Speaker
Ben Broch
Appearances over time
3 episodes
Episodes
3
Podcasts
Quotes & moments
A third-party harness with no privileged access outperforms Anthropic's own Claude Code when running Anthropic's Opus 4.7 model.
Cursor has been perfecting the developer experience since 2022, giving it years of institutional knowledge across multiple models.
Gavin Baker argued on Invest Like the Best that AI token demand is growing 10x next year while supply is only growing 3x, implying prices must rise.
Ben estimated that only around 500,000 people globally currently use any form of agentic coding workflow, out of 8 billion people.
Ben Broch predicted that long-term access to Claude Fable will require a $500–$1,000/month subscription tier, as the current free access is a hook to get users addicted.
Ben observed that Claude's Google Trends score dropped roughly 30–40% over the summer while ChatGPT fell only about 10%, signalling a sentiment shift back to OpenAI.
Ben measured ChatGPT's new full-duplex live voice model responding in under one second in nearly every query, a dramatic improvement over previous generation voice AI.
Current global data-centre compute capacity sits at approximately 0.1 terawatts, making Elon's 100-terawatt compute target 1,000 times today's total.
Chamath's firm 8090 reported that inference spend is doubling every 45 days, driven by continuously expanding AI use cases.
The US Department of War named seven AI partners — SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, and AWS — notably omitting Anthropic.
Over 1.3 million Department of War personnel have used the AI platform, generating tens of millions of prompts and hundreds of thousands of agents in only five months.
Ben speculated that Anthropic has already trained a Mythos 2 model scoring in the 80s or even 90s on intelligence benchmarks, but has not released it.
The original Human Genome Project took 13 years to sequence the genome and only achieved 92% completion of the sequence.
Ben observed that to make an AI product 10% better than before now requires spending 100% more tokens, meaning optimisation gains are shrinking.
In crypto, Phantom wallet made more money than Solana itself. The same dynamic is playing out in AI: Cursor and other abstraction layers will capture more value than the model labs running the underlying compute. Every tool call is a potential revenue event.
Grok 4.3 launched at $1.25 per million input tokens versus Anthropic's $5 and OpenAI's $5 — roughly one-quarter the price. It also claimed the number-one spot on Bridgebench with the lowest hallucination rate in the field. The AI race is shifting.
Elon Musk's SpaceX compensation package only vests if SpaceX hits a $7.5 trillion valuation, launches 100 terawatts of compute into space (1,000 times today's global data-centre capacity of 0.1 TW), and establishes a Mars colony of 1 million people. People laughed at the Tesla package too.
Stripe Treasury is now an AI-native bank: hold multi-currency balances, pay anyone in 160 countries by email, earn 2% cashback — and connect it all to your AI agents via the Stripe MCP. Stripe already knows your revenue trajectory better than you do.
The US Department of War chose SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, and AWS as its seven AI partners. Anthropic — widely considered a leading safety-focused lab — is conspicuously absent. AWS, a major Anthropic investor, made the list.
For three years, VC money secretly transferred from light users to power users via flat AI subscriptions. GitHub Copilot's shift to usage-based billing from June 1 is the canary in the coal mine. AI subsidies will disappear just like Uber's — and then everyone becomes a power user or pays the price.
Cursor's new SDK lets anyone embed its full harness, runtime, and models into their own product. Cursor now competes with the OpenAI and Anthropic — the same companies it pays billions to for inference. This is the same collision that happened between cloud providers and Snowflake.
Codex now integrates a browser (via former Atlas) and HeyGen video generation directly inside the app. Engineers prefer it because everything is in one frame — no switching between tools. Claude Code's split UX is costing Anthropic users.
Luca Arrigo is getting his entire genome sequenced and plans to feed the raw data into AI models to find personalised health insights. His business vision: open a local clinic where people come in for a saliva swab, outsource the sequencing to external labs, and use AI to deliver personalised medicine at scale.
Apple Intelligence is 'one of the most pathetic versions of AI in 2026.' Tim Cook stepped down because he had no idea what AI was. OpenAI is building an AI-first phone targeting 2028, and Apple has roughly one year before John Turnis's product-focused leadership has to prove it can compete.
Agents calling tools are the new Google searches. Services like Apify that gate access to hard-to-scrape platforms (Facebook, TikTok, Reddit) become indispensable once agents can't access data themselves. Owning proprietary data sources is the most defensible position in the next five years.
OpenAI and Anthropic face fierce competition and a hard compute ceiling — a 10x growth from their current valuations would require consuming all the world's energy. SpaceX's market is literally off-planet. All three hosts picked SpaceX when asked to choose one investment.
Grok 4.5 hits benchmark parity with Claude Opus 4.8 at roughly 70% lower cost per task. For high-volume agentic workloads, this isn't just impressive — it may make Grok the default grunt-work model in every serious builder's stack.
Use Claude Fable for strategic planning, GPT-5 as a middle manager checking the work, and Grok 4.5 for cheap token-burning grunt tasks. This three-layer stack combines frontier intelligence with real cost efficiency.
Cursor extracts more performance from Opus 4.7 than Anthropic's own Claude Code does — despite having no privileged access. The harness is a different discipline from model training, and Cursor's years of multi-model developer experience give it an edge the labs can't easily replicate.
Analysis
What they talk about
- Technology 63%
- Business 22%
- Education 4%
- Government 4%
- Health & Fitness 4%
- Society & Culture 3%
Connections
Shows they appear on and people they share episodes with. Drag to explore.
