AI for the rest of us

Welcome to the AI community for everyone.

Hello Friends,

We hope you all had a good Easter. Charles’ eldest was home from University and, having spent Easter Sunday with Charles’ mum, they spent a lovely day on Monday walking through some truly spectacular local bluebell woods.

Hannah has also been enjoying some down time with her family, travelling back to The Forest of Dean for the Easter weekend. Hugging trees, climbing hills and walking a total of 50km. I’m a woodland creature!

Following the Artemis II mission to the other side of the moon has been a highlight of the past 2 weeks. Outside of the war news doom feed and AI thrashing, the Artemis II mission has provided some much needed awe and inspiration about what technology can do and how it can help us reach further - just look at these pictures! Maths nerds are the best (says the nerd with the maths degree)!

For those of you based in London, our April meetup is scheduled for Wednesday 22nd. This month we’re talking about “Claude Code for the rest of us”. We’re diving into how knowledge workers outside of software development are using Claude Code to help them in their roles. Kicking off the meetup Liam Darmody will be demoing the agents he’s built to help him manage his business.

Have a wonderful week!

Hannah and Charles

Get the #1 Bestselling book 'From Cloud Native to AI Native' for free — normally $26.99

What’s Charles reading this week?

AI has made shipping code cheap. Verifying it is another matter. In my latest piece for LeadDev, I explore the hidden costs of the "prompt and review" model being rolled out across engineering teams, from developer burnout and de-skilling, to the deeper problem of code landing in production that nobody fully understands.

On Tuesday Anthropic released a preview of its new frontier model, Mythos, which it says is too dangerous for general release. Instead it will be used by a small coterie of partner organisations including Amazon Web Services, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks, for cybersecurity work.

“Our new model is so good, it’s too dangerous to release to the public” sounds like it could be marketing BS, but in this case it might be true. “During our testing, we found that Mythos Preview is capable of identifying and then exploiting zero-day vulnerabilities in every major operating system and every major web browser when directed by a user to do so. The vulnerabilities it finds are often subtle or difficult to detect. Many of them are ten or twenty years old, with the oldest we have found so far being a now-patched 27-year-old bug in OpenBSD—an operating system known primarily for its security.”

I imagine everyone saw that the entire Claude Code CLI source code leaked thanks to an exposed map file, but in case you missed it Samuel Axon, writing for Ars Technica, has a good summary. I was deeply amused to see Anthropic representatives using copyright takedown requests after the leak. Anthropic have been saying for a while that Claude Code is mostly coded by Claude, and I don’t think LLM-generated code is copyrightable at least in the US. Also my irony meter just exploded.

I want to share a link to Helen Lewis’ musings in her newsletter on the story of Alex Preston. He has been caught using AI to generate copy for a book review he passed off as his own, and published in a respected publication. I don’t have much to add, but I found myself nodding along in agreement to her opening paragraph:

Like a lot of writers, I’ve been thinking about what the acceptable limits of LLM usage are—both in terms of my personal comfort, and what I’m willing to accept from others without considering it ‘cheating,’ There are some functions that LLMs do that I’m fine with (spellcheck, interview transcription, natural-language version of a search engine) and some that are, in my view, totally appalling (using an LLM to generate text and passing it off as your own).

Lewis links out to a piece from The Spectator’s Sam Leith, who she says “rather marvellously, did a journalism and just emailed Alex to ask him WTF happened.” Alex’s reply is worth reading in full. “That someone with his reputation would risk trashing it by using AI in his writing at all slightly astonishes me,” writes Sam. “It is gloomy testament to how tempting these tools seem to be as a shortcut.”

I do understand the temptation. Writing is a skill increasingly devalued. I’ve just had one publisher I work with unilaterally cut my rate by 25% (I demurred), and I’ve just seen via LinkedIn that another organisation, Gradle, has laid off its entire marketing team saying they can be replaced with AI. The temptation to use an LLM as a shortcut to doing sub-standard work at pace exists. But where’s the joy in that?

My view is that the race to the bottom strategy that more and more publishers and companies are pursuing is deeply misguided. I recently suggested to John Cricket, on the Coding Chats podcast, that authentic technical writing from practitioners is genuinely valuable, arguably more so now than before given the amount of change AI is causing. I also think there's a diminishing market for generic writing. My instinct is that we’ll soon tire of badly written AI slop (whether that’s music, writing whatever) and desire something made by humans again. In the same episode I also talked about why I think the social media narrative that we don’t need junior programmers any more is, to put it politely, boloney.

You might remember that last time we reported OpenAI was allegedly abandoning side quests. How’s that been going? Well it has just bought TBPN, a tech video talkshow that gets views of 75-100k, and apparently had ad revenue of $5m last year and aimed at $30m this year: the price was apparently ‘low hundreds of millions’. OpenAI sure could use a better comms strategy (if anyone from OpenAI is reading I’m available, and my rates are very reasonable), but owning this particular channel won’t help it affect the public narrative. I guess the new focus on focus starts real soon now.

