🧙🏼 OpenAI's autonomous research agent

Howdy, wizards.

Welcome to the year of AI agents. We’ve barely entered February, and OpenAI just dropped its second agent this year — Deep Research. Initial consensus seems to be that this one is tangibly more ready for real-world use than Operator.

Here’s what’s brewing in AI.

DARIO’S PICKS

1. OpenAI introduces Deep Research, an advanced research agent

After the release of o3-mini last week, OpenAI promptly dropped another, even bigger release: Deep Research.

Deep Research is a new agentic feature that finds, analyses, and synthesises hundreds of online sources and creates reports that would’ve taken a skilled human several hours — in 10 minutes. The feature is built on the upcoming reasoning model o3, which is optimised for web search and data analysis; it searches and interprets not just text, but also PDFs and images from around the web.

Deep Research goes beyond simply summarising information; rather, it acts as a researcher would do: it explores ways to find information, changes its approach along the way (e.g. finding workarounds to obtain information when faced with a paywall), digs deeper into areas that interest it, finds novel connections, and solves problems. When it cites sources, it uses legitimately high-quality sources, accurately quotes them and provides links not just to the source but often directly to the highlighted quotes.

The feature just went live for Pro users (yes, that’s the $200/month subscription), who get 100 queries per month, reflecting the high computation cost the feature currently requires. Plus/Team plans will get access next, but dates for this are TBA.

In case you’re wondering how this feature stacks up against Gemini’s Deep Research feature launched in December. Ethan Mollick notes OpenAI’s version is way more agentic, exploratory and self-directed; where Google is mostly summarising many different sources, OpenAI’s Deep Research hunts down specific concepts and works around problems to find the best answers (also while consulting a ton of sources). Casey Newton at Platformer calls Google’s deep research “shallower in every way” and that ChatGPT’s version “blows it out of the water”.

Examples and demos:

• OpenAI’s livestream for Deep Research. Includes demos like

  • asking for a market analysis for mobile operating systems (04:00)

  • recommendations on skis to buy for skiing in Japan (08:00), see the side-by-side comparison table it made at (18:40)

  • investment analysis on supersonic air travel (13:50)

  • study on protein modification (14:45)

  • finding a tv-show based on details from the episode (15:40)

• A 30-page report on the evolution of tabletop role-playing games (via Ethan Mollick).

• Sam Altman spent hours looking for an old Honda NSX and was about to give up when Deep Research found it..

• Hubspot co-founder Dharmesh Shah calls the new feature “mind-boggling”. He gives the example of asking it to create a research report with competitive analysis, positioning, product strategy, etc., and receiving an 11,000‐word report complete with tables, data, citations, and insights.

‎ Why it matters‎ ‎ Sam Altman calls this “a superpower; experts on demand”, enabling complex tasks that would take a human hours & days, and would cost several hundred dollars. He also guesstimates that it autonomously “can do a single digit-percentage of all economically valuable tasks i the world”. I don’t know whether this holds true—it seems like a super bold claim. But imagine what happens once these companies start combining the agents they’re creating now (e.g., Operator, Deep Research, the ability to run code, vision, voice, etc.). It feels like the individual components of the somewhat fuzzy concept of AGI are really coming together.

Another thing Altman said which I think is true—and frankly a bit concerning—is that $200/month for ChatGPT Pro doesn’t seem outrageous anymore. If you speed up your work by hours & days, that’s enough value to justify the price for more than just power users. Wouldn’t be surprised if we see a $2,000/month subscription from one of the big AI companies this year.

A key challenge for all these agents is that the internet is heavily and increasingly paywalled, as publishers scramble to safeguard their data from AI. That limits the value of tasks that can be done with access to the open internet. One approach I think will win in this game is what Perplexity is doing for Finance right now, ie combining reasoning models with collaborations with high-quality data sources like Crunchbase and Factset.

TOGETHER WITH MONDAY.COM

2. The future of work management

Picture a world where workflows are finely tuned, automated to perfection, and seamlessly integrated with your favorite apps. It's not just a platform; it's a revelation—a space where managers gain unparalleled visibility into team processes, ensuring each project is a resounding success. Step into the future of work management with monday.com, where efficiency isn't a goal; it's a given.

From startups to industry giants, monday.com has transformed how teams work. Why not let your team be the next success story?

Start your free trial today

QUICKFIRE HEADLINES

Side notes

Here are the other essential updates on AI you shouldn’t miss this week, that I didn’t have time to cover in-depth.

• SoftBank announced a $3 billion annual commitment to OpenAI’s tech and is partnering with OpenAI on a Japanese joint venture, an AI system that will “revolutionize business” called Cristal Intelligence. SoftBank is already an initial backer of the $500 billion Stargate project that I wrote about a couple of weeks ago.

• Not directly AI, but notable media outlets are comparing what Elon Musk is currently doing to an administrative coup, apparently using a similar playbook as his doom-laden Twitter takeover.

• The EU is putting $56 million into developing OpenEuroLLM, its own open-source LLM that will work with the 30 member states’ languages.

• Anthropic launched something called Constitutional Classifiers, a method to defend AI models against jailbreaks.

• The Google Gemini app is now using the 2.0 Flash model—with faster responses and better performance—and uses image generation based on Imagen 3.

FROM OUR PARTNERS

Check out 1440--the fastest way to an impartial point-of-view. The team at 1440 scours over 100+ sources ranging from culture and science to sports and politics to create one email that gets you all caught up on the day’s events in 5 minutes. According to Gallup, 51% of Americans can’t think of a news source that reports the news objectively. It’s 100% free. It has everything you need to be aware of for the day. And most importantly, it simplifies your life.

Sign up for 1440, it's 100% free

THAT’S ALL FOLKS!

Affiliate disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes link to products and other newsletters. Please assume these are affiliate links. If you choose to subscribe to a newsletter or buy a product through any of my links then THANK YOU – it will make it possible for me to continue to do this.