YPO Technology Network AI Brief

This week looked like a fireworks show of AI launches — OpenAI's GPT-5.6, new real-time voice models, Microsoft leaning on its own in-house models. The more important story ran underneath all of it: the AI inside your company has quietly become a black box you can neither see into nor fully trust. Microsoft has begun replacing OpenAI and Anthropic with its own cheaper MAI models inside Excel and O...

The strange truth of AI in 2026 is that the technology keeps clearing bars we thought were years away — Alberta's provincial government just used Claude to scan 466 million lines of code in 20 hours, work that would have taken six and a half years by hand — while the business results stay stubbornly flat. MIT finds 95% of enterprise AI pilots deliver no measurable impact; an NBER survey of more th...

Every boom has a tell, and it is never in the press releases. This week the AI boom's insiders started hedging their own story: Meta announced it will rent out its "excess" AI compute while chipmakers sold off, Oracle's SEC risk factors laid bare the strain of its $300B OpenAI/Stargate commitment, and Mark Zuckerberg told his own employees that AI-agent progress "hasn't really accelerated" as expe...

For two years the question was "how will governments regulate AI?" This month the answer got bigger: the state wants to own a piece, police what the models say, and decide who they may serve. Ownership: OpenAI floated giving the US government a ~$42.6B (5%) equity stake (Alaska-Fund style) and wants Anthropic, Google, and Meta to follow; Altman also called for a US-led "IAEA for AI." The red-line...

An extended, single-thesis episode. For a century the two biggest lines on your P&L — payroll and per-seat software — have been fixed costs sized to peak, sitting there hoping to earn their keep. Stephen Forte's belief: AI turns them into variable costs billed per outcome — per interaction, per order, per resolution. The spine: a fixed cost is a bet on utilization; a variable cost is a bill fo...

The AI stories that get headlines are about models and jobs. The one that hits your P&L first is physical: the buildout ran out of the one thing money can't instantly buy — electricity. The bill is landing: Henrico County, Virginia saw power rates jump 25% overnight because of 37 data centers, with schools asked to conserve — a $5M budget hit. Megawatts, not money: Brookfield 5x'd its Bloom En...

The pink slips are arriving ahead of the product. This week companies cut thousands of jobs and blamed AI — but the technology can't yet do the work those jobs involved. The cuts: British American Tobacco is cutting 9,000 roles; Cisco is cutting while posting record $15.8B revenue; Oracle's filing blames AI for 21,000 cuts. 56% of 2026 layoffs now cite AI. The capability gap: OpenAI's own GeneBenc...

The race to deploy AI agents just outran the controls to manage them. This week three numbers proved it. The breach: Straiker (which raised $64M) found 91% of attacks on production AI agents silently exfiltrate data, and 36% of attacks on coding agents achieve remote code execution. A separate Amazon Q Developer flaw let a booby-trapped repo steal a developer's cloud credentials with no clicks. Th...

The story of the year was supposed to be who controls AI. The real story this week: control and cost split in opposite directions, and your business lives in the gap. The market already switched. US labs fell from 72% to 33% of model traffic on OpenRouter in a year; Chinese models now hold six of the top ten spots. One startup, Lindy, moved 100% of its traffic to DeepSeek. The capability gap close...

For two years, AI was an internal project you rolled out at your own pace. This week, two stories say that era is over: your clients are using AI to grade you, and criminals are using it to rob you. In this episode: Graded by your clients. Thomson Reuters finds roughly $143 billion of professional-services revenue is under active reconsideration, with only 6% of clients satisfied that their provid...

Three deals this week looked unrelated. They are the same deal. A chipmaker bought the software layer, a software giant built its own models, and the biggest model-maker built its own chip — and the strategic logic behind all three is identical. Stephen Forte connects them into one idea: everybody is building their own stack. In this episode: Qualcomm buys Modular (~$3.9B) — why acquiring software...

OpenAI unveiled its first custom chip the same week the market sold off on fears the AI buildout has gone too far. Stephen Forte argues those are the same story told from opposite ends — and that what looks like a bubble is closer to a re-pricing. In this episode: OpenAI's "Jalapeno" chip — built with Broadcom, purpose-made for inference, roughly 50% more cost-efficient than standard AI GPUs in ea...

