NOW WORKSHOPPING
Episode 6
Add a topic: Tag the bot in Telegram — “add to docket: [tweet / article / idea to cover]”
IN THIS EPISODE
ALL TOPICS
AI Written Output Has Already Surpassed All of Human History
Brett Winton, Chief Futurist at ARK Invest, published two charts showing AI written output exceeded all human written output in 2025. Chart 1: cumulative written output since 1500 — the printing press to today. The AI line appears around 2023 and crosses the entire 500-year human curve by 2025. Chart 2: annual output — AI annual output in 2025 already exceeds total human annual output, and the projection goes vertical. Elon quote-tweeted: 'AI content will vastly exceed all human content.' 22 million views. The critical point nobody is saying: the charts are on a log scale. The AI line appearing slightly above the human line could represent a 10x, 100x, or larger real gap. The visual dramatically undersells the reality. The question: if AI is already producing more text than all of human history combined — every year, accelerating — what happens to the information ecosystem? Signal-to-noise collapses. Attention becomes the only scarce resource. Human-verified content becomes a premium category. This is the world you're building your content business in right now.
OpenAI Kills Sora — Three Months After Disney's $1B Bet
OpenAI announced March 24: 'We're saying goodbye to Sora.' Three months earlier, Disney had signed a $1B, three-year licensing deal covering 200+ characters including Marvel, Pixar, and Star Wars. Official reason: compute redirected to robotics. Timeline: Sora announced Feb 2024 — goes viral, Hollywood panics. Launched as TikTok-style app December 2025. Disney deal signed. Killed March 2026. 90 days between Disney's billion and the shutdown. What this reveals about OpenAI: they are not optimising for revenue, customer relationships, or enterprise trust. They are optimising for whatever they believe is the frontier — and everything else is disposable. For the AI video category, this is actually a green light: OpenAI just vacated the space. Runway, Kling, and everyone else no longer have to compete with the most-funded AI lab in the world. But the enterprise trust question is real: would you sign a product deal with OpenAI after this?
Voice Is Dead as a Business — ElevenLabs Gets Squeezed From Both Sides
ElevenLabs raised at ~$1.1B valuation on the thesis that high-quality AI voice was hard. This week both assumptions broke simultaneously. Hit 1: Mistral open-sourced a TTS model that beats ElevenLabs — runs locally on 3GB of RAM, free. No API fees, no per-word billing. Hit 2: Google shipped Gemini 3.1 Flash Live — native audio and voice baked into Gemini, faster response times, more natural dialogue, 2x longer memory, rolling out in 200+ countries. ElevenLabs is getting hit from open source below and Big Tech above. The pattern is now so established it should have a name: startup proves use case viable → lab ships free version → category commoditised → startup either moves upmarket or dies. This happened to image gen (Midjourney → free APIs), memory infrastructure (see item 3), coding assistants, and now voice. The question: what is the new defensible layer, and how long before that gets commoditised too?
Google PM Open-Sources Memory Agent — Kills a Startup Category Before Lunch
A Google product manager open-sourced an 'Always On Memory Agent' built on Gemini 3.1 Flash-Lite and Google's ADK. No vector database — pure LLM-driven persistent memory. Runs continuously, captures what matters, injects context into future sessions. Same week: Anthropic shipped AutoDream for Claude Code — a background sub-agent that auto-consolidates and prunes memory files, pairing with AutoMemory for a full capture/refine memory loop. The memory startup category was supposed to be worth building companies around. Mem.ai raised $23.5M. Rewind AI raised $10M. Claude-mem (open source) has 27.7k GitHub stars. All of them are now competing with something a Google PM shipped using free tools. The data point that makes this concrete: 80% accuracy on the LOCOMO benchmark vs 29% for file-based memory approaches — the open-source version isn't just cheaper, it's measurably better. What's left defensible in AI infrastructure when the PM who built the model gives away the layer you were selling?
The GitLab Founder Who Became His Own Clinical Trial — And Próspera
Sytse Sijbrandij co-founded GitLab ($15B IPO). Diagnosed with osteosarcoma — bone cancer in his T5 vertebra. Ran out of standard-of-care options. No trials available. Instead of accepting that, he applied startup logic: maximum diagnostics across multiple institutions, AI-designed treatment protocols, parallel experiments instead of sequential trials. Made 25TB of his own health data publicly accessible on Google Cloud. Presented at OpenAI Forum: 'From Terminal to Turnaround: How GitLab's Co-Founder Leveraged ChatGPT in His Cancer Fight.' Now building evenone.ventures to scale this model for others. The jurisdiction that makes this scalable: Próspera, a ZEDE in Honduras with its own regulatory framework. Already running: Minicircle (gene therapy trials for longevity), Unlimited Biotech (Phase I/II VEGF/follistatin regimen, dosing participants Jan 2026, data readout April 2026). Próspera enables trials 10-100x faster than FDA. Sytse's AI-designed protocols + Próspera's jurisdiction = a stack the current system cannot compete with. The political fault line: if a terminal patient consents to an AI-designed experimental protocol, who has the right to say no?
Terafab — Elon Builds His Own Chip Factory
Elon announced Terafab on March 21, 2026 — a $25B vertically integrated chip fabrication plant built jointly by Tesla, SpaceX, and xAI in Austin. Target: 1 terawatt of AI compute per year. The reason: chip suppliers can't keep up. Quote: "We either build the Terafab or we don't have the chips. And we need the chips. So we build the Terafab." Fully vertically integrated — chip design, lithography, fabrication, memory, packaging, testing all under one roof. Feeds xAI, Tesla Dojo, Optimus robots, and SpaceX space data centers. Elon changed his X bio to Terafab.ai.