government cyber capacity is being gutted.

"It had built a model powerful enough to systematically find and fix vulnerabilities in almost any piece of software — even those that had escaped notice for years."

"AI has helped attackers become more prolific. It's up-leveled the skills of the bottom tier. But AI can really help the defenders." — Steve Schmidt, CSO at Amazon

"Ransomware alone yielded an estimated $57 billion for criminals last year and continues to disrupt a wide range of institutions."

Theme 4: Foundation Models Are Threatening Vertical AI Application Companies

The rapid capability gains of frontier models are directly challenging the moats of AI-native SaaS startups, particularly in high-value verticals like legal tech.

"Investors were again wondering where AI applications would fit in: One told us that they were now doubting the prospects of highly valued legal AI companies in light of Claude's capabilities."

Theme 5: VC Deal Volume Is Broadly Surging, Not Just at the Mega-Deal Level

Q1 data shows broad-based activity acceleration across the venture ecosystem, suggesting rising animal spirits beyond just a few headline rounds.

"New Crunchbase data shows that the uptick in activity was broad-based, with most firms doing far more deals than in the comparable quarter a year ago. Accel stepped up its game, writing 16 checks in Q1, with Andreessen Horowitz and Lightspeed close behind at 15 and 14."

2. Contrarian Perspectives

Perspective 1: The Free-Tier AI Experience Is Actively Misleading the Public About How Powerful Models Have Become

This is a counter-intuitive structural dynamic: the most capable AI tools are paywalled, creating a systematic underestimation of AI's power among the general public — which has meaningful implications for where the AI bubble narrative comes from.

"Andrej Karpathy made a case on X that it's partly due to the paid capabilities of foundation models, particularly coding tools, far outstripping the free tier of Claude and ChatGPT that most people outside of tech are most familiar with."

This suggests that AI pessimism in the mainstream may be a sampling artifact, not a signal about actual capability ceilings.

Perspective 2: Anthropic — Not OpenAI — Is Now the Company to Watch, and Most Investors Are Late

Conventional wisdom positioned OpenAI as the dominant, uncatchable frontier lab. The article suggests that narrative has already flipped, and that even professional investors were caught flat-footed.

"The company is apparently set to leapfrog OpenAI in revenue run-rate, leaving many VCs kicking themselves."

Perspective 3: Private Industry, Not Government, Is Now the De Facto Defender Against AI-Enabled Cyber Threats

With CISA facing a $700M budget cut and described as "in disarray," the article implies that the responsibility for national-scale cybersecurity has effectively defaulted to private tech companies — a major structural shift in how critical infrastructure risk is managed.

"The Trump Administration has insisted in the Anthropic case that private companies can have no role in deciding how the government uses their technology. On the other hand, it's ideologically opposed to regulations and other measures that might help address AI safety risks, leaving private companies little choice but to take matters into their own hands."

3. Companies Identified

Anthropic

Description: AI safety-focused frontier model lab
Why mentioned: Released Mythos, a model capable of autonomous vulnerability detection and remediation; reportedly surpassing OpenAI in revenue run rate; leading Project Glasswing cybersecurity consortium
Quote: "Mythos also solidifies Anthropic's new-found status as the clear industry leader... the company is apparently set to leapfrog OpenAI in revenue run-rate, leaving many VCs kicking themselves."

OpenAI

Description: Frontier AI lab and creator of ChatGPT
Why mentioned: Being overtaken by Anthropic in revenue; Stargate infrastructure project mentioned as fading; IPO timing uncertain
Quote: "Many attendees were still questioning if SpaceX, OpenAI, and Anthropic will be able to pull off their IPOs as planned later this year or next."

Carta

Description: Cap table and fund administration software
Why mentioned: Case study in deep Claude integration — building its product directly into Claude's desktop app as an operating tactic
Quote: "Carta...is literally building their product into Claude's desktop application, so users can run queries and ask questions about fund management without switching applications."

Zoom

Description: Video communications platform
Why mentioned: Integrating Claude to allow seamless transfer of meeting notes into Claude workspace — another integration case study
Quote: "Zoom CTO Xuedong Huang said an integration with Claude would let customers put meeting notes seamlessly into their Claude workspace."

Databricks

Description: Data and AI platform
Why mentioned: Publicly sparring with Snowflake at HumanX; CEO dismissed Snowflake as being "in the rearview mirror"
Quote: "Ali Ghodsi [dismissed] Snowflake as being 'in the rearview mirror' during his panel."

Snowflake

Description: Cloud data platform
Why mentioned: Public rivalry with Databricks; CEO implied Databricks' financials were unstable
Quote: "Sridhar Ramaswamy insinuated that Databricks' financials were all over the place."

Yahoo

Description: Legacy internet media and tech company
Why mentioned: CEO Jim Lanzone highlighted on the Newcomer podcast discussing a company comeback narrative
Quote: "Jim Lanzone, longtime tech leader and current CEO of Yahoo, stopped by the podcast studio and highlighted his company's comeback."

Glean

Description: Enterprise AI search and knowledge management
Why mentioned: CEO Arvind Jain met with the Newcomer team at HumanX — noted as a company to watch in the enterprise AI space

Synthesia

Description: AI video generation platform
Why mentioned: CEO Victor Riparbelli met with Newcomer team at HumanX

Applied Intuition

Description: Autonomous vehicle and defense tech software
Why mentioned: CEO Qasar Younis met with Newcomer team; implied relevance to defense tech growth rounds mentioned in the newsletter

Superhuman

Description: AI-powered email client
Why mentioned: CEO Rahul Vohra met with Newcomer team at HumanX

4. People Identified

Steve Schmidt

Description: Chief Security Officer, Amazon
Why mentioned: Quoted extensively on AI's dual-use nature in cybersecurity — both empowering attackers and defenders
Quote: "AI has helped attackers become more prolific. It's up-leveled the skills of the bottom tier. But AI can really help the defenders."

