{{first_name | Hi}}, your strategic AI update is here.
Researchers used Anthropic's Claude to break into OpenAI this week, the House shelved its only AI bill, and Dreamforce said last year's models are plenty. EY asked 202 boards and executives: 98% have an AI governance policy, 47% already bypassed it for a deployment that couldn't wait.
Innovation Snapshot
"Organizations are applying yesterday's governance rules to today's interactions with AI."
— Richard Jackson, EY Americas Assurance AI Leader
In today's lineup
Robots
Catch the agent nobody approved
Real ROI from Gen AI
Map what every agent actually touches
People
Raj Rajamani is teaching manufacturers what their agents are actually doing
Love
Before it was a harness, it was a sandworm
The Halo Effect by Phil Rosenzweig
Reading time: 3 min.
ROBOTS 🤖
How are robotics and AI changing industries? We break down the latest news, tools, and innovations for you.
Top Insights
Researchers broke into OpenAI using Anthropic's Claude in a few days of agent time and hours of human time, for a $6,500 bounty.
House GOP leaders shelved the one plan to regulate AI, so the governance policy your board signs is the only one coming.
Dreamforce attendees say last year's models are enough, and a slower frontier would help them absorb what they already bought.
DealBook names 'liability exposure' as AI's new investor risk, and it lands on whoever deployed the model, not only whoever built it.
AI trade rebounds on unusually bullish options bets as the Nasdaq-100 posts its best day in six weeks, one day after the Fed hike.
Catch the agent nobody approved
EY surveyed 202 board members and C-suite AI decision-makers at U.S. companies with $1B or more in revenue for its AI Risk and Governance Survey, published this week. 98% have a formal AI governance policy. 47% have bypassed it at least once, for a deployment that couldn't wait.
The gap widens with agents. 91% now run agentic AI through a pilot or a full deployment. 85% admit those agents take action without a human watching in real time. 26% cannot say, today, whether an unauthorized agent is running somewhere inside their own company.
What the disciplined minority do differently:
Update the governance policy for what agents actually do, not what last year's software did
Build a way to catch an agent nobody approved
Treat a bypassed policy as a reason to fix the policy, not the deployment
47% have already proven their own policy doesn't hold. The other 53% just haven't found out yet.
How is your AI governance policy actually enforced?
Real ROI from Gen AI
How human-centric AI delivers business results
Sachsenmilch, a dairy plant processing 4.7 million liters a day, used Siemens' Senseye AI to catch a failing pump early, saving low six figures in downtime.
CJ Logistics America put OneTrack's AiOn agents into daily operations across 40-plus warehouses, cutting clock-in gap time 45% and lifting units per hour 18% within the first weeks.
A top-10 U.S. hospital runs 24% of all patient payments through Grace, Intelligent Contacts' AI collections agent, two months after go-live, with payments collected up 63%.
Tool Spotlight
Your weekly briefing on tools that create competitive leverage
Map what every agent actually touches
Maps every AI agent as a live blueprint: what it calls, what data it touches, what permissions it holds, and flags the moment it strays. $34M seed led by Redpoint, with CrowdStrike's Falcon Fund. Best for CIOs who need an honest answer before the next audit.
Runs customizable compliance agents against specific frameworks, HIPAA, ISO, NIST, instead of one generic policy. $20M Series A led by GV, 30-plus new agents planned for 2026. Best for compliance leaders who need the framework to match the agent.
Three more proving the policy holds
PEOPLE 👥
Meet the innovators turning bold ideas into real-world impact.
Transformation Champion
Raj Rajamani is teaching manufacturers what their agents are actually doing
Rajamani ran product at CrowdStrike and SentinelOne before co-founding JetStream Security with three other security veterans. JetStream shows what each AI agent is doing right now, on a factory floor or in a warehouse, not what it was approved to do at setup. It closed a $34M seed in March 2026 led by Redpoint Ventures, with CrowdStrike CEO George Kurtz and Wiz CEO Assaf Rappaport among the angels.
LOVE ❤️
Practical wisdom, growth tactics, and a must-read book that will challenge the way you think.
Before it was a harness, it was a sandworm
In 2023 I started opening talks with the sandworms from Dune. This is what AGI is going to feel like, I'd say. Huge, indifferent to you, coming whether you're ready or not. Next slide: a Fremen rider with hooks. We need to be this guy.
We didn't have the word "harness" yet. Nobody remembers a slide about APIs. Two years later the whole industry landed on the same word.
Three rules from that deck still hold. Name the thing before you try to control it. Nobody pays for the worm, they pay for the ride. A person picks the direction, every time. All 12 slides are in the piece.
Transformative Reads
One book, handpicked from my conversations with friends, industry leaders, and tech innovators:
The Halo Effect by Phil Rosenzweig. One good quarter makes every decision look brilliant. One bad quarter makes the same decisions look reckless. Nothing about the decisions changed. The 98% with a governance policy on paper get the same treatment the day someone checks. Perfect for: CFOs and CIOs who need to know what's working before the audit does.
In Culture
The line is for the label. Every agent we put in production still has to walk through the boring door first.
THANK YOU!
Lighthouse takes us many hours to create, and we don’t run ads or charge anything. We are dedicated to making it the best AI transformation newsletter possible.
We just ask for one thing: tell one friend about Lighthouse today!
Want daily AI & Innovation insights? Connect with me on LinkedIn or X.
Got suggestions? Just reply to this email!
Much Love,
Matt and the Future Works team
