Hi,
welcome back. This is the second issue, and this week the AI industry crossed a line nobody saw coming a year ago.
Let's dig in.
———————————————————————————
The story this week
Anthropic just passed OpenAI in business adoption. But the lead is fragile.
According to Ramp's May 2026 AI Index, which tracks actual corporate card spending across 50,000+ US businesses, Anthropic now holds 34.4% business adoption, ahead of OpenAI's 32.3%. The crossover happened in April.
The numbers are striking. Anthropic climbed from 0.03% of businesses in June 2023 to 34.4% by April 2026. OpenAI peaked near 36.5% in mid-2025 and has been slowly declining since. (Source: Ramp AI Index, May 2026.)
One product is driving most of this growth: Claude Code. A recent analysis estimated that 4% of all GitHub public commits worldwide are now authored by Claude Code, double the percentage from just one month prior.
The cost signal is real. Uber's CTO revealed the company spent its entire 2026 AI budget in just four months, largely on Claude Code and Cursor, with engineers reporting monthly API costs between $500 and $2,000 per person. (Source: reported at a developer conference, May 2026.)
So Anthropic is winning. But there's a paradox baked into the win. Claude Code works too well.
When a productivity tool becomes so valuable that an organization's $3.4 billion R&D operation can't afford to keep the lights on, cost scrutiny follows. And cost scrutiny pushes enterprises toward cheaper alternatives.
Three things to watch in the coming months.
The first is usage caps. Anthropic just introduced new restrictions for Claude subscribers using third-party agent tools. The era of "unlimited" subscriptions for autonomous workflows is ending.
The second is token economics. Anthropic makes more money when you use bigger models. That misalignment between vendor and customer interests can only grow as agents consume more tokens per task.
The third is OpenAI's response. Codex is being aggressively promoted with free usage incentives, and power users disenchanted with new limits are an obvious target.
The race isn't over. It's just shifting from "best model" to "best unit economics."
———————————————————————————
Three signals worth your time
Microsoft's DELEGATE-52 study shows agents fail beyond 20 hops. Across 52 professional domains, researchers found that top models lost substantial document content or corrupted outputs across extended task chains. Only Python programming consistently met Microsoft's readiness threshold after 20 delegated interactions. The finding that matters: agentic systems equipped with more tools often performed worse, not better. Enterprise enthusiasm for fully autonomous agents is running ahead of current technical reality.
SAP just bet €100M on agents that execute, not assist. At Sapphire 2026, SAP launched its "Autonomous Suite" with 50+ domain-specific Joule AI assistants orchestrating 200+ specialized agents across finance, supply chain, procurement, and HR. The €100M partner fund is meant to accelerate deployment. CEO Christian Klein's key line, quoted at the event: "If AI runs payroll, financial close, or supply chain planning, 80% accuracy is not good enough." Watch for whether SAP can hit the 99.9% bar enterprises actually need.
"AI super-users" save 9 hours per week and earn 3x more promotions. According to Writer's 2026 Enterprise AI Adoption survey, covering 1,200 employees and 1,200 execs, AI super-users save nearly 9 hours weekly, 4.5x more than AI laggards at 2 hours. They were also 3x more likely to receive both a promotion and a pay raise in the past year. 87% of leaders say super-users are at least 5x more productive. The skill gap is becoming a career gap, fast.
———————————————————————————
The tool this week
Beehiiv
This newsletter runs on Beehiiv. After several weeks of building three publications across three languages on it, it's the platform I'd recommend for anyone serious about short-form editorial work.
The reason is the bundle. Most newsletter tools make you choose between good email deliverability, a clean web presence, and built-in growth features. Beehiiv ships all three, plus a recommendations network that brings organic subscribers without ad spend. For a solo operator, that consolidation matters more than any single feature.
If you've been thinking about starting something, this is where I'd start.
Start your own → Beehiiv
———————————————————————————
A thought to carry into the week
The new productivity gap is not about access. It's about depth.
Here's the uncomfortable finding hidden in the Writer survey. AI access is now nearly universal at large companies. 97% of executives say their company has deployed agents. 52% of employees use them.
And yet only about one in ten employees says AI has actually transformed how their work gets done.
Translation: most people are using AI for surface-level tasks like summarizing, drafting, and formatting. The super-users are using it differently. They're rebuilding workflows. They're delegating multi-step processes. They're operating at a different layer.
The gap isn't about who has Claude or ChatGPT. Everyone does. The gap is about who has spent the hours figuring out what to actually delegate, and how to verify it.
If you're feeling like AI hasn't quite landed for you yet, that's fine. The work is in front of you, not behind you.
———————————————————————————
That's it for this week. More signals next Monday.
If a colleague would find this useful, forward it. That's how we grow.
Some links in this newsletter may earn a commission at no extra cost to you.
— AI Quiet Signal