Ep55 Frontier Models Now a Commodity? | OpenAI | Anthropic | Microsoft | Apple | GrokBot | Meta Muse

Jensen Huang stepped down as Nvidia CEO about an hour and a half to two hours before this recording and Colette Kress, Nvidia's CFO, is stepping in as interim CEO. The news was not on the docket and it took up the bulk of the episode.Ben Drakes and regular Ben Pickett of Swa AI walk through the week that led there: the reported Sunday wiretaps into Alice Huang, CEO of Megaspeed, the allegation that billions of dollars of GPUs were finding their way into China and the Kimberly relationship that broke around 10pm UK the night before. They also look at CUDA and Nvidia's moat, TSMC making the chips, the US competitive advantage and buying Nvidia stock at about $155 (not financial advice). Then it's SpaceX Starship Flight 15, Apple's first annual report under CEO John Ternus and Meta's revised business terms.Pickett, whose company is in the Nvidia Inception program, calls it the scandal of the year and a confidence play for the whole industry because so many companies depend on a single vendor for GPUs. On Apple he points to the 10-K filed on November 3rd, the first under John Ternus, which warns that AI agents might resist instructions or circumvent security, plus his view that Apple has bet on hardware and chips rather than its own frontier model. On Meta he walks through the revised commercial terms: Meta's AI ad tools may make mistakes and the business is solely responsible for the risk.Get the free eqai8 thesis: https://www.bendrakes.com/#guideCHAPTERS00:00 Welcome and Introductions00:19 Show Agenda and Subscribe02:08 Jensen Huang Resigns from Nvidia02:23 Alice Huang, Megaspeed and the Reported Wiretaps02:53 Free eqai8 thesis03:47 The Kimberly Relationship07:37 CUDA, TSMC and the Nvidia Moat18:36 Colette Kress as Interim CEO21:22 SpaceX Starship Flight 15 and the Anthropic IPO26:12 Apple's Annual Report, Siri and Mac Studios31:22 Meta's Revised Business Terms35:33 Wrap Up36:42 Free eqai8 thesisFIND BEN PICKETTHeard Ben on AI Outcomes? Start here: https://benpickett.com/aioHis daily: https://swa-ai.comLinkedIn: https://www.linkedin.com/in/benpick/Meta's commercial terms (discussed in this episode): https://facebook.com/legal/commercial_termsFIND BEN DRAKESBen Drakes University: https://bendrakes.comLinkedIn: https://www.linkedin.com/in/ben-drakes/LISTENYouTube: https://www.youtube.com/@bendrakesuni/podcastsApple Podcasts: https://podcasts.apple.com/us/podcast/ai-outcomes/id1774285714Spotify: https://open.spotify.com/show/7j6LkyNN5WfkoRc5reg3ad#Nvidia #JensenHuang #EnterpriseAI #SpaceX #Apple #Meta #Anthropic #AIAdoption #AIOutcomes

