Presented by
Good morning, AI enthusiasts. Uber's C-suite is at war with itself over AI, and the rest of the corporate world is watching. This week, the company's COO said there's no measurable link between AI spend and useful features. The CEO, on the same earnings cycle, called AI staff "employees with superpowers" and slowed hiring.
Both executives looked at the same AI budget overruns and reached opposite conclusions. If one of tech's most data-driven companies can't agree on whether AI is working, what does that say for everyone else trying to answer that question?
In today's recap:
Uber's C-suite splits on AI ROI
Hassabis narrows AGI timeline to 2029
Set up Claude Code as your daily driver
Anthropic details Claude's production sandboxing
4 new AI tools, prompts, and more
ENTERPRISE AI
Uber's C-suite can't agree on AI
Recaply: Uber just surfaced the year's clearest corporate split on AI ROI, with its COO saying there's no link between token spend and better features while the CEO calls AI staff "employees with superpowers."
Key details:
COO Andrew Macdonald told investors Uber burned through its 2026 AI budget ahead of schedule. More AI usage didn't mean the company shipped better products at the same rate, he said.
Uber blew its entire 2026 AI budget early, per Macdonald's earnings call remarks, while still unable to connect that spend to any measurable customer outcomes. That makes Uber the first major tech firm to say this publicly.
CEO Dara Khosrowshahi pushed back. He told a separate call that Uber "underestimated AI in 2025," and that each AI-enabled employee now does the work of several, with the company slowing new hiring as a result, per Times of India.
Both statements came from the same company within 24 hours, with no explanation from Uber on how it plans to reconcile them.
Why it matters: The AI ROI debate just got its most public example. Uber is spending heavily and has wide AI adoption across its engineering org. But it still can't link that spend to customer impact. The CEO and COO looked at the same data and reached opposite conclusions. Both are probably partially right, which makes the measurement problem harder to solve, not easier. This is what enterprise teams will be wrestling with all year.
PRESENTED BY OYSTER
Got questions about hiring globally?
The best person for your next role might not live near your office — or even in the same country.
And more companies are realizing they don’t need to open entities everywhere just to hire great talent globally.
Instead, teams are using EOR to hire internationally faster, stay compliant, and avoid the operational headache of setting up local infrastructure before they’re ready.
That shift is changing how companies think about growth, hiring, and expansion altogether.
Oyster’s EOR helps companies hire, pay, and support employees in 180+ countries while Oyster handles payroll, compliance, taxes, and local employment requirements.
AI RESEARCH
Top AI scientists signal what they don't understand
Recaply: DeepMind's Demis Hassabis just narrowed his AGI timeline to 2029, while Anthropic's Chris Olah told the Vatican that researchers "keep finding things that are mysterious" about the systems they've built themselves.
Key details:
Hassabis made his 2029 claim via Axios. He's known for using a stricter definition of AGI than most tech leaders, which makes his timeline more significant than it looks.
Olah leads AI interpretability at Anthropic. He spoke at the Vatican during the reading of Pope Leo XIV's AI encyclical "Magnifica Humanitas," where he cited evidence of AI self-reflection and said large-scale job loss is a real possibility.
His Vatican talk was a rare moment of candor. Olah said that interpretability work keeps turning up things no one at Anthropic fully understands about their own models.
Both men spoke on May 27, 2026. Two of AI's most credible technical voices, each flagging what they don't know, on the same day.
Why it matters: Hassabis and Olah aren't hype speakers. Hassabis has a stricter view of AGI, so 2029 from him means more than 2029 from a startup founder. Olah built the field of AI interpretability. His "mysterious" admission doesn't usually make it past PR. The fact that both signals came on the same day, from two independent sources, makes the pattern harder to dismiss. When the people closest to the models flag uncertainty, that matters.
GUIDES
Set Up Claude Code as your daily programming partner
Recaply: In this tutorial, you will learn how to configure Claude Code with CLAUDE.md, Skills, and MCPs so it operates as a persistent, project-aware agent rather than a stateless chatbot.
Step-by-step:
Create a
.claude/directory at your project root and add aCLAUDE.mdfile. Keep it short: only include build commands, test steps, and rules Claude can't infer from the code itself. Ask "Would removing this line cause Claude to make a mistake?" and cut everything else.Add
CLAUDE.local.mdat the project root and add it to.gitignore. Use it for your personal workflow notes (local API keys, personal shortcuts, debug flags) that shouldn't be shared with the team.Build a Skill for any workflow you repeat more than twice. Create
.claude/skills/your-skill-name/SKILL.mdwith step-by-step instructions and any reference files the task needs. Claude loads the Skill on demand when you type/your-skill-name.Add team-shared MCPs in
.mcp.jsonat the project root for tools like Slack, Notion, or a database. Runclaude mcp listto see what's installed globally, then add project-specific servers that every teammate should have.Enable plan mode with
Shift+Tabtwice before any complex task. Let Claude read and map the codebase without writing anything, review the plan in your editor withCtrl+G, then approve before Claude writes a single line.
Pro tip: Any time Claude makes a mistake, end your correction with "Update CLAUDE.md so you don't repeat this." Claude Code creator Boris Cherny says Claude is "eerily good" at writing precise rules from its own errors, and this habit compounds faster than any other setup change.
TOGETHER WITH ZAINITH
Done-For-You TikTok Shop Scaling
Zainith Agency is a boutique marketing agency focused exclusively on TikTok Shop.
They’ve helped brands like Momofuku, Obvi, First Day, and Ice Shaker scale TikTok Shop to $15M+ in sales last Q4.
Generate $1M+ yearly revenue for your eCom brand? Claim your free audit below.
TOOLS
Trending AI Tools
🤖 Helio - AI colleagues (Ada and Hopper) that sit in your Slack channels, pick up tickets, and ship work without constant direction
🛒 Capafy - Skill marketplace where builders upload Claude Code and Codex workflows and earn every time someone runs them
🧠 MiniMax M3 - MiniMax's newest multimodal model, releasing with new capabilities across its AI platform
📱 Bonsai Image 4B - PrismML's on-device image generation model at 0.93GB that runs full diffusion on an iPhone, open-weighted under Apache 2.0
NEWS
What Matters in AI Right Now?
ElevenLabs just launched Music v2, its new AI music generation model with improved vocals, instrumentation, and arrangement alongside multilingual support. The release cuts API pricing by 40-50% and adds inpainting controls for targeted edits, with the model trained on licensed data.
AWS just released Strands Agents, an open-source AI agent SDK that lets developers build agents in about 30 lines of code. The framework is model-agnostic, supporting Amazon Bedrock, Anthropic, and OpenAI, and connects to external APIs through a simple @tool decorator.
Anthropic just published details on how it sandboxes Claude in production using gVisor containers and three defense layers. Claude Code in auto mode catches about 83% of overeager behaviors, and Claude Opus 4.7 holds prompt injection to around 0.1% per attempt.
Apple revealed its upcoming AI glasses, featuring a dual-camera system for photos and gesture tracking, no AR display, Siri integration, and four frame styles in acetate. The glasses are planned to ship in 2027, tethered to an iPhone for heavy compute.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
