AI for the rest of us

Welcome to the AI community for everyone.

Hello friends,

It’s been a fun week for Hannah entertaining her niece and nephew again. We did Go Ape on Wednesday (covered in mud), made a fairy palace on Thursday (covered in paint) and then played some extreme hide and seek on Friday (covered in bramble scratches and more mud).

Exhausted and bruised Hannah is now doing laundry and enjoying a quiet weekend. A weekend where there is no need to say “I’m going to go and read this in the other room where no one is screaming” only to have the screaming person follow you and scream louder. Massive love to all our readers who are parents surviving the school holidays.

Charles saw a T-shirt on his way into London that said, “I think I have seized the wrong day,” which made him laugh so hard he almost choked on his coffee. It has suddenly dawned on him that GOTO Copenhagen is five weeks away, he’s on holiday for one of them, and at GOTO he is teaching a brand new, as yet not fully written sustainability masterclass and giving a talk, and needs to keep multiple clients happy in the meantime. As a result he is a bit of a stressball, although thanks to there being huge amounts of blackberries around already, he has made his first blackberry and apple crumble of the season. Yum! There are still places available for the masterclass - he promises it’ll be finished in time! - Copenhagen is lovely, and GOTO looks amazing, so do sign up if you can make it. Honestly, it's hard to imagine a more important topic in software right now.

Have a fabulous week!

Hannah & Charles

What’s Hannah reading this week?

Clay is an AI company, their most recent marketing campaign explains how their account research agents can help sales teams find signals in the noise of so many customer calls. However, this AI product company has strong feelings about using AI tools at work. Anyone who prioritises thinking and communicating gets a standing ovation from me so I LOVED reading Clay’s AI writing policy.

Varun Anand, Co-founder of Clay, shared his thoughts on LinkedIn whilst also giving credit to the author of the policy Sophie Alpert.

1. You must stand behind every idea and sentence

It is your responsibility to make sure that the entire document is representative of your own thoughts before you share it.

2. Writing is thinking

Spending time on the writing process teaches you more about your topic. If you circumvent this process, you will walk away with a poorer understanding of the subject matter.

3. More time should be spent writing a document than consuming it

If you generate a document from a short prompt then ask your readers to go through the longer output, you are disrespecting their time. They can talk to ChatGPT themselves if they want to.

4. Longer is not better

AI makes it easy to generate long docs, and it loves padding them with sentences that say nothing. If you're producing docs from a short prompt, consider just sharing the prompt.

Although I spent the majority of this week entertaining tiny humans I did find time for a few excellent conversations on AI driven software development. The team at Tessl have started to share more about their own software factory, called kikimora. I can't wait to hear more about their story.

“We’re happy to be the guinea pigs, let’s see if this works”

“We intentionally ran really fast, straight into many walls”

Tessl have the incredible ability to keep evolving their product and company alongside the emerging platforms, practices and patterns of agentic software development. It started with “spec-driven development” and it’s now all about the “dark factory”. Dark factories is a term coined in manufacturing where some factories are so highly automated that you don’t need the lights on, there are no humans that need light to work, only machines working in darkness.

In conversation this week I reflected on the changing roles of product managers and engineers if we’re moving towards a world of dark software factories. Typically product managers handle a lot of coordination and feature delivery, with product directors and above looking at more strategic product positioning, direction and business development concerns. As the coordination overhead vanishes and the speed of feature delivery accelerates these roles could also collapse. Every product person needs to be strategic, because their role needs to be more focussed on what and why, and less about who and when.

For me this sounds like a perfect future. My career would have died a thousand times if you asked me to be a product manager in the backlog management / feature management sense. I don’t write tickets, stories, PRDs… I hate that work. I LOVE product strategy though. Finding problems, gaps and opportunities, and sizing them up. Where should we spend our time? What direction are we heading? Who competes with us and how do we win? I can’t get enough of that work!

So, for all the product people like me, I think dark factories could be a fast track to the next level. When the coordination overhead starts to shrink, it’s time to step up!

What's Charles reading this week?

I studied English literature at university rather than computer science, and made my way up in the field by doing any work I could. After a stint in desktop support, I found myself working for an investment bank at the height of Visual Basic for Applications/Microsoft Excel/Microsoft Access. Not infrequently someone who was computer-savvy but not a programmer would build an application for their own use using one of the Microsoft Office programmes, and then find it getting adopted. A lot of my work was being asked to “productionise” these applications. Most of my fellow programmers hated this kind of job, but I actually rather liked it, perhaps because I still remembered the excitement of my first “hello world” experiments. In this excellent blog from ThoughtWorks CTO Rachel Laycock, she describes similar feelings and how she sees this playing out again with GenAI.