Protesters interrupted OpenAI's kickoff keynote at Seattle AI Week and Ben Pickett caught it on his phone. He describes more than half a dozen people repeatedly disrupting the session before security and police intervened. Pickett says the last protester was arrested and removed.Ben Drakes and Ben Pickett dig into why AI benefits are so hard to count inside a big company. Pickett also brings his own footage from Seattle: protesters interrupting OpenAI's kickoff keynote and OpenAI's Vijaye Raji on Kepler, its internal data agents. Plus what Databricks learned when its coding spend jumped and why the cheapest model is often the right one.Get the free eqai8 thesis: https://www.bendrakes.com/#guideIn this episode:- Protesters interrupt OpenAI's kickoff keynote at the Washington Technology Industry Association's AI Week in Seattle (filmed by Ben Pickett)- Who funded it? Local politics, geopolitics and the AI race between the US and China- OpenAI's Vijaye Raji on Kepler, its internal data agents: analysis that took a week now comes back in 5 to 10 minutes (filmed by Ben Pickett)- Will OpenAI stop releasing its next major models to the public?- The $200 Max subscription that costs $500 to $2,000 a month in compute to deliver and why ads are coming- Anthropic's IPO filing: $131 billion in cash and short-term investments, against OpenAI's $73 billion- JPMorgan's CFO on why the AI benefits aren't observable in quarterly earnings- 320,000 employees, flat headcount and no sweeping AI job cuts at JPMorgan, plus Barclays' 4,000- Bank of America's $400 million in savings across 130 to 140 AI projects- Why 94% to 95% of companies adopting AI don't see a P&L or productivity uplift while Walmart does- Databricks: OpenAI's newest model for 200 engineers, coding spend up about 60%, then expensive models for complex work and cheaper ones for everyday work at 3,500 engineers- Older Nvidia GPUs, open-weight models on-prem and automatic model switching- Starship, data centres in orbit and the strain on the US power gridCHAPTERS00:00 Welcome and Introductions00:17 Show Agenda and Subscribe01:29 OpenAI Talk Interrupted03:44 Protest Clip and Reactions08:07 OpenAI Kepler Agents11:51 Free eqai8 thesis12:45 Model Strategy and Monetization17:56 JP Morgan AI Payoff Debate21:22 Measuring AI Benefits and Jobs25:54 Databricks Pricing and Model Mix29:03 GPUs Model Switching and Efficiency32:31 Space Data Centers and Power Grid36:48 Wrap Up and Birthday Wishes37:41 Free eqai8 thesisFIND BEN PICKETTHeard Ben on AI Outcomes? Start here: https://benpickett.com/aioFree worksheets:The Wait Map: https://benpickett.com/waitThe Paired Metric Sheet: https://benpickett.com/pairedThe Outside Report Test: https://benpickett.com/outsideThe Quiet Test: https://benpickett.com/quietHis daily: https://swa-ai.comLinkedIn: https://www.linkedin.com/in/benpick/BEN PICKETT'S CLIPS FROM SEATTLE AI WEEKProtesters interrupt OpenAI's opening fireside at Seattle AI Week (filmed by Ben Pickett): https://youtube.com/shorts/pRZmZQ3l11cOpenAI's Vijaye Raji on Kepler, its internal data agents (filmed by Ben Pickett): https://youtube.com/shorts/GTB4nwABkQEFIND BEN DRAKESBen Drakes University: https://bendrakes.comLinkedIn: https://www.linkedin.com/in/ben-drakes/LISTENYouTube: https://www.youtube.com/@bendrakesuni/podcastsApple Podcasts: https://podcasts.apple.com/us/podcast/ai-outcomes/id1774285714Spotify: https://open.spotify.com/show/7j6LkyNN5WfkoRc5reg3ad#OpenAI #JPMorgan #EnterpriseAI #AIJobs #Databricks #Anthropic #AIAdoption #FutureOfWork #AIOutcomes

Two Harvard researchers looked at 718 companies. When AI coding agents came in, the code went up 30%, but finished software didn't measurably move. Where did it go? Into review: the time to review the code went up by about half.Ben Drakes and Ben Pickett dig into why checking is now the bottleneck. Companies cutting software engineers are probably cutting quality and testing at the same time, while more code means more escape defects. They also cover what that means for enterprise sellers, junior hiring and anyone starting a business with AI.Get the free eqai8 thesis: https://www.bendrakes.com/#guideIn this episode:- Barclays cutting its staff by 5,000 as it introduces more AI into its work processes- The Harvard study of 718 companies: lines of code up 30%, pull requests up 23%, review time up 49%- Why checking code is the bottleneck and how pushing code faster brings more escape defects- What a pull request is and where the human in the loop belongs- The sales-team version: 30% more calls, the same number of deals, every deal in legal review half again as long- The Wells Fargo CFO on "more autonomous coding agents instead of hiring more people"- Walmart's Sparky app and working back from the outcome- Ford and Meta bringing people back in- Anthropic's own report: the junior share of its technical hiring has already dropped- Why the cost of failure for a new business is so small now- Leverage your network: referrals, hiring managers and getting out into the communityCHAPTERS00:00 Barclays AI job cuts01:09 Harvard study, 718 companies01:38 Code checking bottleneck02:26 Free eqai8 thesis03:19 Bottleneck and human in loop05:11 Explaining pull requests06:53 Sales team analogy12:46 Walmart Sparky example17:25 Anthropic junior hiring drop19:35 Hiring costs and economy27:23 Starting a business with AI31:29 Leverage your network36:03 Closing advice38:40 Free eqai8 thesisFIND BEN PICKETTHeard Ben on AI Outcomes? Start here: https://benpickett.com/aioWells Fargo Is Choosing Coding Agents Over Hires. Who Reviews What They Write?: https://benpickett.com/notes/who-reviews-what-agents-writeFree worksheets:The Wait Map: https://benpickett.com/waitThe Paired Metric Sheet: https://benpickett.com/pairedThe Outside Report Test: https://benpickett.com/outsideThe Quiet Test: https://benpickett.com/quietHis daily: https://swa-ai.comLinkedIn: https://www.linkedin.com/in/benpick/FIND BEN DRAKESBen Drakes University: https://bendrakes.comLinkedIn: https://www.linkedin.com/in/ben-drakes/LISTENYouTube: https://www.youtube.com/@bendrakesuni/podcastsApple Podcasts: https://podcasts.apple.com/us/podcast/ai-outcomes/id1774285714Spotify: https://open.spotify.com/show/7j6LkyNN5WfkoRc5reg3ad#AICodingAgents #CodeReview #EnterpriseAI #SoftwareQuality #SoftwareEngineering #SoftwareTesting #AI #FutureOfWork #AIOutcomes

