It's Wednesday, June 10th. Anthropic put its most capable public model, Claude Fable 5, in everyone's hands. Cursor's SDK now lets you hand your agent your own functions as tools and nest subagents as deep as you want. ChatGPT's memory rewrites itself in the background so it stops going stale.
Every Wednesday, we break down the latest AI tools we recommend adding to your stack.
TOOL SPOTLIGHT
🛡️Anthropic Opens Up Its Most Capable Model Yet
Anthropic launched Claude Fable 5 on June 9, a Mythos-class model it has made safe for general use. It tops nearly every benchmark Anthropic tested, from software engineering to vision and scientific research, and the longer the task, the bigger its lead. On risky topics like cybersecurity or biology, it quietly hands the response to Claude Opus 4.8.
Point it at long, multi-step work. In early testing, Stripe used it to run a codebase-wide migration across 50 million lines of Ruby in a day, work it estimated would take a team over two months by hand.
Reach for it on hard analysis and vision tasks, where it sets new highs on finance-reasoning benchmarks and can rebuild a web app's source code from a screenshot alone.
Expect a fallback, not a refusal. On cybersecurity, biology, chemistry, or distillation queries, Claude Opus 4.8 answers instead, which Anthropic says happens in under 5% of sessions.
$10 per million input tokens and $50 per million output tokens, less than half the price of Mythos Preview. Available everywhere today as claude-fable-5, and included on Pro, Max, Team, and Enterprise plans through June 22 before usage credits apply.
The benchmark sweep will grab the headlines, but the fallback is the real move. A frontier model that routes its own riskiest questions to a safer sibling is Anthropic betting that capability and restraint can ship in the same release.
TOOL SPOTLIGHT
🖥️ Cursor's SDK Lets You Hand Agents Your Own Tools
Image from Cursor
Cursor shipped a batch of updates to its TypeScript and Python SDKs on June 4 that make agents easier to run in production scripts and CI. You can now expose your own functions to the agent as tools, route its actions through an auto-review classifier, swap how it stores state, and nest subagents as deep as the work needs.
Pass your own functions through local.customTools and the SDK serves them to the agent over a built-in MCP server, so a function definition replaces standing up your own server. Tools you define once are visible to every subagent in the run.
Turn on local.autoReview to send a headless agent's actions through a classifier you steer in plain language, allowing read-only checks while always pausing destructive calls like deletes.
Let subagents spawn their own subagents to any depth, each keeping its own prompt and model, and tie every run to your logs with a new requestId.
Free to adopt. Run npm install @cursor/sdk or pip install cursor-sdk to upgrade. Scripts pinning the retired composer-2 model route to Composer 2.5 automatically.
Custom tools get the billing, but auto-review is the unlock. A classifier that always pauses a destructive call is what turns a headless agent from a liability into something you can leave running in CI.
TOOL SPOTLIGHT
💬 ChatGPT's Memory Now Rewrites Itself So It Stops Going Stale
Image from ChatGPT
OpenAI rebuilt ChatGPT's memory on June 4 around a background process it calls dreaming. Instead of only jotting down what you explicitly ask it to remember, ChatGPT now synthesizes what matters from your conversations on its own, updates it as time passes, and shows you the result on a reviewable memory page.
Skip re-explaining yourself. ChatGPT keeps your preferences, projects, and constraints current across chats and revises them as facts change, so a trip "next week" becomes a trip you "took in July" once it passes.
Open the memory summary to see, add, or correct what ChatGPT knows about you, and tell it which topics to raise and when.
Keep twice the memory capacity if you are on Plus or Pro, or revert to the old saved-memories system from Settings, Memory, Saved memories.
Rolling out to Plus and Pro users in the US first, with Free and Go plans and more countries following over the coming weeks. Update your app to get it on iOS and Android.
A system that rewrites what it knows about you in the background is useful and a little unnerving in equal measure. The feature that matters isn't the dreaming, it's the reviewable memory page, so open it and see what it thinks it knows.
FROM OUR SPONSORS
💻 Artificial Analysis Coding Agent Benchmarks
Image from Luma
The wait is finally over for developers trying to optimize their AI workflows. At Artificial Analysis, we have tackled the definitive question on every builder's mind: which AI model and harness actually work best together?