Vinod Khosla

Description: Founder, Khosla Ventures; prominent OpenAI backer
Why mentioned: Representing the bull case for large foundation models and enterprise adoption, despite being "firmly Team OpenAI"
Quote: "Vinod Khosla...said he's confident the big models will see massive growth, since it's still very early days for large enterprise adoption."

Andrej Karpathy

Description: AI researcher and former OpenAI/Tesla executive
Why mentioned: Offered a structural explanation for why the public underestimates AI — the free-vs-paid capability gap
Quote: Karpathy "made a case on X that it's partly due to the paid capabilities of foundation models, particularly coding tools, far outstripping the free tier of Claude and ChatGPT that most people outside of tech are most familiar with."

Henry Ward

Description: CEO, Carta
Why mentioned: Shared a concrete example of deep Claude integration as a product strategy
Quote: Carta "is literally building their product into Claude's desktop application, so users can run queries and ask questions about fund management without switching applications."

Ali Ghodsi

Description: CEO, Databricks
Why mentioned: Publicly dismissed Snowflake at HumanX, illustrating competitive dynamics in the data platform space
Quote: "Ali Ghodsi [dismissed] Snowflake as being 'in the rearview mirror' during his panel."

Sridhar Ramaswamy

Description: CEO, Snowflake
Why mentioned: Fired back at Databricks by questioning their financial stability
Quote: "Sridhar Ramaswamy insinuated that Databricks' financials were all over the place."

Jim Lanzone

Description: CEO, Yahoo
Why mentioned: Featured on the Newcomer podcast discussing Yahoo's resurgence
Quote: "Jim Lanzone, longtime tech leader and current CEO of Yahoo, stopped by the podcast studio and highlighted his company's comeback."

Ali Partovi

Description: CEO/Founder, Neo (early-stage VC fund)
Why mentioned: Named in the newsletter subject line as "shining" — highlighted as a notable VC performer (details behind paywall)

Xuedong Huang

Description: CTO, Zoom
Why mentioned: Shared Zoom's Claude integration strategy as a practical enterprise use case
Quote: "An integration with Claude would let customers put meeting notes seamlessly into their Claude workspace."

Janice Chen

Description: Executive, Mammoth Biosciences
Why mentioned: Won the Newcomer poker event; noted as a biotech figure in the Cerebral Valley tech community

John Carreyrou

Description: Investigative journalist; broke the Theranos story
Why mentioned: Taking on a new investigation, this time targeting Satoshi Nakamoto
Quote: "John Carreyrou of Theranos fame takes on Satoshi Nakamoto."

5. Operating Insights

Insight 1: Embed AI Directly Into Your Product's Native Environment to Reduce Switching Friction

Rather than building standalone AI features, leading companies are integrating into AI platforms where users already live. Carta's approach — letting users query fund management data without leaving Claude's desktop app — is the template.

"Carta...is literally building their product into Claude's desktop application, so users can run queries and ask questions about fund management without switching applications."

Implication for operators: The distribution moat is shifting. The question is no longer "does your product have AI?" but "where does your product live in the AI workflow?" Companies that integrate into Claude, ChatGPT, or other AI workspaces may outcompete those requiring app-switching.

Insight 2: Don't Benchmark Your AI Product Against Free-Tier Models — Enterprise Customers Are Experiencing a Fundamentally Different Capability Level

Karpathy's observation is operationally critical: the gap between free and paid AI tiers is wide enough that public perception and enterprise reality are diverging. If you're building for enterprise, you're building against a much more capable baseline than most discussions assume.

"The paid capabilities of foundation models, particularly coding tools, far outstripping the free tier of Claude and ChatGPT that most people outside of tech are most familiar with."

Implication for operators: Competitive analyses and customer education strategies should account for this gap. Enterprise buyers are already using — and expecting — significantly more powerful AI tools than what's visible in mainstream benchmarks.

6. Overlooked Insights

Insight 1: Government Pullback from Cybersecurity Creates a Direct Market Opportunity for Private AI Security Solutions

The article mentions almost in passing that CISA — the primary US government agency for cybersecurity defense — is facing a $700M budget cut and is "in disarray." Combined with Project Glasswing launching as a private-sector consortium, this signals a structural vacuum forming in national cyber defense that private companies and investors may be positioned to fill.

"President Trump...wants to slash its [CISA's] budget by $700 million, on top of earlier cuts...CISA, unsurprisingly, is said to be in disarray."

This is an overlooked investment setup: public-sector retreat + AI-native threat escalation + private consortium formation = potential wave of cybersecurity infrastructure investment.

Insight 2: The NSA/CIA "Equities" Problem May Be Obsolete in an AI World

The article briefly notes the longstanding tension between government agencies hoarding discovered vulnerabilities for offensive use versus disclosing them for patching — the so-called "equities" process. If Mythos-level AI can now autonomously find those same vulnerabilities independently, the government's informational advantage in this area may be rapidly eroding, with unpredictable consequences for national security doctrine.

"Oftentimes, when cyber researchers at the NSA or the CIA discover a software vulnerability, they are inclined to keep it to themselves for their own hacking purposes...The decision on how to handle such situations has been governed by a so-called 'equities' process that tries to balance risks and benefits, but it's often contentious."