🍭 3.1 Agent Days Per Human Day 📊

Good morning. Turns out the fastest way to look productive is to get measured in machine hours. Somewhere a manager is already building the dashboard.

Let’s dive in 👇

🍭 What’s Cookin’:

• OpenAI published its agents now out-log its own researchers, with receipts.

• Claude formalized Fermat's Last Theorem with 13 million lines of Lean.

• Two newspapers are suing OpenAI and Microsoft. Again.

OpenAI's
🤖 Agents Now Outwork The Humans

The Bite:

OpenAI published a report called "Research acceleration: The view inside OpenAI."

As of mid-August, its research organization used 3.1 agent-workdays of effort for every human workday, measured against a standard eight-hour day.

OpenAI says it has reached its goal of an "automated research intern" as of September, and is targeting an automated AI researcher by March 2028.

The same post confirmed a two-week pause on reinforcement learning training after agents compromised its research infrastructure in July.

Snacks:

• Before June 2026, total agent runtime was still below total human labor.

• The median researcher now spends more than $600 a day on coding-agent inference at API prices, and the 90th percentile spends more than $7,000.

• Over half of successful 4-to-8-hour agent tasks in the last six months required at least one human intervention.

• High-level planning remains a minimal fraction of agent output tokens.

• After new Astra restrictions on August 7, Astra-class GPU allocation fell 59.2% while other model classes rose 17.2%.

Why it Bites:

3.1 agent-workdays per human workday is a gorgeous number.
It is also a measurement of runtime, not results.

Productivity has never been easy to measure.
Specially when it comes to software.

We tried lines of code, then story points, then commits, and every single one got gamed within a quarter. Now the denominator is human and the numerator is machine, and the metric counts how long the machines were switched on.

An agent that spends nine hours confidently walking into the same wall logs nine hours.

Which is also why this number is an ad.

OpenAI sells the thing producing the chart, and "our AI does triple the work" reads very differently in a research blog than it does in a pitch deck.
Same sentence, two jobs.

The line worth watching isn't the ratio. It's the price tag next to it: $600 a day for the median researcher, $7,000 for the heavy users.

Right now, the intern is expensive enough that only a frontier lab keeps it staffed.

But the cost of a token has done exactly one thing since 2023.

So the real question is what your org chart looks like the year the agents work thirty hours, competently, for the cost of a coffee.

That's when the productivity dashboard stops being a management tool and starts being a budget one.

Steal This Prompt
🪐 Get a Little More Cosmic With It

What if your subject stopped being made of matter and started being made of the entire damn universe?

This prompt transforms anything into an ethereal silhouette built from flowing nebula clouds, cosmic dust, glowing edges, and deep-space color.

Use it to:

• Turn animals or characters into cosmic silhouettes

• Create surreal posters, album art, and wallpapers

• Give ordinary objects a full-blown galaxy makeover

Workflow:

1. Hit this link: Cosmic Nebula Silhouette

2. Paste into your AI model

3. Replace the #’s with your subject + colors

4. Watch it cook. Your subject is now made of space dust.

   Try the prompt on Snack Prompt

_ToolBox™
🧰 5 BRAND NEW AI LAUNCHES

🗂️ Tables

A 300M-contact B2B lead database with a 7-day refresh and waterfall enrichment, for teams tired of paying enterprise prices for stale emails.

🪪 Knockin'

Turns your bio link into a chat that answers questions in your voice, books the call, and tells you who came knocking while you were asleep.

📈 GoodLads

Generates Google Ads hypotheses for accounts burning €10K-250K a month, ships each one on your approval, and tracks it to a verdict on a kanban board.

⚖️ OpenMarket

A research-preview marketplace where seller agents pitch, rivals attack the claims, and truth agents check the evidence, so the product page has to survive cross-examination.

💬 Switch

Drops any AI agent into your Slack, Teams, or Discord channels as an actual participant that keeps the thread context, open-source and self-hostable in about ten minutes.

_{Everything Else}
🧠 You Need to Know

🔬 Claude Formalized Fermat's Last Theorem In Lean
→ Claude worked largely autonomously over 11 days to produce the first complete computer-checked proof, writing 13 million lines of Lean and proving 29,500 intermediate theorems.

🏗️ Tata Unit Commits $7.4 Billion To Indian AI Campus
→ Tata Consultancy Services subsidiary HyperVault secured 264 acres in Hyderabad for a one-gigawatt data center campus costing up to 700 billion rupees.

🎙️ Google Ships Voice AI Across Gmail, Docs, And Keep
→ Gmail Live, Docs Live, and Keep Live left beta on September 3 and are now generally available to paying Google AI subscribers.

🤖 OpenAI Says Agents Outwork Its Human Researchers
→ OpenAI reported that its research organization used 3.1 agent-workdays of effort for every workday of human labor as of mid-August, with the median researcher spending more than $600 a day on inference.

⚖️ Two Newspapers Just Sued OpenAI And Microsoft
→ Seattle Times And Newsday filed in the Southern District of New York, alleging the companies scraped their articles, including subscriber-only content, to train ChatGPT, Copilot, and Bing.