AI for the rest of us

Welcome to the AI community for everyone.

Hello friends,

Join us for our monthly meetup in Central London on Wednesday 18th March. This month we’re focussed on Agentic Automation with Kiran Pendem and Minisha Goel. We’ll be discussing when to use Agentic Automation and how powerful agents like Claude Code can help everyone, not just for developers. As always, it’s sure to be a practical and inspiring evening. Massive thanks to our new meetup sponsor Leapter for supporting the community!

We were saddened to learn of the passing of Sir Antony Hoare, a foundational architect of modern computing, who died on 5th March 2026 at the age of 92. He is best known as the inventor, in 1959, of the Quicksort algorithm which he implemented at Elliott Computers in 1960 for a bet with his boss, as recounted in this lovely personal reminiscence by Jim Miles. In this interview clip, he explains how he implemented it. Quicksort has been used in implementations of the Unix sort command, in various language libraries, and as a Hungarian Folk Dance.

Hoare won the Turing Award in 1980, and the ACM page about him has several video clips from a 2021 interview. He gave a famous lecture titled "The Emperor's Old Clothes" in which he said:

“I conclude that there are two ways of constructing a software design: one way is to make it so simple that there are obviously no deficiencies, and the other way is to make it so complicated that there are no obvious deficiencies.”

He is eminently quotable. On ALGOL-60: “Here is a language so far ahead of its time, that it was not only an improvement on its predecessors, but also on nearly all its successors.”

There have been some very good and touching posts in his memory, including "Commemorating Tony Hoare, Inventor of QuickSort," and "In Memoriam: Sir Antony Hoare (1934–2026)" by Dag Spicer at the Computer History Museum.

Enjoy your weekend and we’ll hopefully see some of you in person on Wednesday.

Hannah and Charles

What’s Charles reading this week?

I've written a 3rd article in the sponsored series with Kin Lane for The New Stack. I'm continuing to enjoy working on this series. The core argument, drawn from systems thinker Stafford Beer, is that a system's true purpose is revealed through its behaviour, not its design spec. That means before you trust an AI agent in production, you need robust sandboxes and contract tests to understand what it actually does.

If you’ve ever written a line of Python, you’ve probably used chardet. It’s the library that figures out which character encoding a file is using. For years, it’s lived under the LGPL licence—the "strictly business" licence that corporate lawyers tolerate but don't exactly love.

But this week, maintainer Dan Blanchard decided it was time for a glow-up. Or a heist. Depending on who you ask. Blanchard released chardet 7.0.0, declaring it a ground-up rewrite under the much friendlier MIT licence. He didn't lock himself in a dark room for six months to write it though, instead using Claude Code to do the heavy lifting.

Then a ghost appeared. Mark Pilgrim, the original author who famously "disappeared" from the internet in 2011, emerged from the void to file a GitHub issue. “In the release 7.0.0, the maintainers claim to have the right to ‘relicense’ the project. They have no such right; doing so is an explicit violation of the LGPL.”

The issue here is that "clean room" design requires the person writing the new code to have zero exposure to the original. Claude, however, has definitely "read" the chardet source code. In fact, it’s probably memorised it. Using Claude to rewrite chardet is like asking a parrot to "summarise" a poem it’s heard 10,000 times—is it a new poem, or just a very talented bird infringing on your copyright?

Simon Willison points out that the structural similarity between the versions is less than 2%, but the legal similarity might be 100%. We’re about to find out whether "The Robot Did It" is a valid legal defence.

Last year Superhuman, the company behind the writing software Grammarly, released “expert review”, an AI tool that presents editing suggestions as if they come from established authors and academics, both dead and alive, none of whom consented to have their names appear within the product.

It’s fair to say writers weren’t terribly impressed.

Casey Newton drily noted, “I’ve long assumed that before too long, AI might take my job. I just assumed that someone would tell me when it happened”. With typical restraint tech journalist Kara Swisher, whose advice the feature claimed to offer, told Casey Newton over text that her response to execs at Grammarly would be, “You rapacious information and identity thieves better get ready for me to go full McConaughey on you. Also, you suck.”

Maureen Ryan meanwhile penned a vivid, visceral open letter. “I am writing a book right now, a really challenging endeavor that no doubt someone in Silicon Valley will think it’s fine to steal the day it’s published…I do not want to hear even the slightest amount of bullshit about how you didn’t need to ask for permission to use our names and use our identities in your review function. You just wanted to steal, so you stole. You wanted to take things without permission, so you did.”

Following the, you would think, entirely predictable backlash Superhuman made a sudden reversal. In a Wednesday LinkedIn post, CEO Shishir Mehrotra publicly apologised, saying that “over the past week, we received valid critical feedback from experts who are concerned that the agent misrepresented their voices”. “We hear the feedback and recognize we fell short on this,” he added, neglecting to mention that the firm is facing a class action suit. Julia Angwin, an award-winning investigative journalist who founded The Markup, a nonprofit news organisation that covers the impact of technology on society, is the only named plaintiff in it. Writing for Wired, Miles Klee notes the suit “does not call for a specific amount in damages but argues that damages across the plaintiff class are in excess of $5 million”.

