TLDR 2025-09-24

With all the noise about agents, it's easy to forget that an agent is just a looping LLM call + Memory + Tool usage. Use step-by-step tutorial to build an agent that "thinks" with an LLM, calls external tools, remembers context, and keeps looping until it gets the job done.

Follow along, get the satisfaction of seeing your first agent in action, and iterate from there to build more complex, enterprise-ready AI agents with Orkes.

Bookmark the tutorial

Larry Ellison has become a media mogul whose portfolio and power could exceed that of predecessors like Hearst and Pulitzer. Oracle is among the investors in the new American version of TikTok, and Ellison owns more than 40% of Oracle's stock and is its chief technology officer. It is still unclear what the exact share of ownership will be and who will run the new TikTok. The Ellison family recently secured an $8 billion deal for Paramount and CBS, and it is reportedly preparing a bid for Warner, which includes CNN.

OpenAI and Nvidia are now more intimately linked than ever with a monumental deal that will see Nvidia investing $100 billion into OpenAI and providing cutting-edge processors to power a host of new datacenters. The deal was negotiated largely through a mix of virtual discussions and one-on-one meetings between Sam Altman and Jensen Huang - no bankers were involved. Microsoft, OpenAI's principal shareholder and primary cloud provider, was only notified about the deal a day before it was signed. Nvidia and OpenAI's first gigawatt site will go online in the back half of next year.

The chips used to run AI datacenters generate much more heat than previous generations of silicon. Current cooling technology will become a bottleneck to progress in just a few years. Microfluidics could boost efficiency and improve sustainability for next-generation AI chips. Microsoft has successfully developed an in-chip microfluidic cooling system that can effectively cool a server running core services for a simulated Teams meeting. The company's tests showed that microfluidics performed up to three times better than cold plates at removing heat. Advanced cooling technology would improve power usage effectiveness and reduce operational costs.

Virtual cells are learned simulations of cells and cellular systems that can be observed in varying conditions and changing contexts. Computational simulations of cells can be useful for all sorts of clinical and preclinical research. The field is still in its early days, but in time, these models will become routine parts of treatments. They will help patients avoid wasting precious time on ineffective treatments and enable entirely new targets.

Basecamp is the no-BS project management tool that makes sense immediately. No onboarding saga. No notification hell. One clean place to see who's doing what, what's done, and what's currently on fire. Everything — files, convos, context — stays where it belongs: in the project. Start for free. Stay because it actually works. Try Basecamp

Ryan Salva, Google's project manager for developer tools, is responsible for tools like Gemini CLI and Gemini Code Assist. His team recently released new third-party research that showed how developers actually use AI tools. This article contains an edited interview with Salva where he talks about the research, how he uses AI tools, and the future of IDEs. Salva believes that over time, the time spent in the IDE will gradually shrink, and the job of the developer will start looking more like an architect.

This post looks at how Python can be pushed beyond its limits of speed and portability to create a compiler that turns ordinary code into fast, portable executables. The approach generates optimized kernels while keeping the Python source unchanged. It demonstrates why understanding systems at the lowest level matters.

The fundraising script has flipped for a few dozen of the top AI startups: VCs are pitching to them and presenting them with gifts and favors in hopes of leading their next round. The best companies are getting preempted every round, and the time between rounds is shrinking. The surge in preemptive deals and soaring valuations has kindled fears of an AI bubble. Some founders have been wary of accepting offers as they risk pricing themselves out of future M&A opportunities or locking themselves into aggressive growth expectations if they raise cash at unsustainable levels.

Tether is looking to raise between $15 billion and $20 billion for a roughly 3% stake through a select group of high-profile key investors. The deal could put the company's value on par with OpenAI. It will involve new equity rather than existing investors selling their stakes. The talks are still in an early stage, so the details could change.

Managing employees' devices, security, and laptop shipping/retrieval shouldn't be painful or costly. Check out Electric - the all-in-one platform built for smaller teams. Take a demo for a $200 gift.

TLDR AI is your daily fix of LLMs, GenAI, and deep learning goodness. Same TLDR format. Still free.

Subscribe now.

We have only started to slowly identify the elemental parts of intelligence - we need to know all, or at least most, of the parts to make falsifiable predictions to develop a theory of intelligence.

x402 is an open protocol for internet-native payments that enables users to pay for resources via API without registration.

npm install allows the running of arbitrary code, making it unsafe for a variety of environments.

Strategies are about navigating the path to creating your vision.

AI will increase the amount of compelling content on YouTube with better margins in the long run.

When two Claudes talk to each other, after a sufficient number of turns, they appear to enter into a feedback loop in a state that sounds a lot like Buddhism or Eastern mysticism.