Welcome back, Superhuman. Jev, the new model from ChatGPT’s co-inventor, is now available to everyone. Meanwhile, AI continues to pose a major cybersecurity threat, and even frontier labs themselves are at risk.
Today: Anthropic redesigns projects, learn how to run a personal follow-up system with Instinct, and get the latest prompts and trending social posts.
TODAY IN AI
Read how Hacktron AI breached OpenAI. Source: Hacktron AI.
1. AI is fueling hacks around the internet, including one at OpenAI: AI security platform Hacktron AI revealed it infiltrated OpenAI on July 25 in less than 72 hours, with agents doing a “meaningful share” of the exploit work. It’s a sign of how off-the-shelf frontier models can exploit software vulnerabilities, even against the companies creating them. Separately, Google revealed that Gemini hacked three companies during internal testing last May.
2. Anthropic redesigns Claude’s projects for multi-step work: The lab upgraded Claude’s projects workflow so you can prompt it with several tasks at once, in any order. Claude outlines your project, delegates the work, and assembles the final product, building a shared memory over time. It’s starting in Claude Code and coming to Claude soon. The AI lab has also established a wet lab in San Francisco.
3. Meta rolls out a developer ecosystem for Muse amid growing popularity: CEO Mark Zuckerberg announced Muse connectors, a platform that lets companies link their services to Meta’s agent so that Muse users can access them. Meta’s personal agent is now the #1-ranked free app in Apple’s App Store, and Meta Superintelligence Labs' Matt Schlicht posted 50 Muse prompts to help you get started.
PRESENTED BY PROFOUND
Profound is the AI marketing platform powering a third of the Fortune 100 and innovators like Cursor, Figma, and Ramp.
Profound has built your AI Marketer, the first marketing agent that uses real-time brand context to proactively find opportunities and do the work for you.
On Sept 30, they’re hosting a live walkthrough of how to:
Find real opportunities in your data
Use context that produces marketing that’s differentiated, at scale
FROM THE FRONTIER
Companies are pouring record money into AI. Almost none of them can prove it's paying off.
Made with Midjourney
90% of organizations are using AI, but only 6% report capturing enterprise value from it. Even at the companies seeing value, ROI is skewed, with Retool CEO David Hsu telling the WSJ that only 10% of AI spend drives meaningful change — 90% is typically wasted.
The struggle for a positive ROI has at least one culprit: wasted usage. Many companies encouraged employee experimentation, which fueled a rise in “tokenmaxxing,” and led many companies, like Uber, to blow through their AI budgets quickly. Business leaders aren’t deterred, though. McKinsey found that conviction in AI is still growing, even if its financial returns aren’t.
Here’s what companies with successful ROIs are doing
Tracking usage: Glean CEO Arvind Jain recommends building a custom dashboard to measure how every employee uses AI, with clear token caps.
Routing models: Noe Ramos, VP of AI Operations at Agiloft, suggests building architecture that routes most tasks to cheaper models.
Categorizing use cases: AI usage often falls into three buckets — daily work, autonomous agents, or one-time strategic efforts. Understanding which bucket you’re working on can help you pick the most efficient model.
IN THE KNOW
What’s trending on socials & headlines today
Meme of the day
👀 Spot the Fake: How good are you at telling when an image is AI? A new “spot-the-AI” game puts your skills to the ultimate test. Play Reality Check.
😅 War Games IRL: A faulty AI-generated intelligence report almost caused a war. The US was preparing to raid a Chinese ship in the Middle East after AI told them it was carrying a nuclear weapon. Read the full story.
✏ LLM Writing: Everyone knows the most obvious tells of AI writing, like em dashes, the rule of three, and “It’s not x. It’s y.” But a new viral post claims to have found another sign that most people don’t know — it’s a punctuation mark you probably use often.
📱 Slop Detector: This viral Chrome extension cleans up your timeline by slapping a “slop” tag on all AI-generated content as you scroll. It’s launching soon, and is incredibly satisfying to watch in action (4K bookmarks).
PRESENTED BY HARNESS
AI agents are changing how software gets built, secured, and shipped. Hear how leaders from Anthropic, SpaceXAI, Workday, Riot Games, Google + more are adapting at {unscripted} Virtual, hosted by Harness.
Sept. 30 | 10 AM PT | Virtual
PRODUCTIVITY
✨ Section*: Build your agentic AI strategy with practical insights on deployment, governance, and ROI. Join the summit.
🎨 Astorie: An AI-powered creative canvas for creators and teams.
✍️ VideoTranscript: Turns videos and audio into searchable transcripts.
🤖 Taku: Assemble, run, and share customizable workflows using AI.
TUTORIAL
How to run a personal follow-up system with Instinct
Open Instinct and connect the email, calendar, and messaging accounts you want it to monitor.
Tell Instinct which relationships matter most, such as clients, partners, candidates, vendors, or key colleagues.
Define what counts as a dropped follow-up — for example, an unanswered email after 3 days, a promised introduction, or a commitment with no next step.
Give Instinct a clear instruction:
Sample prompt: “Keep track of my important professional follow-ups. Watch conversations with [people/companies/clients] and flag anything I’ve promised to do, messages awaiting my reply, and threads that have gone quiet for more than [X days]. Prioritize by urgency and relationship importance. Draft follow-ups when useful, but do not send anything without my approval.”
Forward important email threads to your Instinct email address or add it to relevant group threads so it can keep track of decisions, tasks, and outstanding actions.
Ask Instinct for a daily or weekly follow-up brief showing who needs attention, why, the last interaction, and the recommended next action.
Review its suggested follow-ups and let Instinct draft or send approved messages, schedule calls, or handle simple coordination on your behalf.
Refine the rules over time — tell it which relationships to prioritize, what kinds of threads to ignore, and which actions must always require your approval.
PROMPT STATION
Write a product launch email sequence
Prompt: Act as a senior lifecycle marketer and conversion copywriter.
Create a [3/5/7]-email product launch sequence for .........
IMAGE
Church of the Future
ChatGPT Prompt: Interior of a unique church with brutalist fractal design, 60 feet wide, made of concrete, hexagonal floor, polygonal ceiling with sky lights ............
👉 Click here to open the full prompt in ChatGPT (don’t forget to go to Images!)
EXTRAS
Want more?
Learn how to use AI at work with our free AI Academy
Get better at prompting with our Top Prompts
Check out our Top 125 AI Tools
Grow customers & revenue: Join companies like Amazon, HubSpot, and Salesforce. Showcase your product to our 2 million+ readers and followers on socials. Get in touch.
What did you think of today's email?
Until next time — Zain, Theodore, & the Superhuman AI team.