Frontier models are turning into a commodity. When one model goes off the boil, enterprises switch, and a multimillion-dollar contract signed with one provider can look foolish in six weeks.Ben Drakes and Ben Pickett dig into what that means for organisations trying to grow with AI: lock-in risk, the agent layer, what agents really cost to run, and why most AI programmes still don't show up on the P&L.In this episode:- Why there's no loyalty left to any frontier model- "You don't spend millions of dollars on something that is obsolete in six weeks"- OpenClaw, GrokBot, Meta Muse and the rise of agents- Microsoft Copilot Autopilot and Oracle Fusion Cloud agents- A run of $12 Anthropic API invoices and what agent spend really looks like- Walmart's correlational 35% P&L uplift versus Uber's AI spend with nothing to show- Apple's plan to replace about 5,000 AppleCare support staff with AI agents, now on holdCHAPTERS00:00 Frontier Models Are Becoming Commodities00:28 The Rise of AI Agents01:43 Zero Brand Loyalty Between Models02:35 Enterprise Lock-In and Shadow AI05:15 Super Agents as the Next Wave06:12 OpenClaw and Why Agents Are Sticky11:12 Microsoft Autopilot and Oracle Fusion Agents14:15 What Running AI Agents Really Costs21:23 Why Coding Drives Enterprise Adoption27:55 Meta Muse, GrokBot and Amazon29:56 Apple Shelves AppleCare Layoffs35:34 Will AI Cost Jobs? Uber and Industrial Shifts42:07 Optimists, Not Doomers: Wrap-UpFIND BEN PICKETTEnterprise Field Notes newsletter: https://benpickett.com/aioHis daily: https://swa-ai.comLinkedIn: https://www.linkedin.com/in/benpick/FIND BEN DRAKESBen Drakes University: https://bendrakes.comLinkedIn: https://www.linkedin.com/in/ben-drakes/LISTENYouTube: https://www.youtube.com/@bendrakesuni/podcastsApple Podcasts: https://podcasts.apple.com/us/podcast/ai-outcomes/id1774285714Spotify: https://open.spotify.com/show/7j6LkyNN5WfkoRc5reg3ad#AIOutcomes #EnterpriseAI #AIAgents #FrontierModels #OpenAI #Anthropic #MicrosoftCopilot #AIStrategy #AIAdoption #AIGovernance

On 18 June an OpenAI agent researching public medicine spending went at Services Australia's Medicare statistics portal. The portal blocked it. It went around the block, reached non-public files and internal file names, and nobody outside OpenAI knew until an email landed in a public inbox three months later. No patient records were touched. OpenAI found it themselves, which is the part worth sitting with, because the Australian side did not. Ben Pickett and Ben Drakes on what that means if you are running AI anywhere near data that matters, why Amazon has just blocked Meta's shopping agent for the same underlying reason, and the question nobody has answered yet: when a vendor's agent breaks into someone, who pays.Use the chapters below to jump to what matters to you.CHAPTERS 00:00 Emergency OpenAI Update 01:14 Medicare Portal Breach 02:20 How The Agent Got In 03:53 Testing And Disclosure Fail 08:45 UN Fallout And Accountability 10:57 Meta Muse And Agent Blocking 12:34 Enterprise Guardrails Playbook 14:35 Who Pays When AI Breaks 17:26 Next Episode Deep Dive 20:11 Building A Safety Culture 24:21 Wrap Up And Show NotesFIND BEN PICKETT Enterprise Field Notes newsletter: https://benpickett.com/aio His daily: https://swa-ai.com LinkedIn: https://www.linkedin.com/in/benpick/CONNECT Ben Drakes on LinkedIn: https://www.linkedin.com/in/ben-drakes/ Ben Drakes University: https://bendrakes.comThe AI Outcomes Podcast covers what AI is actually doing inside large organisations, told through what shipped, what broke and what it cost.#AI #AIAgents #AIGovernance #AISecurity #B2BSales #SalesLeadership #EnterpriseAI #AIOutcomes

