Tesla put a driverless taxi on Austin roads this week

In this edition:

• This week: Tesla pulled the human out of the vehicle, Microsoft shipped its own model on its own chip, and Bain put the AI ROI shortfall on the record

• Under the radar: OpenAI admitted it is seeing early signs of recursive self-improvement

• What's on the calendar: Microsoft Build wraps, the May jobs report, Anthropic's IPO window, and WWDC

THE WEEK IN AI

THE WEEK IN ONE SENTENCE

Capability landed in three places this week, and the argument moved to where the value went. Tesla started running an unsupervised commercial robotaxi in Austin with no human in the vehicle. Microsoft shipped its first frontier reasoning model trained from scratch on its own chip. Bain & Company put in print that the AI deployments are working as designed and the cost savings are not arriving on the income statement.

THREE SIGNALS

01 • Platform

Microsoft now operates the whole stack

At Microsoft Build on Tuesday, Mustafa Suleyman's team released MAI-Thinking-1, Microsoft AI's first in-house reasoning model. It is 35 billion active parameters, 256K context, trained from scratch on Microsoft's own data, and it runs on Microsoft's own Maia 200 silicon. Microsoft claims independent raters prefer its outputs to Claude Sonnet 4.6 and that it matches Claude Opus 4.6 on SWE-Bench Pro. It is one of seven first-party MAI releases. MAI-Image-2.5 is live in PowerPoint. MAI-Code-1 is live in GitHub Copilot.

The buyer-side change is structural. Microsoft now owns the silicon, the model, the runtime (Foundry), and the distribution (Microsoft 365 and GitHub). A year ago the CIO question was "how exposed are we if OpenAI prices change?" The same question now has a Microsoft answer, a Microsoft chip, and a Microsoft contract attached. The lock-in question moved up a level, from a model decision to a platform decision.

AI READY PRO · FREE UNTIL FRIDAY

I will get you AI trained in 30 days for free

After several months of running it quietly with top AI operators and teams, I’m excited to launch AI Ready Pro today to our newsletter subscribers like you.

It’s a 30-day personalized AI training program that is personalized to you and where you are in your AI journey. It’s the culmination of 100’s of hours spent teaching AI to top Fortune 500 AI teams and operators (and working with Mark Cuban).

Each day, you receive a 10-15 minute exercise to complete to improve your AI skills.

This isn’t vague AI theory or a way to pitch you a tool. It’s 30 days of learning to use AI in your real work. So you come out the other side with the skills to use AI to actually improve you and your teams output.

Until this Friday (May 15), it’s free for newsletter readers who complete the extended assessment. This assessment will help us personalize the learning experience to you.

If you lead a team, AI Ready Team is open today too, also for readers first. It’s the same engine, but to train your whole organization. You’ll get a detailed view of:

• Your team’s AI readiness level + detailed strengths & weaknesses

• Ranked list of AI automation opportunities

• Get tactical advice on how to advance AI efforts in your org

Leading a team? Take the Team assessment instead.

02 • Adoption

Bain put the ROI shortfall on the record

A Bain & Company report covered by Bloomberg on Monday summarized its findings in one line: "The technology worked. The value didn't arrive." The survey found that most enterprise AI deployments are running as designed, the models are doing what the implementations were built to do, and the cost-reduction numbers boards approved are not landing on the income statement on the schedule promised. The piece picked up more than 450,000 views across the social cuts since the weekend.

The framing is the thing to track. Bain reads the forecasts as too optimistic, not the technology. The shift, if it happens over the next two quarters, will be that named companies start reframing their public AI ROI story from cost reduction toward revenue and quality. I want to be honest that the bear case is real here: a consulting firm telling enterprise buyers their forecasts were wrong is a finding with a built-in audience, and the methodology will matter more than the headline.

03 • Autonomy

Tesla pulled the human out of the vehicle

On Wednesday, Tesla AI Director Ashok Elluswamy confirmed Tesla launched a fully unsupervised commercial Robotaxi service in the greater Austin area. The post crossed 938,000 views and 7,000 likes within the day. There is no human operator in the vehicle, and Day 2 of operations held without incident in live commercial use rather than a closed test loop.

Waymo runs the same kind of service in San Francisco with the same outcome (no driver), and Waymo took most of a decade, plus a per-mile cost structure that depended on a small geofence to get there. Tesla claims that the same outcome is achievable on a different cost curve and across a larger area. The number worth watching over the next two quarters is the per-incident rate, because that is the data that decides whether the regulators in the next two cities follow Austin's permitting pattern or pull the brake.

UNDER THE RADAR

The most under-reported story this week is an OpenAI safety admission. In a public note flagged by Peter Wildeford on Tuesday, OpenAI said it is seeing "early signs of recursive self-improvement in today's systems" and called recursive self-improvement "potentially the most consequential frontier safety issue of the coming decade." The note did not make front pages. Wildeford's repost did most of the visible distribution.

Recursive self-improvement is the scenario where a model meaningfully helps build the next model, and the next model helps build the one after, and the loop compresses the timeline between releases. The reason it matters at the company level, not just the lab level, is that it is the strongest argument against forecasting AI ROI on the cost structure of the model you bought this quarter. If the model in your Q4 contract is meaningfully better at meaningfully lower token cost than the one in your Q1 contract, the business case attached to either is a snapshot, not a plan. The Bain finding earlier in the week is the buy-side version of the same problem.

What I keep hearing from operators 12 to 18 months into a deployment is that the static forecast is the part that ages worst. The OpenAI admission is the first time a major lab has put the dynamic version of that argument in writing.

QUOTE OF THE WEEK

"If your advisor hasn't logged 100+ hours in a modern agentic IDE, stop listening to them."

The line lands in research and lands harder outside of it. The same calibration gap shows up in every layer of decision-making about AI work that has not been done by the person making the decision.

SPONSORED BY CLUTCH

Hire secure AI teammates that work 24/7.

Hire pre-built AI teammates. Give your engineers and operators a platform to ship their own AI apps. Stop losing sleep about what is running where.

Clutch is the platform behind both: pre-built agents for the workflows your ops team should automate first, plus the integration plane your team's vibe-coded apps and Claude Code projects plug into. One platform. Real production. Visible and safe by default.

Built for ops, engineering, and security teams that are tired of the shadow-AI surface area inside their own company.

WHAT’S ON THE CALENDAR

• Microsoft Build wraps Thursday with Agent 365 and Frontier Tuning sessions, where named enterprise customers usually surface. Watch for any cost-cap announcements on Copilot billing following last week's controversy.

• The May jobs report lands Friday morning. First labor read of the post-Q1 AI capex cycle. Watch for any signal on tech employment relative to the AI infrastructure hiring wave.

• Anthropic's confidential S-1 filed last week, with a possible public IPO window opening as early as June 16. If the filing goes public on that timeline, it would be the first frontier-lab public listing on the books and the first disclosure of full financials — revenue, margins, compute costs, and customer concentration.

• WWDC opens Monday June 8 with iOS 27 and the redesigned Siri. The surfaces most likely to move the consumer AI story. Watch for whether Apple's on-device strategy closes the gap on what Google and Microsoft have already shipped.

REPLY

Hit reply and tell me whether your team's AI deployment is being scored on the cost case it was funded on, or on a different value case you've started building. I read every reply.

Have a good weekend,
Haroon