Daily AI Briefing — June 25, 2026

OpenAI’s custom inference chip, agents at work, Anthropic’s pull on Google talent, and the AI infrastructure race.

DDiego Varela|25 jun 2026|3 min de lectura
Daily AI Briefing — June 25, 2026
Daily AI Briefing cover

Cover photo by Taylor Vick on Unsplash.

A short daily audio briefing for Diego Varela on the AI news that actually matters: OpenAI’s custom inference chip, agent workflows, Anthropic talent pull, and the infrastructure race.

Audio: generated for Telegram delivery. Local archive path: /Users/diegovarela/.hermes/hermes-agent/daily_ai_briefing_2026-06-25.mp3

Headlines

  • OpenAI and Broadcom unveiled “Jalapeño,” a custom inference chip for running LLMs more efficiently.
  • OpenAI published new research arguing AI agents are moving toward longer, more complex workplace tasks.
  • Reports say senior Google AI researchers Jonas Adler and Alexander Pritzel are leaving for Anthropic.
  • Amazon committed another $13B to AI infrastructure in India as compute buildout remains the industry bottleneck.
  • Figma added AI-assisted code, motion, and shader workflows at Config 2026.

Transcript

Daily AI Briefing for June 25, 2026.

Good morning, Diego. Today’s AI news is less about one shiny chatbot demo and more about the plumbing: chips, agents, budgets, and who controls the talent.

First, OpenAI and Broadcom unveiled Jalapeño, OpenAI’s first custom inference chip for large language models. The key word is inference: this is about making the day-to-day running of models cheaper and more efficient, not just training the next monster model. If it works at scale, OpenAI gets more control over cost, latency, and supply chains. Translation: fewer existential conversations with GPU procurement spreadsheets.

OpenAI also published research on how agents are changing work. The claim is that agents are moving from quick single-turn answers toward longer, messier tasks across roles. That lines up with what enterprises are asking for: not “write me a paragraph,” but “take this workflow, use tools, check your work, and leave an audit trail.” The near-term question is reliability, not magic. Agents that save time are valuable; agents that confidently create cleanup work are interns with better branding.

On the Google side, the notable signal is competitive pressure. TechCrunch and The Decoder both report that senior Google AI researchers Jonas Adler and Alexander Pritzel are leaving for Anthropic. Talent moves are not product launches, but in frontier AI they matter because the research teams are still small enough that individual departures can move roadmaps. It also suggests Anthropic remains an unusually strong magnet for researchers focused on model capability and safety.

Google’s own official AI blog was quieter in the last day, but the infrastructure story continues: earlier this month it announced major data-center investment, including Alabama expansion. That matters because every frontier lab’s product plan is now also an energy, chip, and data-center plan.

Elsewhere, Amazon announced another 13 billion dollars for AI infrastructure in India. That is not just cloud bragging; it is part of the race to put compute closer to big markets and enterprise customers. And Figma’s Config updates added AI-assisted code, motion, and shader workflows, another reminder that AI is being embedded into professional tools rather than arriving as a separate destination.

Bottom line: today’s theme is industrialization. The winners will not only have better models; they will have cheaper inference, disciplined agent workflows, enough data centers, and enough researchers who have not been poached by the lab next door.

Sources