Good morning. It's Monday, April 20, and we're covering the metric shaping the AI boom, Dorsey’s AI-led workforce cuts, the push toward iris-scan digital IDs, and more.

Tinder, Zoom Add Eye-Scan IDs: Apps partner with Sam Altman’s World to verify users via iris scans, issuing anonymous World IDs to combat AI bots and deepfake scams amid rising online fraud.

Anthropic Launches Claude Design Tool: The research-preview product uses Claude Opus 4.7 to generate and refine prototypes, slides, and assets collaboratively, expanding AI’s role in end-to-end design workflows.

Cursor Eyes $2B Funding Round: Sources say the AI coding startup is nearing a raise led by Thrive and Andreessen Horowitz at a $50 billion valuation, as revenue targets exceed $6 billion in 2026.

OpenAI Strikes $20B Cerebras Deal (Paywall): The company will spend over $20 billion on AI chips and data centers while gaining equity, aiming to cut NVIDIA reliance and secure massive compute capacity.

Sandra Bullock Urges AI Adoption: She called on Hollywood to embrace AI creatively but cautiously, as industry debates intensify over misuse risks, labor protections, and evolving production tools.

Musk Proposes AI-Driven Universal Income: Elon Musk argues government-issued payments could offset AI-driven job loss, claiming surging productivity from robotics would outpace money supply growth and avoid inflation.

Cerebras Files for U.S. IPO: The NVIDIA rival disclosed April 17, 2026 plans to list on Nasdaq as revenue hit $510 million, signaling renewed IPO momentum amid booming demand for AI chips.

Salesforce has designed a four-layer architecture for companies that want to operate at agentic scale. Slack handles human-agent collaboration. Agentforce builds and deploys agents across every channel. Customer 360 brings decades of business logic to bear. And Data 360, Informatica, and MuleSoft ground every agent in unified, real-time context. It's not a patchwork of AI tools — it's a coherent system built to drive real business impact.

The agentic enterprise isn't coming. It's here. Explore Agentforce.

A nonprofit nobody had heard of two years ago is now setting the tempo for hundreds of billions in AI infrastructure spending — and its own researchers aren't sure what their chart means. METR, a 30-person safety org based in Berkeley, publishes the "time-horizon" chart: a measure of how long an AI agent can work on a software engineering task before getting stuck.

The chart's reach has outrun its precision. METR also published a randomized trial last year showing AI coding tools made developers 19 percent slower — a finding partially reversed in a follow-up now pegging the real effect as a modest speedup. The methodology has drawn pointed criticism, the benchmark covers only software engineering tasks, and METR's own researchers put the odds of an intelligence explosion beginning this year somewhere between less than 1 percent and 10 percent. → Read the full article here. (Paywall)

Most tech executives obscure the reason behind layoffs. Block CEO Jack Dorsey is doing the opposite — describing in precise detail the exercise that led him to cut 40% of his 10,000-person workforce. Over the 2025 holiday break, Dorsey and his leadership team experimented with models including Anthropic's Opus 4.6 and OpenAI's Codex 5.3, reconvened in January, and concluded the company "would not look the same, or be the same size" going forward.

The cuts affect more than 4,000 employees at the $41 billion fintech company, with Dorsey and Sequoia partner Roelof Botha framing the shift as a fundamental break from conventional management hierarchy. Dorsey said they built in a headcount buffer for "mistakes" — and acknowledged they made some. The decision moved from "expiration to execution" in three weeks. → Read the full article here.

Salesforce's four-layer architecture gives companies everything they need to operate at agentic scale — Slack for human-agent collaboration, Agentforce for deployment, Customer 360 for business logic, and Data 360 plus MuleSoft for unified context. Not a patchwork. A coherent system.

The agentic enterprise is here. Explore Agentforce.

Tech companies and researchers, including those tied to Tesla and DoorDash, are paying gig workers up to $25/hour to record themselves doing household chores like folding laundry, according to an April 17, 2026 report. The videos—often captured via head-mounted smartphones—are used to train AI models that can guide robots through complex physical tasks.

Unlike chatbots, which learn from abundant online text, robotics lacks large-scale, high-quality training data, making these recordings especially valuable. Researchers are betting that more data will improve robot performance, but timelines for capable home robots remain uncertain, ranging from a few years to decades. → Read the full article here.

Researchers at MIT Lincoln Laboratory are developing systems that enable divers and autonomous underwater vehicles (AUVs) to collaborate on complex maritime tasks, according to an April 14, 2026 report. The project combines AI-driven perception, navigation algorithms, and new hardware to merge human dexterity and judgment with robotic speed, endurance, and data processing.

Key challenges include limited underwater visibility, scarce labeled sonar data, and low-bandwidth communication that can take minutes to transmit images. Early field tests in U.S. coastal waters and the Great Lakes show progress, though real-time collaboration remains limited. The work targets applications such as infrastructure repair, search and rescue, and military operations. → Read the full article here.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team