Along with a better comms strategy, it also needs a different leader, at least if The New Yorker’s loooong and unflattering profile of Sam Altman is anything to go by. Between them, journalists Ronan Farrow and Andrew Marantz interviewed over 100 people, including Altman, and the report they bring back from this effort is depressing; the words “lying” and “sociopath” are used repeatedly. I’m not sure we’ve learnt anything new here, but it does paint a picture of a man with what we might generously call extremely flexible ethical and political views. Here’s just one part of the article:

He’s unconstrained by truth,” the board member told us. “He has two traits that are almost never seen in the same person. The first is a strong desire to please people, to be liked in any given interaction. The second is almost a sociopathic lack of concern for the consequences that may come from deceiving someone.

The board member was not the only person who, unprompted, used the word “sociopathic.” One of Altman’s batch mates in the first Y Combinator cohort was Aaron Swartz, a brilliant but troubled coder who died by suicide in 2013 and is now remembered in many tech circles as something of a sage. Not long before his death, Swartz expressed concerns about Altman to several friends. “You need to understand that Sam can never be trusted,” he told one. “He is a sociopath. He would do anything.” Multiple senior executives at Microsoft said that, despite Nadella’s long-standing loyalty, the company’s relationship with Altman has become fraught. “He has misrepresented, distorted, renegotiated, reneged on agreements,” one said.

OpenAI also closed its latest funding round, raising $122bn at a $852bn valuation, and SpaceX filed for IPO, reportedly aiming at a valuation of $1.75tr (to date the largest ever IPO was Saudi Aramco in 2019, raising $29.4bn at a $1.7tr valuation). The New York Times reports that Elon Musk is insisting that banks that want to be included in the IPO (given the size, they will need a lot of banks in the syndicate) need to buy subscriptions to his ‘Grok’ also-ran deep-fake-porn generating LLM that he merged into SpaceX earlier this year.

Finally from me this week Oracle laid off an awful lot of people - perhaps as many as 30,000 - notifying them all by an email which, one ex-employee told Johnson O'Ryan at The Register, basically read “Thank you. Go (expletive) yourself,” as is the corporate American way. The company has several declining legacy businesses such as ERP and the main database platform, and is pushing into data centres on the back of the AI boom. Initially it was borrowing heavily against the cashflows from the legacy businesses, and now it’s doing layoffs in that legacy business as well to fund the DCs.

What's Hannah reading this week?

One of the board games we like to play with my niece and nephew involves giving clues in the style of a caveman. We love it! It turns out that caveman speak is not just hilarious but can also save you money! A 16 year old developer named Om Patel taught Claude to speak in caveman, apparently saving up to 75% of tokens. This has of course led to some brilliant experiments on YouTube, reddit and even a Claude skill so you too can enjoy a less eloquent, more efficient neanderthal experience from your AI tools.

“Why waste time say lot word when few word do trick?”

Popular issue tracking tool Linear made a bold statement this week: “Issue tracking is dead”. That’s their business and they’re saying it’s dead. Linear have come to the same conclusion as the rest of us, that many of the methods we use to manage software development today are becoming redundant.

This is happening everywhere. Even with my brand new product BIMP, I find myself trying to balance the pain points and friction of today’s organisations and a desire to “skate to where the puck is going".

Today, coding agents are installed in more than 75% of Linear’s enterprise workspaces. In the last three months, the volume of work completed by agents grew 5x, and agents authored nearly 25% of new issues

Linear have made a decisive move to accelerate towards the future state instead of clinging onto the way we do things today and I admire their courage. I expect we’ll see more of this as the productivity tools we use today become sources of friction tomorrow. My own experience supports this prediction: it would take me more time to write tickets than to do the work so I don’t do it anymore. We need tools that help us ideate, test, decide and act. Ticket shuffling and backlog grooming is not the way.

On the subject of experimentation and testing ideas Andrej Karpathy’s experimentation loop (or Karpathy Loop) provides some inspiration about how we can use AI to discover more, faster, and make better decisions as a result. I can’t wait to try these methods in product development.

It’s not just issue tracking that is dead. Apparently the SDLC is also dead. In this blog post from February Boris Tane lays out the ways that our current processes and practices fail. There are so many insights in this article that I’ve listed them below for you so you don’t miss any!

Every stage of the SDLC is collapsing, we can no longer tolerate handoffs and friction created by siloed tools
If your “requirements” are just context for an agent, then the ticketing system isn’t a project management tool anymore. It’s a context store. And it’s a terrible one. (I wonder if Linear read this post and changed their strategy!!)
Agents write tests alongside the code. Not as an afterthought.
The pull request flow needs to go. (I couldn’t agree more, but we need to solve for peer review and shared ownership another way!)
The deployment “stage” doesn’t just get automated. It becomes an ongoing, self-adjusting process that never really ends. Agents can handle complexity like feature flags, canary releases, progressive rollouts, automatic rollback triggers with ease, no specialist tools or teams required.
The observability layer becomes the feedback mechanism that drives the entire loop.

OpenAI took me by surprise this week when they published “Industrial Policy for the AI Age. Ideas To Keep People First.” Is this just PR or do they really want to share prosperity broadly? I do truly hope that we will see the frontier companies taking their responsibility seriously.