Bezos bets $100B on factories

You probably saw the Wall Street Journal headline: Jeff Bezos is in talks to raise $100 billion for a fund called Project Prometheus. Most of the coverage focused on the number. The more interesting part is what he's not doing with it. He's not building another model, not funding another AI startup. He's buying old manufacturing companies in aerospace, chipmaking, and defense, and rewiring them with AI.

Bezos co-founded the venture with Vik Bajaj, a former Google exec. They've already landed $6.2 billion in initial funding and Bezos has been meeting with sovereign wealth funds in the Middle East and Singapore to raise the rest. The targets are specific: pre-production processes like prototyping, design, and materials R&D. Not assembly line robots. The stuff that's too messy and variable for traditional automation but increasingly in range for AI that can reason about physical systems.

For four years, the default AI investment thesis has been: fund the teams building the models, the infrastructure, the tooling. Bezos is skipping that entire layer. He's betting the real returns come from buying underpriced businesses in industries that haven't been touched yet and running them differently. Think of it like a classic buyout fund, except instead of the usual playbook of cutting headcount and loading on debt, the value creation comes from AI making the operations fundamentally better.

That's a genuinely different thesis than what most of the market is chasing. And it reframes who's actually exposed. The thousands of mid-market manufacturers who figured they'd get around to AI eventually are now potential acquisition targets. Bezos doesn't need them to adopt AI. He just needs to buy them before they do. Meanwhile, VCs still racing to fund the next foundation model might want to glance at where the richest person on earth is actually putting his capital.

There's something clarifying about watching someone with essentially unlimited options choose not to build an AI company. The Axios breakdown of Project Prometheus reads like a bet that the AI gold rush already has too many miners and not enough people buying the mines.

We're hosting an official Claude Code workshop with Anthropic on April 10 in Midtown Manhattan. Half a day, 30 seats, hands-on. You bring a laptop, Anthropic engineers and AI strategists and builders from Tenex bring the expertise, and you leave with something you actually built — not a deck about what you could build someday.

Who it's for: Engineering leaders (EMs, VPs, CTOs, platform leads) and business leaders (CPOs, heads of digital transformation, senior AI sponsors). Up to 2 attendees per org.

What you'll do: Live demo of enterprise Claude Code workflows, then a guided build session where your agents write the code and you direct them. Show and tell at the end, networking lunch after.

Cost: Free. Application-based — we're keeping it small so everyone gets direct support.

Allie K Miller — the AI advisor behind workshops that have trained millions of professionals, named one of Time's 100 Most Influential in AI — ran a session recently that one of our community members took detailed notes on. Her core framing stuck with us: most people treat AI like a search bar they visit when they have a question. Miller treats it like a hire she onboards.

What that looks like in practice: In a live demo, she talked through a Dunkin' Donuts campaign idea out loud and Claude built a branded deck with speaker notes in under five minutes, launching three separate AI researchers at once to cover different celebrity partnerships. The key: the AI already had the brand's colors, fonts, and slide structure memorized from a previous setup. She didn't re-explain anything. And in another workflow she demonstrated, an agent pulled support emails, synthesized FAQ answers, ran them through her brand voice, and published the results to her website without any human touching it. That's what the full setup unlocks.

The framework: Miller structures her entire AI setup like a hiring process. Culture fit, knowledge base, skill set. Most people jump straight to asking Claude for stuff. She spends the upfront time teaching it who she is, how she works, and what great output looks like before she assigns a single task.

How to run the same playbook:

1. Write the job description. Claude's desktop app and Claude Code both let you create a permanent instruction file (called CLAUDE.md) that shapes how Claude behaves in every conversation. Think of it as a first-day onboarding doc. Miller's version tells Claude to act as a chief of staff: ask follow-up questions after completing a task, flag when instructions are underspecified, and maintain standing opinions about formatting, tone, and output structure across sessions.

2. Hand over the institutional knowledge. Miller built what she calls a Context Vault — a folder of documents that give Claude deep knowledge about you. Your values, career goals, business objectives, communication style. The idea is that every conversation builds on what Claude already knows about you instead of starting from zero each time.

3. Build repeatable workflows. Miller calls these "skills" — saved instructions that Claude can run on command. The Dunkin' deck wasn't a one-off prompt. It was a skill that already knew the brand guidelines, slide structure, and speaker note format. When someone needed another deck a week later, they just asked for it and Claude handled everything without being re-briefed.

4. Connect it to your tools. Claude can plug into your existing apps — Gmail, calendar, Slack, Chrome, Stripe — through connectors called MCPs (Model Context Protocol). That's how Miller wired up the workflow where Claude pulls emails, writes FAQs, and publishes them to her website autonomously.

The free resource: Miller's Context Vault has 8 copy-paste prompts for building your AI's knowledge base. Low bar to start. Try it this week.

27 Claude Code features most people miss — A clean, visual cheat sheet covering the features that turn a casual user into a power user. Covers everything from slash commands to agent orchestration in a format you can actually reference while working. Grab your copy now. Bookmark it

How to build skills in Claude Code (for non-developers) — Step-by-step video walkthrough of creating, running, and publishing skills without writing code. If the Allie K Miller section above got you interested in skills but you're not sure where to start, this is the on-ramp. Watch it

Deloitte surveyed 3,235 execs on AI adoption. It's not pretty. — Alex Lieberman walked through the findings and didn't hold back. Most enterprises are still barely past the pilot phase. If your org feels slow on AI, you're in crowded company — and that's either comforting or terrifying depending on your competitive position.

World models: the AI paradigm forming behind the LLM hype — Packy McCormick and Pim De Witte's deep dive into a new class of AI that simulates physical environments instead of predicting text. If you want to understand what comes after the chatbot era and where the next wave of $1B+ funding is going, this is the piece. Read it (long)

Andrew Ng built a Stack Overflow for AI agents — Context Hub gives coding agents access to current API docs instead of hallucinating outdated syntax. Open source, now covers 1,000+ APIs, and widely adopted by developers. If your team's agents keep making up code that doesn't work, this is the fix.

"Local agents are a dead end" — Sergey Karayev argues the future is hundreds of cloud-based agents running 24/7, triggered by bug alerts, Slack messages, and each other. Provocative take, but the reasoning holds up. Thread on X

Browser Use CLI 2.0 — Browser automation at 2x speed and half the cost, with 1.5M views on the launch video. Worth watching the 30-second demo to see where agent-controlled browsers are headed.