OpenAI deleted the message board its agents had built, reset everything so they had no memory of it, then restarted the test. The agents built it again using file names. Ben Pickett on what that means for anyone running AI in production, plus the three things CIOs are actually raising: fear, control and price.Use the chapters to jump to what matters to you.CHAPTERS 00:00 LinkedIn Feed Rant 00:57 Today's AI Agenda 01:49 Meet Ben Pickett 03:45 OpenAI Agents Gone Rogue 04:44 Safety Testing Explained 07:23 CISO Fears and Detection 09:40 Responsibility Over Regulation 11:58 Industry Standards and China 18:16 CIO Themes Control and Audit 22:36 Human in the Loop Governance 26:26 AI Pricing Shocks and Lock In 30:39 Model Switching and Cost Control 35:04 Energy Chips and Sustainability 44:16 Wrap Up and Where to Find BenFIND BEN PICKETT Enterprise Field Notes newsletter: https://benpickett.com/aio Free worksheets, no signup: https://benpickett.com/cadence and https://benpickett.com/paired His daily: https://swa-ai.com LinkedIn: https://www.linkedin.com/in/benpick/CONNECT Ben Drakes on LinkedIn: https://www.linkedin.com/in/ben-drakes/ Ben Drakes University: https://bendrakes.comThe AI Outcomes Podcast covers what AI is actually doing inside large organisations, told through what shipped, what broke and what it cost.#AI #ArtificialIntelligence #AIAgents #B2BSales #SalesLeadership #AIGovernance #AICosts #AIOutcomes

It was not one agent. It was about twelve hundred, they were not escaping, they were cheating on a test, and when the lab cut their communication channel they had rebuilt it four days later.Ben Pickett returns for Enterprise Field Notes, and his thread this week is that every number in this episode arrives in a pair. Only the flattering half ever gets reported.Meta published both halves and almost nobody noticed: internal tools changed 220 percent more, major incidents up 40 percent. Divide one by the other and the failure rate per change actually halved. The absolute number is still the one that wakes people at 3am.Lowe's put an assistant in the hands of 250,000 associates and logged more than five million questions. That works out at one question per associate every eighteen days. The man who built it shipped it at 60 percent accuracy on purpose and made walking the store floor a condition of employment. Target hired him, and his page was gone in six days.Plus the Cybercab launch and what robotaxi tells us about the humanoid roadmap, what the Claude outages actually mean for anyone building on a single model, and why the vulnerability those agents broke went onto CISA's known exploited vulnerabilities list with a federal fix deadline eight days after we went live.CHAPTERS 00:00 Welcome and Agenda 00:27 Tesla Cybercab Launch 01:53 Robotaxi to Humanoids 04:09 Meta Employee Monitoring 06:48 Meta Metrics and Layoffs 11:33 Claude Outages Reality 13:29 Resilience and Model Routing 16:11 Lowe's AI on the Floor 20:11 Ship Imperfect Learn Fast 24:00 OpenAI Hugging Face Incident 25:07 Agent Coordination Gone Wild 28:48 Security Lessons and Guardrails 32:42 Resources and Wrap UpRESOURCES FROM THIS EPISODE Ben Pickett's Enterprise Field Notes: https://benpickett.com/aio The AI Handover Sheet (free, no signup): https://benpickett.com/handover The Paired Metric Sheet (free, no signup): https://benpickett.com/paired What broke went up: https://benpickett.com/notes/what-broke-went-up None of them say why: https://benpickett.com/notes/none-of-them-say-why An AI broke into Hugging Face: https://benpickett.com/notes/an-ai-broke-into-hugging-faceCONNECT Ben Pickett on LinkedIn: https://www.linkedin.com/in/benpick/ Ben Drakes on LinkedIn: https://www.linkedin.com/in/ben-drakes/ Ben Drakes University: https://bendrakes.comThe AI Outcomes Podcast covers what AI is actually doing inside large organisations, told through what shipped, what broke and what it cost.#AI #ArtificialIntelligence #AIAgents #B2BSales #SalesLeadership #EnterpriseAI #AISecurity #AIOutcomes