There’s a wider observation here, which is that Silicon Valley is a bit of a bubble, and sometimes loses touch with how the rest of the world views the technology it builds. That perhaps helps explain this extraordinary clip in which Palantir CEO Alex Karp says he thinks his AI technology will lessen the power of “highly educated, often female voters, who vote mostly Democrat” while increasing the power of working-class men. “This technology disrupts humanities-trained—largely Democratic—voters, and makes their economic power less. And increases the economic power of vocationally trained, working-class, often male voters,” Karp said. “And so these disruptions are gonna disrupt every aspect of our society. And to make this work, we have to come to an agreement of what it is we’re going to do with the technology; how are we gonna explain to people who are likely gonna have less good, and less interesting jobs.”

Karp, like Musk, is a ludicrously overpaid, outspoken CEO of a richly valued tech company with a cult following and an unsettling store of political power, who uses outrage to drive media attention and inflate the stock price. That still doesn’t guarantee it’s a durable model for success, but Palantir is being awarded UK government contracts.

Since we started with Hungarian folk dance it seems only appropriate to end on a music note. Here is an interesting LinkedIn analogy (hat-tip once again to newsletter reader Chris) between AI and the evolution of pop music, in the broadest sense, which seems designed to appeal to me. It’s short and worth reading and reflecting on.

You can extend the musical part of the observation to classical music as well. The pipe organ's evolution spans centuries, but it was when electric motors replaced manual bellows in the late 19th century, that organists gained unlimited, steady wind pressure, allowing composers like Messiaen and Dupré to write massive, sustained fortissimos that would have been physically impossible before. Bartolomeo Cristofori's invention, the pianoforte, replaced the harpsichord by using hammers instead of plucked quills, allowing dynamic control for the first time. Beethoven pushed early fortepianos so hard he broke strings regularly — his later sonatas were essentially written in anticipation of better instruments. Liszt's thunderous virtuosity was only possible once the iron-frame Steinway could survive it. The celesta, invented by Auguste Mustel, so enchanted Tchaikovsky he kept it secret until The Nutcracker (1892) premiered so he could be first to use it.

It remains to be seen whether AI will allow us to create anything really new.

What's Hannah reading this week?

If you’ve ever worked in product or software engineering you’ll understand the value of feedback loops. We build tests to make sure we don’t accidentally break things. We continuously monitor our system health. We analyse user behaviour. We gather feedback. We iterate, experiment, respond and over time we improve.

What if we asked agents to do that? Could they self improve? Could they iterate, experiment and respond faster than us?

This post is written by an OpenClaw agent called Markus Downe, sharing its journey of self improvement. Macey Barker, a community engineer from Tessl set her agent Markus Downe a challenge:

“Markus has been trawling Moltbook, and the human internet, for evidence of and insight about various agent failure patterns. He has also been developing hypotheses for how to remedy them, and designing and running experiments to see if he’s right.”

Following the agent's own research and experimentation Macey applied the new skills to Markus, and found that they work. Her OpenClaw agent had improved itself.

On a larger scale Polsia claims to be an AI that runs your company while you sleep. An entirely agentic operating model for a new business. Polsia is currently running 3,923 companies autonomously, iterating and improving performance and ROI. You can watch Polsia working in real time on this adorably crappy looking dashboard. If the numbers here are to be believed then Polsia is having some success, reaching over $4m ARR.

Chatting to Luke Marsden (CEO of HelixML) on a wine fueled live stream this week, we talked about Polsia and debated what the future of business might look like if we can use agents to continuously experiment, iterate and improve all aspects of our companies, not just the software. He’s started to experiment with this at Helix and I can’t wait to see what happens.

A small team can go a long way if they’re prepared to unleash the agents. Agents are faster, don’t get bored with repetitive tasks and they can multi-task - you just create more agents! But through this extreme automation do we risk building undifferentiated cookie cutter businesses? Or worse, do we build businesses that attract attention and deliver short term revenue at the cost of everything else. As Seth Godin once said “Enough A/B testing will turn any website into a porn site.”

Geoffrey Huntley, the engineer who invented the “Ralf Loop” shared his worries this week in his post “Software development now costs less than than the wage of a minimum wage worker”. He predicts layoffs, personalised software eating SaaS and a lot of weeping VCs (my words not his). The shift in software engineering is already underway, but it would be foolish to think that other professions won’t follow. This is a topic I’ll be covering at QCon in London this week but with a slightly more positive spin. If you’re at the conference do come and say hello!

One thing feels certain to me: there has never been a better time to build a business. Starting now gives you a killer advantage. You can build a lean, AI-native business from day one while your competitors wrestle with their "transformation" plans and restructures. That’s certainly my plan, and next week, I’ll be sharing some big news in the newsletter about what I’ve been building.