AI for the rest of us

Welcome to the AI community for everyone.

Hello friends,

It was amazing to see so many of you in person for our meetup this week. We ran out of pizza, drinks and chairs!!! We love to see the community growing like this, but as the organiser it means Hannah has some work to do. The meetup needs a new home that can accommodate more people.

Do you know any companies with event spaces who can take 100-150 people? If you do, we'd love an introduction.

Charles was in Epsom on Thursday to watch his brother-in-law play Belle's father Maurice in Disney’s Beauty and The Beast at the Epsom Playhouse, in a production directed by Charles’ sister-in-law Lian. It was superb.

This week Hannah is travelling to Malmö in Sweden for Øredev and then immediately flying to Salt Lake City, Utah for Kubecon. If you’re attending either of those events make sure you come and say hello!

Have a wonderful week,

Hannah and Charles

At this week’s meetup we hosted our usual open mic slot. Here’s what the community wanted to share:

A free Perl and Raku workshop on Saturday 21st Nov (Register)
Make better decisions, work with Florence Stanley, Founder of Why and Wherefore
Paul Markham wants you to own your audience, check out followers.email
Enter the competition “Show us your Chinatown , 2046”
AI for the rest of us CFP closes on TODAY: Propose a talk or workshop

Convert Agent Failures into Evals

Nick Ferrell shared his experience at the meet-up this week. Read more about how to use evals in your agentic system design.

Read The Blog

What’s Hannah reading this week?

Before we get into the serious AI news of the week I wanted to share this silly little tune that made me laugh: LGTM - Looks good to me.

Yes, it’s AI generated, sorry Charles!

“Aaaah...” - Charles

This week at our meetup in London Tara Ojo Agyemang from Google introduced us to WebMCP. I had assumed that WebMCP was part of regular MCP, but it’s actually completely separate. WebMCP helps AI Agents reliably interact with your website or online app.

This is a new reality we all have to come to terms with. A lot of the traffic on the internet today is AI Agents doing things on behalf of their human users. It will only increase over time. Investing in Agent Usability is not optional if you want to remain relevant.

After the talks we had a great discussion about what the agentic web means for us as users and builders. If I’m building a website or product I need to be thinking about the users who are interacting via an Agent intermediary. How do I delight those users? This is a new dimension to UX that’s still evolving.

Personally I want to connect with humans, build trust, be memorable… so I need to figure out ways to get past the new doorman, the user’s agent, in a way that helps create that human connection. Quite a challenge.

This week at BIMP HQ we’ve been super impressed with GPT-6.1 Terra, it’s just so fast. We still use the larger models like Astra and Sol for planning and review but Terra is smashing it right now for execution tasks. It’s incredibly cheap and highly accurate.

In cyber-security this week we’ve seen the largest crypto heist of all time (Binance lost $6.3bn), but it’s not clear yet if AI was involved in this breach. Meanwhile popular coding agent Cursor came close to a cyber-security disaster when one of its popular Claude Code extensions was compromised with malware. Wired reports that although thousands of developers downloaded the tainted code, most of them were safe due to a flaw in the malware itself.

This flaw in the malware makes me wonder if this was a novice hacker, using coding agents to build something almost right but not quite… and then using the agentic coding ecosystem to distribute it very effectively.

The Binance attack used a compromised developer laptop to gain access, and if this Cursor / Claude attack had been more successful it would have infected thousands of developer laptops. This has become a powerful pattern for attacks because developers have all kinds of access and it’s that access that the hackers want. They want to use the developer laptop to gain access to production systems, to databases, to code bases and to CI/CD systems.

For me the conclusion is pretty clear. We have to assume every developer laptop is compromised and build our access controls and our dev tools around that assumption. There is no other sane option right now.

What's Charles reading this week?

It's been quite a week.

On Monday, Microsoft announced a buyout, which it calls a Voluntary Exit Offer (VEO) programme, for up to 60,000 employees, or roughly a quarter of its workforce. In a memo to employees Amy Coleman, Microsoft’s executive vice president and chief people officer, wrote that the VEO “gives eligible employees who may be considering a future outside the company a new choice to make that transition now with support”.