“The moment that application becomes something the business depends on, the questions change completely. Is customer data protected? What happens when a dependency fails? Can someone else understand this system in two years’ time? Will it survive an audit? Can it cope with a thousand times more users than it has today, what about millions in one day? How will we know something is wrong before our customers do? Those questions don’t show up in a demo or in the build phase at all unless an experienced engineer is in the room. I certainly wasn’t asking them when I was building my first apps. I only cared about features!”

I found myself reaching for Claude on Monday to debug a weird rendering bug on my own work website, which turned out to be caused by a URL containing the word tel clashing with a CSS selector, which is the sort of thing that drives me mad. Claude did a good job of helping me figure it out.

In the process I also had to do battle with the increasingly flakey GitHub, which was experiencing another bunch of problems at the time, these ones caused by saturated load balancers, a faulty autoscaling policy, and a "latent retry bug in Visual Studio Code”. As ever I feel nothing but sympathy for the people working on it, but doing its best impression of a YoYo/gooseberry in a lift/the sea is pretty concerning, not least for them. As GitHub went up and down, SpaceX-owned Cursor announced an early beta of Origin Code Hosting (getting access to lots of source code is helpful for training). OpenAI and several other startups are also working on GitHub rivals.

The article that caused me all the problems was this one for LeadDev, which is based on some research from Google on Gemini usage at work. The short version is that although a lot of people are using Gemini at work (and by extension other AI tools), they aren’t using it in particularly significant ways outside of some very narrow use cases, like programming. It is obviously early days, but my sense is that people in software continue to overestimate how useful LLMs are in business generally, and also how difficult it actually is to transform a business around more automation.

The BBC’s Lucy Ashton reports that patients in Rotherham are struggling with a new AI GP receptionist because it cannot understand the Yorkshire accent. The system, EMMA from a company called QuantumLoopAI, is being used nationally in the UK. “Healthwatch Rotherham said some patients could not get the system to understand them as it does not respond well to variations in speech such as regional accents or speech impediments,” Ashton writes. Voice is a particularly hard problem in computing, because there are a bunch of factors, including that real-world acoustic conditions are messy - think background noise/echoing/reverberation - plus needing to account for accents, dialects, speaking rate, pitch, mumbling, stammering and so on. All of these affect how well it works, none of which come up in text-based chat. But I’m still surprised more testing wasn’t done ahead of rolling it out. I wonder if similar problems occur with the plethora of AI note taking apps that seem to be springing up in surgeries, hospitals, and pretty much everywhere.

Somewhat related, I'm feeling grumpy that Calendly, the scheduling service I use, has announced that, like it seems basically every company in the world, it is launching an AI assistant and a Notetaker. “The New Calendly. More than just scheduling,” their email says. The thing is, the only thing I want Calendly to do is scheduling. Just really well. Basically this will mean I have to pay more for the service I need because of a bunch of things I absolutely do not want. Sigh.

Last Friday, Google won an auction to acquire a huge amount of Spirit Airlines data, the low cost airline that was shuttered in May, paying $10 million reportedly for AI training. Google is also covering the costs of hiring a third-party service to scrub the data to comply with consumer privacy laws. An alternative bid with similar terms was accepted from Mercor for $7.5 million, should Google fail to follow through on the purchase.

Writing for Ars Technica, Ashley Belanger says that “Google’s privacy commitments protecting Spirit customers don’t seem to extend to former Spirit workers. Panicked by the deal, former flight attendants are now rushing to object, worried that Google has not agreed to strip confidential information that workers fear could possibly be used by Google or a third party to link them to scrubbed data.” Flight attendants are understandably concerned. Belanger links to a court filing from Tuesday from the Association of Flight Attendants (AFA), a collective bargaining unit that represents Spirit workers, that states, “The privacy architecture of this transaction is consumer-facing; its payload is disproportionately employee-facing.” It continues, “Hence, the employee data is far more confidential than the customer data, yet receives far less protection than the customer data.”

You can see why they might be concerned. The Register’s Simon Sharwood points out that Google’s trove includes:

“100 million emails and 500 million items from Microsoft Teams, 17 million OneDrive files and 20.5 million items from SharePoint. The search giant also now owns over 30 million recorded customer service calls, and more than 15 million customer service chat records. 600,000 ServiceNow tickets are another element of the collection, along with 13.7 million active emails addresses from Oracle’s Responsys marketing application, and details of 11 million sales of in-flight Wi-Fi services. There’s also operational data describing over 763,000 flights, five million crew pairings, more than 1.2 million fuel slips, and records describing purchases of 787,452 parts.”