We're still in the first inning of AI, and this week's lineup shows why. Unconventional AI and TypeSafe AI are rethinking the computer and the model from the ground up. ElevenLabs is turning its voice stack into a roughly $29-a-month AI receptionist. Shield AI is reportedly raising at $20B+. Aaron Ross and Aaron Dillon break down where the new AI efficiency trade is forming, and why entry-point discipline matters in defense tech. Unconventional AI is one of the boldest infrastructure bets in the private market. Naveen Rao, the former Databricks AI chief who built MosaicML, raised a reported $475M seed at a $4.5B valuation. His goal is to replace the decades-old von Neumann architecture with oscillator-based, brain-inspired compute. His first model, Un-0, matches state-of-the-art image diffusion while running on a software simulation of that chip, and Rao's stated target is a 1,000x cut in inference power. Our house view is that energy is the binding constraint on AI scaling, and whoever credibly bends that curve owns the next decade of infrastructure. The milestone we would watch is working silicon, not simulation. TypeSafe AI attacks the same cost curve from the model side. Founder Diogo Almeida is a former OpenAI researcher behind the InstructGPT and RLHF work that became ChatGPT. He emerged from two years in stealth with a $40M DCVC-led seed and Jev, a model that outputs structured decisions instead of text. Jev is priced at $0.042 per million input tokens with 70–500 ms response times, and within days Vercel called it the fastest-adopted model in its AI Gateway's history. We believe inference economics have been missing exactly this: computers talking to computers in their own language. ElevenLabs Reception moves the voice leader from infrastructure into outcomes. It's a 24/7 AI receptionist that answers calls, books appointments, and speaks 70+ languages. Our view is that AI applications are converging on outcomes rather than categories. A CEO buying a booked meeting doesn't care whether it came from a voice company, a chatbot company, or a search company. Sales and marketing may be the biggest near-term winner of the AI cycle. Shield AI is reportedly in talks to raise at a valuation of at least $20B, up from $12.7B in March. It is guiding to 80%+ revenue growth and at least $540M in 2026 revenue. Hivemind autonomy software is the pitch, but V-BAT and the upcoming X-BAT make this a hardware business at a software multiple. We frame defense tech as a durable value franchise with real government demand, not an unlimited growth story. Aaron Ross (Ross Pre-IPO) and Aaron Dillon (Managing Director, AGDillon & Co.) close with where AGD is focused today: AI infrastructure, specifically inference and open source, with Positron, Together AI, and Baseten on the whiteboard. In this episode: Why energy is AI's real bottleneck, and Naveen Rao's 1,000x thesis TypeSafe's Jev and the case for models that don't speak English Google's 3.2 quadrillion monthly tokens and why AI compute is still scarce Meta's Muse and the coming wave of personal AI agents ElevenLabs Reception, speed-to-lead, and the convergence of AI applications Shield AI, Anduril, Castelion: growth company or value play? AGD's focus: inference, open source, Positron, Together AI, Baseten Sections 00:00 Intro: four companies changing the game 00:30 First take: Unconventional AI and TypeSafe AI 01:54 Naveen Rao and the energy thesis behind Unconventional AI 03:21 Dynamical systems: a computer that works like a brain 05:52 Google's 3.2 quadrillion tokens and AI agents' "gibber mode" 07:18 Constraint breeds innovation: China, compute, and capitalism 08:59 Sold out: why AI compute is still scarce 10:22 The first inning: Meta's Muse and personal AI agents 13:49 ElevenLabs Reception: a $29/month AI receptionist 16:34 Outcomes over silos: AI applications converge 18:35 Meeting customers where they are 21:27 Would AGD invest in Unconventional or TypeSafe? 22:37 Defense tech hype: Anduril, Saronic, Shield AI 23:50 Shield AI: software story or hardware multiple? 24:56 What's the TAM? Dillon's defense tech skepticism 29:12 Castelion and cutting through the noise 30:36 Where AGD is focused: inference, open source, Positron, Together AI, Baseten 32:25 Close #UnconventionalAI #TypeSafe #ElevenLabs #ShieldAI #AI