Employees at the vice president level and below on select teams in the United States, Canada, India, and the United Kingdom will be eligible, according to the note. A source told Reuters that “the voluntary exit offer will impact key divisions, including AI experiences, Copilot, finance, human resources, Xbox and Windows + Device”. Bloomberg has a similar list of impacted divisions.

This isn’t the same as a redundancy programme though. A spokesman told Fortune, “If you look across the industry, you see that similar voluntary programs have around a 3% to 5% uptake. In our case, that feels accurate to us; we anticipate an under 5% uptake.” But it certainly is an eye-catching figure, particularly given that on Wednesday the firm revealed its profits for the three months to the end of September jumped to more than $36.6bn (£27.5bn), according to the BBC.

Also on Monday Anthropic filed its S-1, effectively a prospectus for investors, as it gets ready for its IPO. It will list on the Nasdaq under the symbol ANT, and will be the first fully AI-focused company to go public.

The S-1 is extremely long - well over 300 pages - and whilst I’ve waded through as much as I can I haven’t had time to read the whole thing. The rest of this is a long summary of what I’ve discerned.

The S-1 shows $16.3 billion in revenue, and a $80.7 billion net loss, although $72.4 billion of that loss is due to an accounting charge to reflect the increased value of convertible notes Anthropic has issued (i.e loans from early investors and strategic partners that automatically turn into company stock).

IPOs are as much about marketing as fundraising. The firm emphasised its growing enterprise business, noting that more than 11,900 customers spent at least $100,000 in the past twelve months; more than 1,900 spent $1 million or more; and roughly 250 customers spent $10 million or more.

Anthropic’s engineers are now contributing 45 times as much code to its codebase as they were in 2024, the company said. “Over 90 per cent of the contributions to our codebase in September 2026 were written by Claude,” according to the S-1. The firm also states that “Anthropic remains small for its reach. Each Anthropic employee supports over $6.6 million in trailing 12-month revenue as of September 30, 2026, which is meaningfully higher than the average for the largest public U.S. companies.”

94 pages of the S-1 are devoted to risk factors. The filing warns about the possibility that sophisticated AI could be abused by threat actors to launch cyber attacks, develop novel biological weapons, or by authoritarian governments to surveil their citizens. Anthropic also pointed to the risk that it could lose control of the AI models it develops: “Sufficiently advanced AI systems could, if not developed, deployed, and operated with appropriate safeguards, oversight and security measures, pose catastrophic or existential risks to humanity.”

Copyright is also mentioned, with the firm saying “If we cannot rely on the fair use doctrine to access copyrighted work for training our models, we may be required to pay significant fees to obtain a license to such training data,” or if an adverse ruling prevents them from distributing models trained on copyright material “retrain affected models from scratch at substantial cost”.

It also highlighted the volatility of government action, specifically noting the Trump administration’s decision to designate the company as a supply chain risk.

There is a lengthy section on power needs and water, starting on page 34. I’m finding it hard to summarise, but broadly it frames these and other environmental concerns as sources of cost, delay, and regulatory risk.

The share structure seemed to me very complicated (although I’m not expert on this) but the main thing I did note is that the IPO will create one share of Class F stock, held by an LLC owned by the seven co-founders. That one share carries 50.1% of the vote and no economic rights. Fortune’s business editor Nick Lichtenberg has a detailed breakdown if any readers want the full details.

The intended listing date hasn’t been disclosed, but rumours suggest it could come as soon as 16th November. Anthropic’s employees should do very well out of the deal. The AI company’s board has granted employees about 91 million restricted stock units with a fair value of about $5.8bn. The philosophy appears to be “we might bring about the end of humanity, but we’ll get really rich first”.

Updates

Early Bird Tickets

Our next 2 day conference is set for 25-26 February 2027. Grab your early bird ticket for £150.

Get Your Ticket

October Meetup

Our October Meetup is scheduled for Tuesday 27th. We'll be talking about The Agentic Web - WTF is that?!

Between Jobs?

Interested in a more human approach to career change and AI? Join the Surviving The Gap community for practical workshops, useful resources and early access to Career Core.

Learn More