Drone Armies, Cursor Agent & More

TLDR;

Technology:

🫢 Anthropic’s Game-Changing MCP Protocol

🍎 Apple to Launch New Siri in 2026

👩‍🔬 OpenScholar, a SOTA Scientific Research Tool

Markets:

🔮 Klarna is Killing it with AI

🤑 Verizon Achieves Landmark Milestone

Product:

👾 An Army of Drones

😟 Wait… Perplexity Isn’t a Search Engine?

😍 Cursor’s New Agent is Amazing

AI ENGINEERING

Anthropic’s Game-Changing MCP

Are there any doubts that Anthropic is the leading AI lab right now? And I’m saying this as a ChatGPT main (although I prefer Anthropic’s API).

Anthropic has introduced the Model Context Protocol (MCP), a new standard aimed at simplifying how AI models interact with data sources, a surprisingly tricky development pipeline as I describe below in the Product section.

Previously, each integration required custom coding. With MCP, a single protocol enables AI to connect with any data source. In the following X thread, an Anthropic employee shows several video examples of how MCP works, like a demonstration showing MCP enabling the Claude desktop app to interact with GitHub, creating repositories, and pulling requests quickly, reducing setup time from weeks to under an hour.

Under the hood, MCP uses a client-server architecture, where clients (the LLMs in this case) connect to servers that securely adapt various data sources. It’s not rocket science, but the key is that it’s an open standard that, if adopted widely, which seems to be the case based on how many companies announced integrations just today, will make connecting data to your LLMs a breeze.

TheWhiteBox’s takeaway:

I just love this.

I sincerely believe that 2025 will be the year of AI engineering. As Generative AI products mature and standardize, 2025 could be the year anyone willing to become a ‘doer’ uses AI to build whatever software they want (another nail in the coffin for SaaS companies).

Importantly, we must put extra weight on the word ‘standardized,’ which is crucial to reducing the noise and helping you use the tools everyone uses.

For instance:

LLM APIs, both proprietary and open-source, follow the OpenAI standard, so they are almost identical except for minor tweaks

With MCP, providing context to your LLMs will also be standardized.

Even GPU kernel development, which is NVIDIA’s moat, by the way, with CUDA, is on the path to standardization thanks to TinyCorp.

All things considered, the development stack around LLMs is maturing enough to start building this new age of LLM-enabled software. It’s coming.

APPLE

Apple To Launch New Siri in 2026

Apple is planning a new Siri for 2026, as reported by Bloomberg.

Internally referred to as “LLM Siri,” it aims to enhance its conversational abilities and perform more complex tasks by leveraging large language models (LLMs). This initiative seeks to position Siri as a more competitive alternative to AI chatbots like ChatGPT.

The company is testing this new Siri iteration on iPhones, iPads, and Macs through a separate application. It plans to replace the existing version eventually. An announcement is anticipated as early as 2025, potentially during the Worldwide Developers Conference,…

Enjoyed the introduction? Subscribe or sign in to continue with this edition.