AI Leadership Weekly

Top Stories

New OpenAI tool to build agents with APIs
OpenAI has taken a big step forward in tying all of its new tools and features together with their new Responses API.

It pulls together their Chat Completions API as well as the Assistants API, making it easier for developers to create advanced agents without having to fiddle around with multiple APIs for a single task. Further to this, they've added three new tools to the mix: Web Search, File Search, and Computer Use.

• Web Search Tool uses a highly-trained model to search the web and effectively cite where it found its results.

• File Search Tool lets users embed data in a convenient way for agents to perform RAG tasks.

• Computer Use Tool rolls in OpenAI's Operator features, allowing these models to take control of a computer or VM.
  

Meta begins testing in-house chip
As part of their plan to reduce reliance on Nvidia, Meta has begun testing its first in-house AI chip.

Creating custom AI silicon is a goal of multiple corporations these days, and feels akin to ASIC development for crypto mining. The hope of these chips is to drive down infrastructure costs for the company (both in terms of chip purchases and power requirements), who has to source current chips from Nvidia, as well as creating chips specifically designed to speed up AI tasks. Reuters reports that the current tests, known as "tape out", cost tens of millions of dollars to perform, and can take 3 to 6 months with no guarantee of success.

Manus, the Chinese agent
Operator was all the rage what felt like months ago now, but now the Chinese agent AI Manus is taking the spotlight.

Manus uses multiple agents in different roles—such as a planner agent, knowledge agent, and executor—and uses multiple LLM models, including Claude 3.5 Sonnet and Qwen. Their siloed approach aims to help with jailbreak exploits, and they claim it competes with and exceeds OpenAI's Operator on the GAIA benchmark.

Researchers at MIT were impressed with the product, but found it unstable at times and prone to crashing. They did say, though, that when it works, it works better than OpenAI’s DeepResearch.

In Brief

• Google’s Gemma 3, made to run on a single card

• Sony experimenting with AI characters on PS5

• Apple delays Siri AI upgrades to 2026

Market Trends

Judge allows comedian to sue Meta, OpenAI
In a case that started back in 2023, a federal judge is allowing part of a copyright suit against Meta and OpenAI to proceed.

The class action was brought by author and comedian Sarah Silverman, as well as authors Richard Kadrey and Christopher Golden. In the suit, their lawyers argue that LLM models and products such as ChatGPT are derivative works which intrinsically can't exist without the creative works of others.

Key parts of the case include evidence that Meta knowingly removed copyright management information (CMI) from their training sets, and trained their model, Llama, in such a way so as to reduce the likelihood of it revealing that copyrighted material was included in its training set. This is on the back of recent revelations that Meta knowingly torrented 85TB of ebooks for the purposes of AI training.

25% of YC startups have AI-generated codebases
In a recent podcast, it's been claimed that up to a quarter of Y Combinator's current cohort of startups have codebases that are almost entirely AI-derived.

There is no end of AI coding tools like Cursor and Wind Surf. Anthropic's CEO also recently predicted that, by year's end, up to 90% of code will be AI-generated. (It's important to note, of course, that these claims are from an executive with a product to sell.) From the podcast, it seems as though these developers are adopting the current trend of "vibes coding" with their tools, which the hosts seem excited by, but whose long-term viability is yet to be seen. They also note that these AI tools remain terrible at debugging.

Tools and Resources

Eraserbot
Eraserbot will actively scan your codebase to create up-to-date diagrams for your documentation.

Reworkd
An AI tool to extract web data at scale.

Perplexity Desktop App
Download the Perplexity Windows app for AI search on your PC.

Recommended Reading

Is AI a bubble?
As the headline reads, "It's complicated."

The field of AI is vast and goes beyond the recent LLM explosion. But it's true that there's a huge gold rush occurring right now, with trillions of dollars being poured into both research and data centre construction, not to mention chip development. So it's worth pondering if this is just another “dot-com bubble.”

Hit reply to let us know which of these stories you found the most important or surprising! And, if you’ve stumbled across an interesting link/tweet/news story of your own, send it our way at [email protected] It might just end up in the next issue!

Thanks for reading. Stay tuned for the next AI Leadership Weekly!

Brought to you by Data Wave your AI and Data Team as a Subscription.
Work with seasoned technology leaders who have taken Startups to IPO and led large transformation programmes.