Episode 49 unpacks a week where three companies answered the same question in three different ways: what does AI actually cost, and who pays for it. Anthropic shipped Claude Sonnet 5 at $2 per million input tokens, close to flagship Opus 4.8 quality, betting most work doesn't need the expensive model to get expensive-model results. Meta started renting out its spare AI compute as a new business called Meta Compute, chasing the same cloud dollars as AWS and Google. Tesla went the other way, capping employee AI tool spending at $200 a week months after ranking engineers by how many tokens they burned.

What's Covered

* Anthropic's Claude Sonnet 5 launch and its near-Opus pricing bet

* Meta Compute: turning spare AI infrastructure into a product

* Tesla's $200-a-week cap on employee AI tool spending, and the Grok exception

Sources

* MacRumors: Anthropic releases Claude Sonnet 5

* TechCrunch: Meta, like SpaceX, looks to turn excess AI compute into cash

* Electrek: Tesla caps employee AI spending at $200/week

This week's episode digs into the moment the AI race moved from pure software to the hardware and control beneath it. OpenAI and Broadcom reveal Jalapeño, OpenAI's first custom chip. Anthropic takes a 29-million-query theft of Claude to the White House. And GPT-5.6 arrives gated to about twenty vetted organizations. Along the way: the tool that turns messy PDFs into clean data, and a prompt move that lets a cheap model punch above its weight.

What's Covered

* OpenAI and Broadcom's Jalapeño chip and the move off Nvidia

* Anthropic's distillation accusation against Alibaba

* GPT-5.6 Sol, Terra, and Luna, and the government-gated rollout

* LlamaParse, the tool of the week

* Metaprompting, the prompt of the week

Sources

* OpenAI unveils its first custom chip, built by Broadcom (TechCrunch)

* Anthropic claims Alibaba illicitly distilled Claude (Tom's Hardware)

* Previewing GPT-5.6 Sol, Terra, and Luna (OpenAI)

* LlamaParse (LlamaIndex)

* Anthropic Economic Index, June 2026

* Users shift from tokenmaxxing to efficiency (CNBC)

* US clears Mythos 5 for 100+ organizations (TechCrunch)

* Nobel laureate John Jumper leaves DeepMind for Anthropic (TechCrunch)

* Google delays Gemini 3.5 Pro to July (Crypto Briefing)

This week's deep dive traces AI's shift from tools you prompt to agents that act on their own. We get into loop engineering and the six pieces that make an autonomous agent work, the open-source canvas Dify and KPMG's rollout of Microsoft Agent 365, Coinbase letting agents trade through the x402 protocol, and Jeff Bezos' Prometheus raising 12 billion to build an artificial general engineer. Then the bill comes due: a Munich court holds Google liable for what its AI Overviews invent, a former xAI engineer sues over Grok safety, and the US government forces Anthropic to pull Fable 5 and Mythos 5 three days after launch, all while OpenAI files to go public near a trillion-dollar valuation.

CHAPTERS

* 0:27 - From Prompting to Autonomous Loops

* 2:10 - Loop Engineering and the End of Prompting

* 3:13 - The Six Building Blocks of an AI Loop

* 6:50 - Dify and Agent 365 Go Mainstream

* 7:43 - Agents With Wallets: Coinbase and x402

* 9:00 - Bezos, Prometheus, and the General Engineer

* 10:11 - When the Liability Shield Cracks

* 10:59 - Munich Holds Google Liable for AI Overviews

* 13:58 - The xAI Grok Safety Lawsuit

* 15:37 - Washington Pulls Fable 5 and Mythos 5

* 18:00 - A Trillion-Dollar IPO Meets a Kill Switch

Sources

* Anthropic statement on suspending Fable 5 and Mythos 5

* Loop Engineering, by Addy Osmani

* The New Stack on loop engineering

* Dify

* KPMG and Microsoft scale Agent 365 globally

* Coinbase debuts an AI agent that can trade and pay for research

* Jeff Bezos' Prometheus raises 12 billion

* Munich court holds Google liable for false AI Overview answers

* Fired xAI engineer sues over Grok safety warnings

* OpenAI files confidentially for an IPO