MLB just banned AI from calling pitches

☎️ Your AI can talk. Can it actually act?

Ask your AI voice agent to change a billing address right now. If it hands off to a human, your AI can talk, but it can't act. One Invisible deployment fixed that gap for a 75-year-old, multi-billion dollar brand, watching containment jump from 23 percent to 71 percent, with customer satisfaction unchanged.

Read the full report on the difference between AI-enabled and AI-native.

‍

DOWN TO BUSINESS

📊 Cisco: AI drafts most of its SEC filing language

Cisco's AI system now drafts 80 to 90 percent of the MD&A section in its SEC filings, a task once handled entirely by finance teams and outside counsel. The tool cross-references prior quarters and flags inconsistencies before a human reviews the language. Cisco plans to extend the same approach toward giving every employee a personal AI assistant.

🎾 Wimbledon: months of website work, done in a month

Wimbledon used an IBM AI tool called Bob to migrate more than 15,000 digital assets onto its new website and app, work that would traditionally take five specialists months, completed by one engineer in a month, with the final transfer taking 47 minutes. The tournament also processes 2.7 million data points per match through a hidden hub nicknamed Court 19.

🏦 Bank of America: AI helps cut efficiency ratio to 59%

Bank of America's efficiency ratio improved to 59 percent this quarter, down from 63 percent a year ago, a gain CFO Alastair Borthwick credits partly to AI tools now used by more than 200,000 employees. The internal assistant Erica for Employees triages routine requests so staff spend more time on exceptions and higher-value work.

🎬 Netflix: AI cuts documentary production time in half

Netflix used AI to produce 17 minutes of footage for a documentary twice as fast and at half the cost, co-CEO Ted Sarandos told investors. Creative teams used the tools mainly in post-production, enhancing crowd scenes and historical battle sequences productions couldn't otherwise have afforded on their timelines. GenAI workflows have now touched 300 Netflix titles this year.

FROM THE EDGE

🎨 Léonard Boussioux: AI makes you sharper, your team less original

Across four studies — short stories, sustainability pitches, humor, and collaborative fiction — AI improved individual creative output, especially for weaker performers, but narrowed the range of ideas groups produced, MIT Sloan Management Review reports. Teams converged on similar solutions when AI entered at idea generation. Reserving AI for narrowing options, not generating them, preserved variety close to fully human teams.

💹 Ralph Koijen: AI doubles how well we explain stock prices

Agentic AI systems reading earnings-call transcripts now explain nearly 20 percent of stock price movement around earnings announcements, up from 8 percent for standard models, University of Chicago researchers found. The systems extract structured signals from the call itself, and the reasoning stays human-readable rather than a black box. Researchers say the approach reveals mechanisms markets have been missing entirely.

💼 Jim Doucette: AI raises the bar for new hires

A new study of the three sectors that recruit the most MBA graduates finds generative AI is raising, not lowering, expectations for new hires. Winning professionals pair domain expertise with AI fluency, judgment, and the ability to synthesize knowledge across functions rather than staying inside a narrow specialty. One bank executive said the shift happened in the last three years.

HOT MODEL NEWS

🤝 Anthropic launches $1.5B AI implementation firm with Blackstone

Anthropic, Blackstone, Goldman Sachs, and Hellman & Friedman have launched Ode, a $1.5 billion company built to embed Claude across private-equity-owned mid-market firms. The venture pairs Anthropic's models with hands-on implementation teams for companies too small to build in-house AI but too complex for off-the-shelf tools. It follows a wave of frontier labs building dedicated deployment arms.

💵 OpenAI CFO proposes scrapping per-token AI pricing

OpenAI CFO Sarah Friar has proposed a new metric, useful intelligence per dollar, to replace per-token pricing as the way enterprises evaluate AI cost. The current model rewards heavier token usage regardless of whether it produces a better outcome, misaligning vendor incentives with what buyers actually want. Outcome-based pricing would let COOs compare AI spend like any other productivity investment.

🔬 Researchers boost AI inference speed 100-fold

Researchers at Peking University have built an all-optical interconnect system that links standard electronic chips to speed up distributed AI inference by more than 100 times, using about one-ninth the computational resources of conventional setups. The system pairs optical communication with algorithms designed specifically to exploit it, rather than adapting existing methods. Scaling it to commercial systems remains unproven.

🧩 Thinking Machines ships open-weight model built to be customized

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, released its first model, Inkling, a fully open-weight system enterprises can download and fine-tune for their own workflows. Rather than chasing frontier benchmark scores, the company built Inkling specifically as a customizable base other organizations can adapt. A lighter version, Inkling-Small, targets lower-cost, lower-latency deployments for simpler tasks.

PLOT TWIST

⚾ MLB bans dugout AI after Mets couldn't stop bragging

Major League Baseball has blocked teams from running AI on dugout iPads, after custom apps expanded beyond video review into recommending pitch calls and substitutions. Former reliever Adam Ottavino said the Mets spent big money on an AI pitch-predictor, then couldn't stop bragging about it. The restriction took effect this week. Their offense ranks 25th in the league.

🎙️ Voice AI nails sounding human, fails at listening

Hume AI, a voice AI company, benchmarked more than 40 leading voice models against a million human ratings and found they've mastered sounding human while remaining terrible at actually listening to one. In one test, a voice agent confirming a suspicious bank transaction couldn't tell a confident "yes" from a hostage-negotiation "yes."

✋ AI brain implant lets paralyzed man feel again

Keith Thomas, paralyzed from the chest down, can now grasp, lift, and feel objects in his hand again through a brain-computer interface pairing AI with electrical stimulation of his spinal cord and brain. Researchers call it a double neural bypass, which sounds like science fiction because it basically is. Some of Thomas's gains held for months after the stimulation stopped.

🥡 DoorDash lets AI order takeout from your command line

DoorDash opened a limited beta of a command-line tool that lets developers order food and groceries through an AI agent, no app required, because apparently that's the future now. A demo showed the tool reading Slack messages and running Python scripts just to order three salads, narrating itself "Flibbertigibbeting" along the way. DoorDash calls it early agentic commerce.