Presented by
Good morning, AI enthusiasts. The company building the AI that's disrupting jobs just committed $250M to help workers through the disruption it's creating. It's one of the more unusual moments in the history of tech.
The OpenAI Foundation, the nonprofit arm of one of the world's most influential AI labs, is now funding the research and programs designed to manage the economic fallout from its own products. The question isn't whether the gesture is genuine. It's whether $250M is big enough to match the scale of what's coming.
In today's recap:
OpenAI's nonprofit arm funds worker safety net
Biohub releases a free world model of protein biology
Get started with Sesame's four voice agents on iOS
Trajectory raises $15M to build AI that keeps getting smarter
4 new AI tools, prompts, and more
OPENAI
OpenAI Foundation commits $250M for worker safety net
OAI
Recaply: OpenAI Foundation just committed $250M to help workers and economies deal with AI-driven job changes, with Wojciech Zaremba co-signing the post.
Key details:
The plan covers three areas: tracking how AI changes the economy, helping workers through near-term job shifts, and building systems to share AI gains more widely.
Funds go to outside groups through grants and open calls, while the Foundation builds its own team to start new projects.
The Foundation is asking the public what they see in their jobs and towns, saying formal data often misses what's really happening.
First grants arrive later this year, with the Foundation sharing results as it learns what works.
Why it matters: The company building AI that disrupts jobs is now funding the safety net for those jobs. That's a rare move. The $250M is enough to fund real policy research and worker programs. Whether it adds up to change or smart PR depends on where the money lands. That it comes from the nonprofit arm, not OpenAI itself, matters too. It keeps some space between the cash and the commercial push driving the disruption.
PRESENTED BY PULLEY
Cap table management that puts your business first
Managing your cap table doesn’t have to be complex. Pulley simplifies equity management for Founders and CFOs with intuitive workflows, accurate, audit-ready reporting, and predictable pricing—so you can plan and scale without surprises. Onboard in days, not weeks, and rely on responsive, expert support every step of the way.
From issuing grants to 409A valuations or ASC 718 reporting, Pulley gives you the clarity to manage equity, make decisions, and get back to work. Experience a platform built for business owners and finance teams: transparent, reliable, and designed to put your company first.
AI RESEARCH
Biohub releases a free world model of protein biology
Recaply: Mark Zuckerberg and Priscilla Chan’s Biohub just released ESM, a free world model of protein biology made up of three parts: ESMC, trained on 2.8 billion protein sequences; ESMFold2, a protein design engine; and ESM Atlas, a map of 6.8 billion sequences and 1.1 billion structures.
Key details:
ESMFold2 runs a full candidate search in about two days and screens top picks in under a day, cutting the first step of drug discovery from months in the lab to days on a computer.
Biohub tested designs against five cancer and immune targets, with hit rates of 36 to 88 percent for compact binders and 15 to 29 percent for antibody formats, all confirmed in the lab.
Alex Rives, Head of Science at Biohub, said the models learned such an accurate view of biology that designs tested on a computer function as predicted when taken to the lab.
All three models are free to any research group, in what Biohub calls the largest use of AI in protein biology to date.
Why it matters: AI structure prediction got the big headlines with AlphaFold. ESM reaches for something harder: designing proteins that don't exist yet, then proving they work. The cryo-EM images showing a designed protein binding EGFR exactly where the model said it would is a real proof point. For researchers on rare diseases or underfunded targets, this removes one of the biggest early blocks in finding drug candidates.
GUIDES
Get started with Sesame's four personal voice agents on iOS
Recaply: In this tutorial, you will learn how to download and set up Sesame's four personal voice agents on iOS, so you can start having low-latency real-time voice conversations with AI that remembers what you talk about.
Step-by-step:
Download the Sesame app from the App Store (available in 39 countries) and sign up for an account. You may land on a short waitlist during the initial rollout, but most users are being let in quickly.
Choose your first agent from the four available characters: Maya, Miles, Simone, or Charlie. Each has a different voice, personality, and style. Pick the one that fits the kind of conversation you want to have first.
