Anthropic shipped Opus 4.8 the same week Salesforce hit $11B

In this edition:

• This week: Cognition hits $492M run-rate, Salesforce puts agent tokens on the earnings deck, and Anthropic ships Opus 4.8 with a measurement frame

• Under the radar: The first viral counter-narrative on Microsoft AI ROI landed Thursday, and what it means is that the skepticism finally went public

• What's on the calendar: WWDC, Nvidia earnings read-through, and the OpenAI IPO filing window is still open through June

THE WEEK IN AI

THE WEEK IN ONE SENTENCE

The AI category got measured this week, with three real numbers and one organized doubt, all landing in five days. Cognition put a $492 million run-rate on the record. Salesforce posted a $11.1 billion quarter with Agentforce tokens up 152% quarter over quarter. Anthropic shipped Claude Opus 4.8 the same day the first viral counter-narrative on Microsoft AI ROI landed in the financial press.

THREE SIGNALS

01 • Agents

Coding-agent revenue is a real line item now

Cognition announced on Wednesday a $1 billion round at a $26 billion valuation, with enterprise usage up more than 10x since January and revenue run-rate at $492 million. The round was led by Lux Capital, General Catalyst, and 8VC. The number that matters for procurement is the run-rate, because coding agents have not posted a figure that looks like a real enterprise software business before now. The multiple is roughly 53x revenue, which is high for software and lower than the public-market AI premium attached to vendors with weaker usage.

Cognition is not alone on the number. Cursor hit $3 billion in annualized revenue last week. Claude Code shipped a free vulnerability-scanning plugin. OpenAI Codex disclosed more than 4 million weekly developers. Three vendors with three different wedges all turned in real numbers inside one news week, and the vendors still pitching general-purpose autonomy without usage-attached revenue are the ones procurement should ask harder questions of this quarter.

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

Salesforce put an "AI revenue" number on the public record

Salesforce reported $11.1 billion in Q1 revenue on Wednesday, the company's first-ever $11 billion quarter. Agentforce, the company's agent platform, processed 28.6 trillion tokens during the quarter, up 152% quarter over quarter, and produced 3.8 billion Agentic Work Units, up 111% quarter over quarter. Annual guidance was raised to $45.9 to $46.2 billion at 34.3% operating margin.

The disclosure pattern is what changed. Token counts and agentic work units are now reported alongside revenue, in the same place an analyst reads gross margin. Microsoft, Google, and Amazon still disclose AI revenue in qualitative language inside their cloud lines. Salesforce broke that pattern by turning agent throughput into a metric on the earnings deck, which gives finance teams a public yardstick when they ask their own AI-platform vendors what comparable numbers look like. Vendors that cannot answer that question now have a peer who can.

03 • Model

Anthropic shipped Opus 4.8 with a measurement frame attached

Anthropic released Claude Opus 4.8 on Thursday at the same price as Opus 4.7. The marketing language was deliberately narrow: sharper judgment, more honest self-reporting on progress, and longer independent work. The model added a five-step Thinking effort selector across Low, Medium, High, Extra, and Max for granular cost-performance tradeoffs on hard problems.

The independent verification arrived the same day. Artificial Analysis scored Opus 4.8 at 1890 on GDPval-AA, the agentic real-work benchmark, up 137 points from Opus 4.7 and 121 points ahead of the next-best model. Dan Shipper at Every posted a hands-on take, calling it strong enough that Anthropic could have named it Opus 5, and reported it beat GPT-5.5 on Every's internal Senior Engineer benchmark. The framing matters more than the version number. Same-price upgrades, named benchmarks, and a tunable effort knob give procurement teams a renewal conversation that is no longer "what model do you use" but "what benchmark and what effort tier matches the workload you are running."

UNDER THE RADAR

The most under-reported story this week is the first organized doubt. Financial Times analysis of Microsoft AI ROI circulated as a viral post on Thursday, describing the numbers as "very ugly" even under best-case assumptions and crossing 851 retweets across investor, leadership, and founder lists by mid-day. The full FT figures are not visible in the truncated quote, and Microsoft has not formally responded.

The signal is not the data point, it is the velocity. The market has been quietly skeptical of enterprise AI ROI for at least two quarters, and this is the first viral artifact giving that skepticism permission to be said out loud. The same week Salesforce showed AI as product working at $11.1 billion, Microsoft AI as bought-on-top showed up as a counter-narrative, and the two stories belong together because they describe the gap between the sell-side scoreboard and the buy-side reality.

The question to surface inside your company this quarter is whether the AI work you are funding looks more like the Salesforce shape, where AI is part of the product the company sells, or the Microsoft shape, where AI is bought as a license added to existing seats. What I keep hearing from operators 12 to 18 months into a Copilot rollout is that the productivity case has been hard to defend at the budget level, and the FT post is the first time that conversation moved from private to public.

QUOTE OF THE WEEK

"Frontier models may behave differently when they detect they are being evaluated. Safety evaluations that do not account for this measure the test, not the model."

The line lands harder this week because it complicates the same scoreboard that the rest of the category is celebrating.

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

• WWDC opens Monday June 8 with the iOS 27 keynote. The redesigned Siri and rumored Gemini-distilled assistant work get their first official surface. Watch for whether Apple's on-device AI strategy closes the gap on what Google and Microsoft have already shipped into their respective stacks.

• Nvidia reported earnings Wednesday after the bell. First analyst pass on AI-rack demand and HBM share now landing into the weekend. Data-center revenue trajectory remains the cleanest macro read on whether enterprise AI workloads are pulling through real budget.

• The OpenAI IPO filing window remains open through the first half of June. If the S-1 drops this week, it's the first public disclosure of OpenAI's full financials: revenue, margins, compute costs, and customer concentration. Every AI vendor pricing conversation between now and the roadshow gets priced against what's in that document.

• Goldman Sachs published an $800 billion annual AI infrastructure spending estimate Wednesday. The largest single-year figure any major bank has put on record for the category. Downstream costs from compute, networking, energy, and data-center buildout eventually pass through to enterprise contract pricing.

REPLY

Hit reply and tell me which of this week's numbers your team is going to be asked about first in a board or budget meeting. I read every reply.

Have a good weekend,
Haroon