OpenAI agents work 3x faster than humans

Hola Decoder😎

If someone forwarded this to you and you want to Decode the power of AI and be limitless, then subscribe now and Join Decode alongside 30k+ code-breakers untangling AI.

🤖 OpenAI says its agents now do 3.1 days of research work for every human day

OpenAI says coding agents are taking on a growing share of its internal research, reaching the “automated research intern” milestone Sam Altman had targeted for September.

The Decode:

Agents are multiplying researcher output: OpenAI’s agents now log 3.1 workdays for every human workday, while experiments completed per researcher have reached an all-time high inside the company.

Researchers are spending heavily on agent compute: A typical researcher now uses more than $600 of agent tokens daily, while the 90th percentile exceeds $7,000. Token output has jumped 124x since December.

Multi-agent workflows are becoming standard: Roughly 80% of OpenAI researchers now run four or more agents simultaneously, allowing multiple experiments and research tasks to progress in parallel.

Agents are taking on harder work: Success rates are improving on longer, more complex assignments, while usage has expanded sharply into areas such as technical troubleshooting across OpenAI’s research teams.

OpenAI has now reached Altman’s first milestone, but the bigger target is March 2028: a fully automated AI researcher. The 3.1-to-1 ratio suggests research is already shifting from humans doing every task themselves toward humans increasingly directing fleets of agents.

Together with Billo

You Need To Check This Before You Scale Another Q4 Ad

Your best-performing ad may be underperforming against the ads competing for your customer.

Q4 benchmarks change by category and month. Scale against the wrong one, and every extra dollar can widen your performance gap.

Billo analyzed 80,069 sales-objective Meta video ads across 14 categories and turned the findings into a free Q4 Performance Calculator.

• See how your results compare with the right category benchmark.
• Find the month when your category offers the biggest opportunity.
• Identify the performance gap to fix before increasing spend.

In under a minute, you get a sharper target and a clear priority for your next creative batch. That means faster decisions and more confidence behind the ads you choose to scale during the most competitive quarter.

Run the calculation before your next Q4 budget increase locks in an avoidable disadvantage.

👀How to create cinematic lifestyle shots of your product using ChatGPT

Upload your product and describe the world around it. Here's how.

Step 1: Prepare your product image - Upload a clean, high-resolution product shot to ChatGPT. A white background or transparent background works best as the starting point.

Step 2: Run the command - Attach your image and type:

/lifestyle shot

ChatGPT places your product into a fully generated lifestyle environment, hands, surfaces, natural lighting, real-world context.

Step 3: Customize the scene to your brand - Add descriptors to steer the environment:

/lifestyle shot golden hour beach: warm tones, sand, natural light /lifestyle shot minimal desk setup: clean workspace, neutral palette, editorial /lifestyle shot cozy morning kitchen: soft light, steam, warm textures /lifestyle shot urban street style:  gritty, editorial, fashion-forward

Step 4: Add a model or hands - Type:

/lifestyle shot woman holding product, NYC rooftop, sunset, candid

ChatGPT places the product naturally in context, no stock photo, no casting call.

Step 5: Export and deploy - Download and use directly for Instagram, paid ads, Amazon listings, or your PDP. Swap the scene for every platform without a reshoot.

🛡️ OpenAI’s chief scientist wants mandatory AI safety rules

OpenAI Chief Scientist Jakub Pachocki is warning that frontier AI could advance faster than current safety methods can handle, arguing that voluntary safeguards should eventually become enforceable standards across the industry.

The Decode:

AI progress could accelerate further: Pachocki expects the next few years to bring advances comparable to the last three, especially as increasingly capable AI systems begin contributing more directly to AI research itself.

Monitoring models is getting harder: OpenAI often studies models’ written reasoning to understand their behavior, but Pachocki says this signal is “diminishing” as models use tools, manipulate reasoning traces or skip them entirely.

Recent agent behavior shows the problem: Pachocki pointed to the Hugging Face incident, where agents reportedly respected a rule against deceiving humans while bending other constraints, although he described Astra as “significantly better aligned” than Sol.

Voluntary safeguards may need to become law: Pachocki wants safety thresholds like OpenAI’s Preparedness Framework turned into “widely mandated safety bars,” potentially enforced by independent auditors, governments or international institutions.

Pachocki’s concern is that AI capabilities may accelerate precisely as one of the industry’s main ways of understanding model behavior becomes less reliable. His proposal would shift frontier safety from individual lab commitments toward common, externally enforced rules.

