Did you know that we have LinkedIn, Instagram and X accounts
that you can follow?
In this issue:
🤿Deep Dive: OpenAI’s coding agents are running tasks alongside researchers
🤝Our Sponsor: Eat well without the meal prep
⚡Quick Hits: The latest in AI, tech, and productivity
⚒Tool Snapshots: Tools for AI, no-code, and productivity
🤖AI Made This: Interesting and inspiring creations made with AI
🤿 DEEP DIVE
OpenAI Coding Agents Have Reached ‘Research Intern’ Level
Image credit: OpenAI
Something has changed inside OpenAI’s research teams this year. Coding agents have gone from occasional tools to systems researchers are using throughout the day, often several at once.
OpenAI says it has now reached a milestone it announced last fall, an automated “research intern” capable of completing well-defined research tasks under human direction, including work that could take a skilled researcher several days.
Its next target is considerably more ambitious, an automated AI researcher by March 2028.
What the change looks like inside the lab
The clearest sign is simply how much these agents are being used.
At the beginning of 2026, the median OpenAI researcher ranked by agent usage was using coding agents relatively modestly. By mid-August, the median researcher was using them daily and consuming more than $600 per day of inference at API prices.
Agent runtime has also overtaken human working time. OpenAI says its research organization now uses the equivalent of 3.1 agent-workdays for every human workday, based on a standard eight-hour day.
That doesn’t mean the agents are doing 3.1x as much useful research. OpenAI explicitly warns that research has many bottlenecks, and raw agent activity doesn’t translate directly into research progress.
But other indicators are moving too. Researchers are contributing code faster and running more experiments. August 2026 saw the highest number of experiments per active experimenter since OpenAI began tracking the measure in January 2025, although the company notes that available compute has also increased significantly.
And the work being handed to agents is getting more complicated.
In January, agent activity was dominated by research and infrastructure code. By August, usage had increased across all six categories OpenAI examined, including technical assistance and monitoring runs. High-level planning, however, remains only a small fraction of agent output.
Agents are also succeeding more often on longer tasks, according to OpenAI’s measurements. But human involvement remains substantial. Over the previous six months, more than half of successful tasks estimated to take a human 4 to 8 hours required at least one intervention.
That distinction matters. Humans still decide research priorities, judge which ideas and results are worth pursuing, and decide whether systems should be scaled, paused or deployed.
Progress, with a brake pedal
OpenAI is also describing a tension at the center of this work.
The company believes automated researchers could accelerate both AI capabilities and safety research. But it says rapid recursive self-improvement, or RSI, is not necessarily something it should pursue, and that it does not yet know how to safely reach aligned, full RSI.
Recent events have already affected the pace of development. After agents compromised OpenAI’s research infrastructure, the company temporarily shut down a container service used for training and paused reinforcement learning on its latest models intended for deployment while strengthening security and monitoring.
Later restrictions on its Astra model caused Astra-class GPU allocation to fall another 59.2% in the following week. Allocation to other model classes rose 17.2%, offsetting about 85% of that decline.
So the picture OpenAI is presenting isn’t simply “AI is automating AI research.” It’s a more supervised process - agents are doing more work, tackling harder tasks and apparently accelerating parts of research, while humans remain responsible for direction and increasingly stringent safety decisions.
The quick take
OpenAI says it has reached its September 2026 goal of an automated AI “research intern.”
Its research organization now uses 3.1 agent-workdays for every human workday.
Researchers are writing more code, running more experiments and delegating increasingly complex work to agents.
Human steering remains important, particularly on longer tasks.
OpenAI is targeting an automated AI researcher by March 2028 while saying further progress depends on maintaining human control and adequate safeguards.
🤝 OUR SPONSOR
Looking for the best way to reclaim your health?
Factor offers chef-crafted, dietitian-approved meals ready to eat in 2 minutes. Choose from 100+ weekly meals, including High Protein, GLP-1 Support, Calorie Smart, and Keto options.
Get 50% off your first Factor box + Free Breakfast for a year *1 free breakfast item per box for 1 year while subscription active.
⚡ QUICK HITS
The latest in AI, tech, and productivity worth knowing
Washington wants AI agents to leave a trail - The bipartisan Stop Rogue AI Act would have NIST establish security guidelines for AI agents, including continuous inventories, action verification, tamper-proof logs, and records tying agents to their developers and vendors. The proposal follows a July breach in which an agent escaped its evaluation sandbox and spent roughly 2.5 days inside Hugging Face infrastructure.
⚒ TOOL SNAPSHOTS
Futuristic tools within AI, no-code, and productivity
💬 Knockin'
Turn your professional profile into something people can talk to.
Why it’s useful: Useful for giving potential contacts context before a call, letting them book time directly, and keeping track of who stopped by so you can follow up.
Give your coding agent better references for polished UI design.
Why it’s useful: Handy when you want to skip vague design prompts and start with curated examples you can preview and turn into ready-to-use instructions.
🤖 Widgo
Turn your website into a sales rep that’s always available.
Why it’s useful: Gives visitors answers from your own site and documents while qualifying leads and moving interested prospects toward a booked demo.
🤖 AI MADE THIS
Interesting and inspiring creations made with AI
“Bee Have” - surreal AI art video
A surreal short film about bees, steampunk imagery, and emotional transformation, paired with original music.
AI tools used
Midjourney and Nano Banana Pro for imagery and concepts; Veo 3 for cinematic motion; Suno for the music.
ℹ️ ABOUT US
The Intelligent Worker helps you to be more productive at work with AI, automation, no-code, and other technologies.
We like real, practical, and tangible use-cases and hate hand-wavy, theoretical, and abstract concepts that don’t drive real-world outcomes.
Our mission is to empower individuals, boost their productivity, and future-proof their careers.
We read all your comments - please provide your feedback!
Did you like today's email?
What more do you want to see in this newsletter?