As the newly released Coding Agent Benchmarks reveal, the specific combination of model and harness matters far more than the industry expects. Testing uncovered massive performance gaps across different setups, with cost per task varying by more than 30x and execution speed differing by more than 7x. These insights prove that the next major leap in coding agent efficiency won't just come from better models alone, but from the deliberate optimization of the model and its harness together.
Ready to talk shop with the builders pushing these boundaries? Join Artificial Analysis alongside special guests from Cursor, Cognition, and Kernel Labs for an evening of lightning talks, panel discussions, and great food and drinks!
In this section, we feature a few standout opportunities leading AI companies, non-profits, policy groups, and other organizations.
🔬 Work at a Lab/AI Safety
Roles at frontier labs, safety organizations, and research institutions
RESEARCH ENGINEER, AI OBSERVABILITY — Anthropic | San Francisco, CA | $320,000–$405,000
RESEARCH ENGINEER, RL INFRASTRUCTURE & RELIABILITY (KNOWLEDGE WORK) — Anthropic | San Francisco, CA | $350,000–$850,000
RESEARCH ENGINEER, MODEL EVALUATIONS — Anthropic | SF / NYC / Remote-Friendly | $320,000–$485,000
AI SYSTEMS ENGINEER, CODEX CORE AGENTS — OpenAI | San Francisco, CA | Comp not posted
RESEARCH ENGINEER, RETRIEVAL & SEARCH (APPLIED ENGINEERING) — OpenAI | San Francisco, CA | Comp not posted
🚀 Work in Industry
Roles at funded AI startups and private companies
SOFTWARE ENGINEER, AI SDK — Vercel | Hybrid (SF / NYC) | $196,000–$294,000
SOFTWARE ENGINEER, AGENT — Vercel | Hybrid (SF / NYC) | $232,000–$348,000
SOFTWARE ENGINEER, AI PRODUCT — Figma | San Francisco, CA | $153,000–$376,000
SOFTWARE ENGINEER, AI PLATFORMS — Figma | San Francisco, CA | $153,000–$376,000
FOUNDING AI ENGINEER — Letterbook (YC) | San Francisco, CA | $120,000–$180,000 + 1.0–2.5% Equity
🏛️ Work in AI Policy / Governance
Roles in government, think tanks, NGOs, and policy organizations
SENIOR RESEARCH LEAD, AI SECURITY PORTFOLIO — RAND Corporation | San Francisco, CA | $167,300–$261,400
RESEARCH LEAD, AI CYBER TESTING & EVALUATION — RAND Corporation | Washington, DC | $146,200–$261,400
SAFETY POLICY MANAGER, GLOBAL AFFAIRS — OpenAI | San Francisco, CA / Washington, DC | $171,000–$280,000 + Equity
POLICY COUNSEL, EMEA — Anthropic | Dublin / London | £200,000–£240,000
🫵 Want your message in front of 200,000 AI builders?
Our partners and sponsors get exclusive placements across the newsletter and access to AIC's in-person network — demo nights, dinners, hackathons, and forums across 180+ chapters.
For all inquiries, send us a note at [email protected].
The AI Collective is built by volunteers across 180+ chapters in 40 countries.
Thank you to the thousands of volunteers around the world who make this work possible. We truly could not do this without you.
🧑💻 About the Editors
About Noah Frank
Noah is a researcher, innovation strategist, and ex-founder thinking and writing about the future of AI and the workforce. His work and body of research explores the economics of emerging technology and organizational strategy. Outside of AIC, Noah heads research for Centaurian AI.
About Joy Dong
Joy is a news editor, writer, and entrepreneur at the intersection of AI and blockchain. Whether she is demystifying complex systems in her newsletter, TEA, or building streamlined solutions through her automation agency, Ownly, Joy’s mission is to make emerging tech accessible and actionable for everyone.
About Lindsay Gross
Lindsay is an AI engineer, researcher, and writer focused on how AI systems behave in practice and what it takes to make them safe. Her work sits at the intersection of AI safety, governance, and product design, and at AIC she writes about the questions that matter most as these systems scale.