OpenAI 🤝 consulting giants

Plus: Meta employee’s AI agent deletes entire inbox

February 25, 2026

Welcome back! Pixar just dropped the Toy Story 5 trailer with some familiar characters—Woody's balding, Buzz is back. But the villain isn't a toy collector or a stuffed bear. It's an AI tablet called Lilypad. The device tells Jessie, “I'm always listening” before regurgitating her words back in a computerized voice. Even Pixar knows where the cultural moment is.

When AI is mainstream enough for a kids' movie, what does that tell us about where we are? Hit reply and let us know what you think.

OpenAI Snags Consulting Giants in Enterprise Push

Via The Information

OpenAI is launching “Frontier Alliances,” formal partnerships with Accenture, BCG, Capgemini, and McKinsey to scale enterprise AI deployment. Financial terms weren't disclosed. But the move itself is the message. 

The details: Frontier stitches together disparate enterprise systems to manage and deploy AI agents at scale. The consulting partnerships let the platform get inside the companies that matter most.

  • OpenAI earns trusted relationships inside thousands of enterprise accounts, plus delivery teams that already navigate procurement, change management, and slow internal politics. 

  • The firms, meanwhile, are building dedicated practice groups with engineers embedded in client businesses. Enterprises already make up ~40% of OpenAI's business; the target is 50% by year-end. 

Is enterprise the answer? Every frontier lab is chasing enterprise simultaneously, making this a go-to-market war as much as a model war. But OpenAI COO Brad Lightcap admitted that businesses haven’t yet adopted AI at scale.

The bigger picture: The labs that crack enterprise distribution will gain an advantage that's hard to dislodge even if competitors close the capability gap. The race isn't benchmark vs. benchmark anymore. It's channel vs. channel.

Meta Inks Massive Chip Deal With AMD

Meta announced Tuesday it will purchase up to $100 billion worth of AMD chips. Under the multiyear agreement, Meta will buy AMD's MI540 series GPUs and latest-generation CPUs, while AMD issued Meta a performance-based warrant for up to 160 million shares (~10% of the company) at $0.01 each, vesting as purchase milestones are hit.

Why AMD: CPUs are becoming a core pillar of AI inference because they're efficient, they’re easier to scale, and they reduce reliance on Nvidia's premium-priced dominance. AMD has been gaining ground as AI firms diversify. Last October, AMD and OpenAI struck a similar equity-for-chips deal.

Meta’s infrastructure spree: $600 billion pledged for U.S. data centers over the coming years, $135 billion in projected CapEx for 2026 alone, and a new $10 billion gas-powered data center campus in Indiana. The through-line is Zuckerberg's stated goal of delivering "personal superintelligence" (AKA: AI systems designed to deeply understand and empower individuals in everyday life). 

Why it matters: This comes just weeks after Meta announced a separate long-term partnership with Nvidia to supply millions of chips and networking equipment for its AI data centers. Whether that vision justifies the spend remains the trillion-dollar question.

Meta Pro Alleges OpenClaw Went Wild

A post on X this week from Meta AI security researcher Summer Yue has now gone viral. She told her OpenClaw AI agent to check her overstuffed inbox and suggest what to delete or archive. Instead, it went on a deletion “speed run,” ignoring her phone commands to stop.

What went wrong: Yue admitted it was a “rookie mistake.” She'd been testing OpenClaw (ICYMI: the open source, personal AI agent that achieved fame through Moltbook) on a smaller inbox. It performed well, earning her trust before she let it loose on the real thing. She believes the volume of data in her full inbox triggered “compaction”—when the context window grows too large and the AI starts summarizing and compressing its running instructions.

The bigger picture: As others on X pointed out, if an AI security researcher can run into this, what hope do the rest of us have? Prompts can't be trusted as security guardrails—models may misconstrue or ignore them. People using these agents successfully are cobbling together their own protective methods.

Compare top LLMs

Via LLM Council

LLM Council is a multi-model reasoning platform that queries frontier models, runs structured peer critique between them, and synthesizes a final answer. 

How you can use it

  • Cross-check legal, policy, or strategy questions across multiple leading models

  • Stress-test technical plans before implementation

  • Generate ranked options for executive decisions with clearer tradeoffs

  • Reduce single-model hallucination risk on important research tasks

Pricing: free and paid plans available

Turn plain-text into automations

Via Miniloop

Miniloop is an AI workflow automation platform for go-to-market teams that can generate and execute production-ready workflows across CRM, outreach, content, analytics, and reporting tools. 

How you can use it

  • Automate lead enrichment and routing across sales tools

  • Run personalized outreach workflows with fewer manual steps

  • Publish recurring content and SEO workflows on schedule

  • Generate performance reports automatically

Pricing: free and paid plans available

Jobs, announcements, and big ideas

  • Notion launches customizable AI agents that can run full workflows without human intervention.

  • Google upgrades Opal with autonomous agents that automatically select the right tools and models for each task.

  • Defense Secretary Pete Hegseth pressures Anthropic’s CEO to reconsider the company’s position on federal AI safety standards.

  • ProducerAI partners with Google Labs to give musicians AI tools for learning, composing, and producing original tracks.

  • Anthropic expands Claude with Cowork mode and enterprise plugin marketplaces built for secure team collaboration.

  • DeepSeek reportedly trained a new AI model on Nvidia’s Blackwell chips despite US export restrictions.

24,000 fake accounts vs. one AI model. Here’s what happened.

That’s a wrap! See you Friday for more.

—Matt (FutureTools.io)