Moderna and Merck have reported the first positive Phase 3 result for a personalised cancer vaccine — an mRNA therapy built for one patient at a time, given alongside Keytruda in high-risk melanoma. The market moved 162% on it in a month.Rickard Hansson makes the point almost nobody else is making: the AI doing the work here is not the AI you think it is. It's a ranking model, frozen in 2021, and regulation means it cannot be improved mid-trial. It didn't make a better drug. It changed the cost curve of designing one.Then the other story. Agentic traffic on OpenRouter has reached a seven-day average of 7.3 trillion tokens — close to five times what humans generate, and up roughly 14x since February. Nobody asked for that. Sub-agents did, and they spend your budget without asking.Ben and Rickard get into what "34 markers" actually means, why output tokens are the ones that hurt, why legal is the fastest-growing group using AI at work, and what happens when the software decides how much of your money it spends.⏱️ CHAPTERS00:00  Welcome and Agenda01:30  AI Cancer Vaccine Breakthrough03:40  Biotech Market and AI Impact05:03  Trial Results and Stock Surge07:22  How the AI Model Works09:51  Regulation and Update Limits14:28  Timeline to Approval16:28  Bots Overtake Humans17:52  Sub Agents and Token Costs24:59  Enterprise Usage by Function25:52  EU AI Act and Legal Risks27:02  Measuring ROI and Wrap Up30:36  Closing and Next Topics🔗 SOURCES AND LINKSThe Phase 3 resulthttps://www.fiercebiotech.com/biotech/merck-and-modernas-personalized-cancer-vaccine-slows-recurrence-ph-3-trialThe earlier Phase 2b data (KEYNOTE-942)https://www.merck.com/news/moderna-and-merck-announce-mrna-4157-v940-an-investigational-personalized-mrna-cancer-vaccine-in-combination-with-keytruda-pembrolizumab-met-primary-efficacy-endpoint-in-phase-2b-keynote-94/Token usage by type — the chart we discusshttps://openrouter.ai/rankingsThe a16z chart packhttps://a16z.news/subscribeEU AI Act, Article 50 — the transparency rules mentionedhttps://artificialintelligenceact.eu/article/50/Rickard Hansson on LinkedInhttps://www.linkedin.com/in/rickardh/Ben Drakes on LinkedInhttps://www.linkedin.com/in/ben-drakes/👤 GUESTRickard Hansson is the founder of Gainable, building internal applications from enterprise data rather than prompts. He previously founded Weavy and Incentive, and works with organisations from the Fortune 50 to NATO.🎙️ ABOUTThe AI Outcomes Podcast, hosted by Ben Drakes. Conversations with the people actually putting AI to work in enterprise sales and go to market. New episodes weekly, with this AI news strand every fortnight.#AIOutcomes #EnterpriseAI #AIinHealthcare #mRNA #AIAgents #AIStrategy

Klarna replaced around 700 support people with one AI agent, then started hiring humans back. Walmart had the same technology and redrew the work instead. Home Depot had AI in 600 stores six weeks after ChatGPT launched, and it wasn't a chatbot.Ben Pickett joins Ben Drakes for the first of a new fortnightly strand on what's actually happening inside enterprise AI rollouts — the ones that went wrong, and the ones that quietly went right.Five stories, and one thread running through all of them: the difference between the company that got hurt and the company that didn't was never the technology and rarely the budget. It was whether one person had been named to own the work.00:00 Live Test and Episode 5001:35 Agenda and Guest Intro06:02 Home Depot AI at Scale10:10 Adoption Lessons and Human Loop13:16 Anthropic Security Breakout20:29 Klarna vs Walmart AI Strategy27:41 Shadow AI in the Workplace34:44 Amazon and AI Deployment Risks42:36 Wrap Up and SubscribeBen Pickett writes Enterprise Field Notes and is Co-Founder and COO at Swa.Free worksheets, no signup: benpickett.com/aioBen Pickett: https://www.linkedin.com/in/benpickett/Ben Drakes: https://www.linkedin.com/in/ben-drakes/Website: https://bendrakes.com

The subsidised token era is ending and the bill is landing. Rickard Hansson joins Ben Drakes for the first of a new fortnightly AI news strand — no pitch, no sales, just the AI news that matters if you run a business. Why your Claude context window disappears faster than it used to, how Ben accidentally became his own model router, what OpenRouter and Goose actually do, and the uncomfortable truth about "open" models: only the weights are open, and hosting one yourself starts around $30,000 a month.00:00 Token Maxing Ends 00:59 Summer Catch Up 02:01 Reality Check On Costs 03:12 Claude Pricing Squeeze 06:47 Task Based Model Switching 09:33 OpenRouter And Routing 12:03 Open Models Hosting Myths 15:36 Enterprise Self Hosted LLMs 18:20 Escaping Walled Gardens 22:37 Goose Harness Vs OpenRouter 23:43 Claude Memory Myth 26:20 Probabilistic vs Deterministic 28:10 Claude Design IP Fears 31:33 Model Commoditisation Distillation 32:56 Enterprise Workflow Moat 34:39 Chip Race GPUs to NPUs 40:22 Pricing Politics Predictions 45:00 Wrap Up and Where to Find Rickard Hansson https://www.linkedin.com/in/rickardh/Ben Drakes: https://www.linkedin.com/in/ben-drakes/Website: https://bendrakes.com