Welcome to this edition of Loop!

To kick off your week, I’ve rounded-up the most important technology and AI updates that you should know about.

‏‏‎ ‎ HIGHLIGHTS ‏‏‎ ‎

Let's jump in!

Sign up for a free trial of Attio, the agentic CRM.

Ask Attio to build a daily workflow that surfaces the deals that need your attention today, like anything with a stage change, a recent reply, or a new signal in the last 24 hours.

Review your pipeline in Claude, synced live from Attio via MCP.

That's it.

Try Attio Now

1. Big Tech is coming for consultancy companies

We start this week with Microsoft and Amazon, who have launched their own AI deployment companies within days of each other.

Microsoft's new venture is called the Frontier Company, which is backed by $2.5 billion in funding and will include 6,000 engineering experts. Amazon is committing $1 billion to its organisation and it will be created within AWS.

Both of these new ventures are based on the forward-deployed engineer (FDE) model, which was pioneered by Palantir. Rather than selling a product to companies, Amazon and Microsoft will temporarily embed their AI engineers within the customer company.

These engineers will then work with existing teams and build AI systems that are specific to that company, so that they can get the most out of the technology.

OpenAI and Anthropic have already launched their own FDE ventures in recent months, which are valued at $4 billion and $1.5 billion respectively.

It's clear that companies are really struggling to deploy AI systems by themselves, and the tech giants see a big opportunity in building that out for them. It's a real threat to the business model of consultancy companies, who already make huge amounts of money doing this.

2. AI saved a man’s life after doctors misdiagnosed him

With so many negative stories about technology, I thought this was a great story that showed the positive ways it's being used.

In New York, an AI model was used to help save a man's life and identified a serious misdiagnosis by doctors, who failed to spot that he was suffering from heart failure.

When the 45-year-old arrived at the hospital, he was struggling to breathe and had an abnormal ECG - but doctors didn't see anything conclusive and sent him home with an inhaler.

That same hospital was trialing EchoNext in the background, which can analyse ECGs and spot signs of heart damage that are missed by humans. The model flagged that the man could be suffering from heart failure, and a follow-up scan found his heart was pumping just 10% of its blood, which eventually led to a heart transplant.

The model has been approved by the FDA and will be available for free through OpenEvidence, which is a popular medical chatbot that's used by around half of American doctors.

For years we've been told that AI could help us spot disease and save lives, but so far most of those claims have felt like marketing. This is a great example of how the technology has made a real difference, and with EchoNext now available for thousands of doctors, we should expect a lot more stories like this one.

3. China reveals its answer to Claude Mythos

360 is a Chinese cybersecurity firm and they claim that they've built a domestic answer to Anthropic's Mythos, with a new AI system that can automatically find software vulnerabilities.

Mythos was previewed in April and found thousands of flaws in the software systems that we all rely on, which caused enough alarm that the US government ordered Anthropic to suspend exports of a weaker version.

360's founder describes their new cybersecurity tool as a strategic weapon that "cannot be held only by others" and is needed to support China's domestic industry.

Interestingly, he admits that Chinese models are trailing the American ones by 20-30%, so 360 is compensating this with an agent-based approach. They have done this by wrapping these weaker models with detailed security guidance, vulnerability databases, and automated tools.

I'm a bit skeptical of the claims here, as 360 says its tool found 3,432 vulnerabilities but only 105 were confirmed by Chinese authorities, and none have been independently verified.

The main takeaway here is that both countries now treat vulnerability-finding AI as a national strategic asset, and we should expect an arms race in automated cybersecurity tools - with defenders on both sides racing to patch what these systems uncover.

4. Notion adds interactive HTML blocks for AI agents

If you've been using AI agents to help with your work, you've probably been forced to read reams and reams of AI-generated text - which becomes quite tedious and mind-numbing after a while.

In the last few months, I've noticed that people are becoming really sick of markdown files and how much text they need to read. This leads to most people skim reading them at best, or just accepting whatever the agent has suggested - even though there's the risk that it's completely wrong.

To get around this, people are starting to ask their AI agent to write the plan as a markdown file first, and then turn that proposal into an interactive HTML page - allowing them to actually see how a feature would work, rather than reading through hundreds of words.

I've been doing the same with Claude Code and it's a lot easier to review the work that was done, and push back before anything gets built. But the downside is that these HTML files are hard to share, as most platforms don't render them natively.

Notion has clearly spotted this trend, as they've just announced that they now support HTML blocks directly within their platform - allowing you to easily share these interactive files with anyone on your team.

Their native AI agents can also use them to build things like a ROI calculator, a quiz, or an org chart for new hires. Notion has an early lead here, but I wouldn't be surprised if other productivity tools - such as Confluence and Google Docs - follow suit before the end of the year.