Start a voice session by tapping the microphone and speaking naturally. Sesame is designed for low-latency flow, so it responds quickly and can pivot mid-sentence as it pulls in real-time search results to back up what it says.
Use the on-screen features during your conversation: tap the search cards to see image results, save key takeaways as notes, or switch to text mode if you're somewhere you can't speak out loud.
Enable Incognito mode for sensitive topics you don't want saved. In normal mode, each agent builds its own memory over time, so conversations get more useful the more you use it.
Pro tip: Each of the four agents keeps its own separate memory, so you can use Miles for work topics and Maya for personal ones and they'll stay distinct. This makes them feel less like a single AI and more like four different people you can go to for different things.
TOGETHER WITH STORECLAW
Don't Leave Millions on the Table
Every day without AI, your store falls behind. StoreClaw helps e-commerce sellers automate growth with AI that monitors competitors, optimizes listings, automates marketing, and tracks real profit across Shopify, Amazon, and more. No complex setup or extra hires — just smarter operations, higher conversions, and more revenue.
ENTERPRISE AI
Trajectory raises $15M to build AI that learns from use
Recaply: Trajectory just raised $15M to build a continual learning platform, targeting one of the biggest gaps in AI today: every model stops getting smarter once training ends.
Key details:
Trajectory logs every time a company's AI fails in the real world, like a customer query that gets handed to a human, then uses those failures to post-train a new model as often as every week.
The seed round was led by Conviction at a $115M valuation, with backing from Bessemer, Radical VC, BoxGroup, Google DeepMind chief scientist Jeff Dean, and Stanford professor Fei-Fei Li.
The three co-founders came from top AI labs: CEO Ronak Malde joined Google DeepMind from the $2.4B Windsurf deal, Arjun Karanam built Apple Vision Pro AI, and Michael Elabd worked in Google DeepMind robotics.
Trajectory has 11 people and one paying customer: Decagon, an AI support startup whose post-trained models now beat frontier models on its core tasks.
Why it matters: Most AI teams ship a model and watch it fall behind. Trajectory's bet is that coding showed the template. Tools like Cursor got much better by training on real user data, and that loop is a big reason they pulled ahead so fast. The hard part is that coding is easy to verify. Code runs or it doesn't. Whether this works in fields where success is fuzzier is the real test Trajectory has to pass.
TOOLS
Trending AI Tools
🎤 Sesame - Four personal voice agents on iOS with low-latency real-time conversations, search, notes, and per-agent memory
🔬 Biohub ESM - A free world model of protein biology covering 6.8B sequences, with structure prediction and protein binder design tools for researchers
🤖 Trajectory - A continual learning platform that post-trains AI models on real-world failures weekly, helping companies build AI that improves with use
⚙️ Google Coral Board - Google's newest edge AI dev board supporting on-device speech translation, natural language hardware control, and vision-based music generation
NEWS
What Matters in AI Right Now?
Cognition revealed that Devin hit more than 10x enterprise growth since January, with run-rate revenue now at $492M, and also closed a $1B+ round at a $26B valuation, led by Lux Capital, General Catalyst, and 8VC.
Sesame launched an iOS preview of its personal agents app with four voice agents now in the App Store, including low-latency real-time search, notes, text input, and a deep-dive mode for in-depth topics.
Robinhood launched Agentic Trading and an Agentic Credit Card, letting AI agents trade equities and make purchases on a user's behalf, with spending caps, a live activity feed, and a virtual card that earns 3% cash back.
OpenAI announced GPT-5.2 and GPT-5.3-Codex will be cut from Codex on June 2 for ChatGPT users, with GPT-5.5 becoming the new default free-tier model while both older models stay on the API.
Anthropic pushed reliability fixes to Claude Code after tracing recent quality issues to three separate changes: a reasoning downgrade, a caching bug that made Claude lose context, and a verbosity change that hurt coding output, all fixed as of April 20.
YouTube launched a custom feed feature for signed-in US users, letting them build a personal content channel from a text prompt, pinned as a chip at the top of their Home page.
Google debuted the newest Coral Board, an edge AI dev board for on-device use cases including speech translation, natural language hardware control, and vision-driven music output.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