AI agents just crossed the line from demo to deployment — and that changes what a CEO has to decide this year. The pilot era is ending; the question shifts from "should we try AI" to "how do we deploy agents to everyone, and who supervises them." In this episode, Stephen Forte covers: The deployment proof — Samsung is rolling out ChatGPT Enterprise and OpenAI's Codex coding agent to every employee...

Three second-order effects of the AI buildout are landing on business leaders at the same time — on your people, on what gets built next, and on who's allowed to use any of it. In this episode, Stephen Forte covers: The AI-jobs story flips — Gallup finds tech workers who rarely use AI are about 3x more likely to be laid off (~18% vs 6%), while Forrester says 55% of companies that restructured arou...

The strategic AI question is no longer "which model do we use." It's "where does the model run, and who pays for the tokens." This week the AI inference startup Baseten raised roughly $1.5 billion at up to a $13 billion valuation for the unglamorous business of running other companies' models. Meanwhile token prices are collapsing about 10x a year, and enterprise AI bills are going up anyway. In t...

A weekend deep dive away from the news cycle. The question underneath this week's "who controls AI" headlines isn't the supplier's question — it's yours: if every company on earth can buy the exact same foundation model you can, where does durable advantage actually come from? Efficiency from "using AI" is real but not durable, because everyone gets it. This episode braids three expert frameworks...

The week that asked who controls AI ends by zooming all the way in — to the individual. On June 17, Noam Shazeer, co-author of the 2017 paper that introduced the transformer (the architecture under ChatGPT, Gemini, and Claude) and co-lead of Google's Gemini, announced he is leaving Google for OpenAI — less than two years after Google paid a reported $2.7 billion to bring him back from Character. A...

All week the question was who controls AI. Today it got three answers, and none of them is "the market." First: four days after the largest IPO in history, SpaceX agreed to acquire Anysphere — maker of the AI coding tool Cursor — for $60 billion in stock, folding a leading agentic coding product (reportedly ~$2B in annual recurring revenue) into the same house as xAI's Grok models and Colossus sup...

In a single week, three capitals placed three very different bets on who controls AI. At Bercy, the French government unveiled a "systemic" sovereignty plan: its domestic intelligence service (DGSI) is terminating its contract with US data-analytics giant Palantir in favor of French firm Chapsvision, and a conversational assistant built on Mistral AI is being rolled out to roughly one million civi...

For the first time, the United States has applied export controls to an AI model itself — not a chip. The Department of Commerce is forcing Anthropic to cut off access to its frontier Fable 5 and Mythos 5 models for foreign nationals worldwide, including H-1B visa holders working inside the US, citing the models' ability to autonomously find and exploit software vulnerabilities. Performance no lon...

Anthropic shipped Claude Fable 5 — its first broadly available Mythos-class model — with safeguards that silently degraded responses to suspected distillation attempts, documented only deep in a 319-page system card. Researchers caught it, the backlash landed, and Anthropic reversed within 24 hours: flagged queries now fall back to Claude Opus 4.8, with visible notification. The lesson for executi...

Three capabilities arrived this week and they belong in the same conversation. Visa embedded its global payment network directly into ChatGPT — agents can now check out at any Visa-accepting merchant with tokenized credentials and user-defined controls. Anthropic published "When AI Builds Itself," with internal data showing Anthropic engineers ship 8x as much code per quarter as before, more than...

Anthropic released Claude Fable 5 to the public on Tuesday — the first Mythos-class frontier model with general availability. One million token context, one hundred and twenty-eight thousand max output, reasoning always on, and the long-horizon memory management that makes multi-day work possible. Available day-one on Amazon Bedrock, Snowflake Cortex AI, and Databricks Unity AI Gateway. Free insid...

Two vendor moves landed this week that change how AI shows up on your statement and what tools your team can open. Anthropic split Claude Code billing into interactive seats plus a separately metered Agent SDK credit pool — same playbook Microsoft just ran with GitHub Copilot. Google rewires NotebookLM into a real agent and quietly kills the Workspace AI Ultra Access add-on with a July 7 transitio...

Three institutions reached the same conclusion this weekend — nobody wins at AI alone. Apple opens WWDC today with Tim Cook's final keynote. The headline: a completely rebuilt Siri running on a custom 1.2-trillion-parameter Gemini model licensed from Google at one billion dollars per year. Apple — four hundred billion dollars in cash, forty years of disciplined engineering — concluded it cannot bu...