5. Meta's new AI can read your thoughts and type for you

Meta has released a new AI system that can decode brain activity and type out your thoughts, without the need for an implant.

To get this working, participants wore a special helmet that can read their brain signals as they think about typing. The AI system then analyses those signals and instantly turns them into sentences.

With this second version of Brain2Qwerty, the team achieved 61% word accuracy on average, which is a huge jump from the 8% accuracy that's common with other non-invasive tools.

These are some very impressive results and they're approaching the accuracy of surgical brain implants, which is surprising to see.

Meta has released the training code for Brain2Qwerty v2, and its research partner has released the underlying dataset that was used, which should help other neuroscience teams to build on their work.

MEG machines are still huge and expensive, so this technology isn't going to reach the average person anytime soon. But for the millions of people with brain conditions, it could be genuinely life-changing.

Anthropic will use AI to create medications for “neglected diseases”

The company has announced that it will start developing its own medication, with a focus on creating treatments for "neglected diseases".

The move puts Anthropic in an unusual position, as it also sells its Claude Science workbench to pharma customers who it will now be competing with.

They join a growing list of AI companies that are chasing drug discovery, including Google DeepMind's Isomorphic Labs and AI-first companies like Insilico.

To back this up, Anthropic has stepped up hiring over the past year - with several researchers being poached from Big Pharma companies and top academic institutions.

It'll be interesting to see what diseases Anthropic prioritises, although I'd temper expectations here.

No AI-designed drug has yet made it through clinical trials and FDA approval. Experts have also pointed out that AI models still can't replace the slow, expensive lab work that's needed to prove that a drug is safe and effective.

While AI is genuinely useful for speeding up the search for new drug candidates, it will take a long time for the effort to payoff. Regardless, this is a good move from Anthropic and they've got a lot of funding to try and make this work.

How to build an advanced voice agent

In just a few hours, I was able to build an AI coach that helps me prepare for upcoming interviews, think through big decisions out loud, and explore ideas in greater depth.

I've turned it into a free guide that walks you through the entire workflow, so that you can build your own voice agent - even if you're not technical.

It covers everything from setting up the tools, to customising the agent for your specific use case, to my tips for getting the most out of the technology.

AssemblyAI are also giving Loop's readers $50 in free credits, which is more than enough to build and test your first agent.

I shared a link to the guide in last Thursday's email, but I'm resharing it again just in case you missed it. You can read the guide using the link below.

I'd love to hear what you think of the guide and what you ended up building - just hit reply to this email and let me know.

⚛️ Realta Fusion becomes the first company to generate electricity directly from a fusion reaction

🧬 US scientists create a manmade cell, for the first time

🤖 Anthropic makes Fable 5 available again, after a greenlight from the US government

💬 Mark Zuckerberg tells staff that AI agents haven't progressed as quickly as he'd hoped

💾 South Korea's tech giants will spend over $550 billion to boost RAM production

🧪 Anthropic launches Claude Sonnet 5 and a new tool for scientists

☁️ Meta could start its own cloud computing business

🎨 Google releases a faster, cheaper image generator

This startup is building its own custom AI chips, which are designed to run models faster and more cheaply than Nvidia's GPUs. The company was founded in 2022 by two Harvard dropouts, who left to become Thiel fellows and take on Nvidia.

The company sells what it calls "frontier inference clusters", which are full systems that bundle its chips with custom racks and software. If you're not familiar with inference, it's everything that happens after you submit a prompt to a model.

This has become the biggest bottleneck and cost for AI companies, which is why investors are closely watching anyone who promises to speed it up.

The company has now raised $800 million in total, including a $500 million round in December that valued it at $5 billion. Its backers now include AI heavyweights like Geoffrey Hinton, Fei-Fei Li, and Andrej Karpathy - alongside the billionaires Stanley Druckenmiller and Peter Thiel.

After TSMC was able to successfully manufacture the AI chip, Etched has booked $1 billion in contract orders - which is remarkable for a company that struggled to raise any money back in 2023.

Back then, the founders were operating month-to-month and close to running out of cash, as every major investor passed on their pitch that AI would need to use specialised chips.

That bet looks much better today, with competitor Cerebras enjoying a breakout IPO and even OpenAI building its own custom chip with Broadcom.

The product is still being tested with customers, so those orders haven't been proven at scale yet, but Etched is certainly one to watch.

This Week’s Art

We’ve covered quite a bit this week, including:

If you found something interesting in this week’s edition, please feel free to share this newsletter with your colleagues.

Or if you’re interested in chatting with me about the above, simply reply to this email and I’ll get back to you.

Have a good week!

Liam

Share with Others

If you found something interesting in this week’s edition, feel free to share this newsletter with your colleagues.

Liam McCormick is a Senior AI Engineer and works within the AI team at Bright. He identifies business value in emerging technologies, implements them, and then shares these insights with others.