20VC with Harry Stebbings — searchable YouTube transcripts & AI Q&A

Trillion or Bust: The New VC Game | CoreWeave Sell-Off Creates Panic | General Atlantic Eyes IPO

The venture capital market is increasingly concentrating on a few ultra-high-growth AI companies, leaving many profitable, fast-growing startups unfundable, the speakers argue. They contend that mega-funds require billion-dollar positions, making smaller outcomes unattractive, and that the AI trade is showing cracks, as seen in CoreWeave's sell-off and Anthropic's slowing growth. They also highlight code review as the most important software category, with the build-to-production pipeline being the key opportunity.

The Untold Truths of Venture | Who Wins & Loses in the Next Decade of VC | Sameer Dholakia

Sameer Dholakia argues that the AI market will be larger than oil because intelligence powers all humans, and he believes the current AI wave is bigger than the internet and cloud combined. He advises investors to focus on next-twelve-month revenue multiples rather than current ARR, citing investments in Anthropic and Lora as examples where high entry prices were justified by rapid growth. He also predicts a permanent shift toward companies staying private longer, with an IPO unlock expected in 2027-2028.

Horizontal vs Vertical Agents: Who Wins? | Model Routing: Commodity or Increasing in Value?

Harrison Chase argues that building reliable agents remains extremely difficult because LLMs are non-deterministic black boxes, and the value in agent systems lies in the harness layer that manages context, memory, and routing. He recommends starting with context engineering and harness tweaks before considering model post-training, and he is bullish on vertical agents where domain experts can build effective evals. He also predicts that memory will become a common feature in all products, though the best implementation is still unclear.

Valar Atomics Sues Day One Ventures | Meta's Back to the AI Frontier | NVIDIA's $30B Reflection Deal

The hosts argue that pro rata rights are a valuable option that early investors should not casually waive, despite pressure from large funds like Sequoia. They illustrate this with the Valor Atomics dispute, where Day One Ventures fought to retain its pro rata rights after being asked to waive them, and note that even investors who publicly support waiving rights often do not waive their own. The discussion also covers Meta's new Muse Wonder model, which is seen as performative since it lacks a public API, and Nvidia's potential $30 billion acquisition of Reflection, which would consolidate open-source model development.

Atlassian CEO: Why Agents Won't Replace Interfaces | Why The Jury’s Still Out on Specialised Models

Mike Cannon-Brookes argues that agents will not replace user interfaces; instead, design and experience become even more critical in the AI era. He emphasizes that Atlassian's strategy focuses on building the best enterprise context graph and evolving design to handle agentic workflows, while remaining skeptical about the long-term benefits of specialized models. He also notes that cost discipline and model routing are key to delivering value, and predicts a shift from information asymmetry to imagination asymmetry.