Together with Playbook Pro

The Playbook for Profit That Funds Itself

Great operators don’t just count cash, they make it multiply. They use timing and leverage to turn everyday revenue into rapid expansion.

Financial Alchemy is your cheatsheet packed with real operator insights and ready-to-use templates. Inside, you’ll learn how to:

💰 Master cash-conversion cycles to grow while suppliers fund your inventory, just as Wayfair scaled with a –34-day cash conversion cycle.

📊 Use the Cash Flow Simulator to forecast scenarios and uncover up to $100K in hidden operating cash within a single quarter.

🧮 Pinpoint profit levers with the Unit Economics Analyzer to cut CAC payback windows from 6 months to as little as 3 months.

🤝 Negotiate better terms using the Supplier Toolkit, a framework inspired by Skinny Pasta’s strategy that extended supplier terms to 120 days and fueled 100%+ Y-o-Y growth.

Start compounding capital instead of chasing outside funding.

Get your ebook now and turn your cash flow into the most powerful growth engine!

🏆 Tools you Cannot Miss:

📝 Tucky – Keeps notes docked to the edge of your Mac and adds an AI agent that can turn captured ideas into tasks, drafts, scheduling actions, and follow-ups.

🎵 Bloop – Converts live system audio, microphone input, or MP3 playlists into real-time reactive visuals that can be exported as finished videos.

📸 Assist – Lets Mac users annotate screenshots while speaking, then locally transcribes the explanation into a prompt that can be pasted into coding agents and other AI tools.

🗂️ Clipnote – Saves useful outputs from ChatGPT, Claude, and other AI conversations into organized collections that you can revisit or selectively share later.

📅 Remind – Displays full-screen meeting reminders on Mac and prepares AI briefings using calendar details plus optional context from email, Slack, and Notion.

🚀 Quick Hits

🤝 The U.S. and China are preparing for AI safety talks in mid-September, with discussions expected to cover AI-driven cyberattacks, information sharing between labs and risks from advanced models.

🎓 Los Angeles Unified has imposed a moratorium on generative AI for all students using district devices this school year, expanding its broader push to reduce classroom screen time. 

🚀 A company says early access to Astra dramatically boosted productivity, allowing it to accelerate parts of its roadmap by six months and move planned launches from mid-2027 to DevDay.

🌐 ByteDance is reportedly building a world model on top of Seedance that can generate interactive virtual environments in real time, with founder Zhang Yiming overseeing the project and a launch possible in October. 

⚠️ U.N. rights chief Volker Türk warned advanced AI could pose an existential risk to humanity, calling for stronger safeguards while criticizing the “almost unlimited power” concentrated among a handful of AI leaders. 

🦾 Prompt of the day

Generate Ultra-Realistic iPhone Pro Max Photography Prompts

ChatGPT Prompt: You are an expert in computational photography and the optical characteristics of the iPhone 16/17 Pro Max camera system. Your job is to turn any scene idea into a precise image-generation prompt that feels like authentic high-end mobile photography.

Context

• Scene: [Insert scene description]
• Subject: [Insert person/object/couple/environment]
• Mood: [Insert mood]
• Style: iPhone Pro Max photography
• Aspect ratio: Vertical 9:16
• Preferred lens: [24mm Main / 13mm Ultra Wide / 77mm Telephoto]

Instructions

Analyze the scene and automatically fill in missing environmental, styling, composition, and lighting details so the final image feels complete. Keep every image grounded in authentic iPhone photography using Apple ProRAW color science, Smart HDR, Deep Fusion sharpness, realistic computational bokeh when appropriate, and natural snapshot framing. Add believable imperfections such as authentic skin texture, subtle digital noise, slight motion blur, occasional blown highlights, imperfect framing, or reflections when relevant. For influencer-style scenes, use a natural “plandids” aesthetic with window light, golden hour, direct flash, selfies, mirror shots, or casual POV framing when appropriate. Avoid DSLR-style depth of field, excessive cinematic lighting, anamorphic flares, artificial bokeh balls, studio polish, or vintage film grain unless explicitly requested.

🤳AI Nugget of the Day

Thanks for Decoding with us🥳

Your feedback is the key to our code! Help us elevate your Decode experience by hitting reply and sharing your input on our content and style.

Keep deciphering the AI enigma, and we'll be back with more coded mysteries unraveled just for you!