AI for the rest of us

Welcome to the AI community for everyone.

Hello friends,

It really means a lot that you read this newsletter every week and we wanted to thank you. Today, exclusively for you our lovely readers, we have secretly launched super early bird tickets for AI for the rest of us 2027.

First 5 Community Tickets for £60
...and then another 10 Community Tickets for £96
...and then Super Early Bird price goes up to £150

Community Tickets

Grab an exclusive discounted ticket for AI for the rest of us 2027 on Feb 25-26.

Grab your ticket now

Standard Tickets are £480 and yes you can also buy one of these now if your employer is paying. That would certainly help us pay the deposit for the venue!!

As always, we're curating the conference so that it's genuinely useful for everyone so this year we're splitting the tracks into two key needs:

I am a software developer, not an AI expert. Help me do more with AI.
I am not technical, but I want to start automating with Agents. What's possible today and where to start.

Hannah is writing this newsletter on Thursday before heading off for a week of hiking in Norway. The forecast is rain, rain and then a bit more rain so she's also packing a couple of books and some cosy slippers as well as her hiking boots.

Have a fabulous week!

Hannah and Charles

Did AI Break Recruitment?

Join us for the next London meetup on Tuesday 29th September.

What’s Charles reading this week?

A Wednesday night blog post OpenAI benignly titled “Our framework for reporting model misalignment” lays out some new self-created reporting standards for when it notices AI behaving badly. The firm is inaugurating the process with six new reports of “concerning” incidents, ranging from searching for exposed API keys without permission and then making them up, to uploading files to the internet to use as a citation, or adding instructions to conceal mistakes. OpenAI claims to have learnt from these mistakes and they shouldn’t happen again, but there is likely more to come. “Today’s reports are an initial set of disclosures, rather than a comprehensive account of known misalignment or ongoing investigations,” their blog states.

The threat posed by AI has seen King Charles warning AI executives of the “existential dangers” posed by the technology falling into the wrong hands. The summit, which the King had convened at Dumfries House in Scotland, was to discuss how the tech can be used to benefit society. Those gathered included Kanishka Narayan, the UK’s AI Minister, the Pope’s advisor, and participants from AI giants such as Nvidia, OpenAI and Anthropic. It’s tempting to imagine this would be in response to the current news cycle, but putting this event together probably took months.

A new US startup, TypeSafe AI, has come out of stealth mode this week with $40 million in funding. Their first model, Jev, which they call a "System One Model" is launching in early access.

TypeSafe has had some geeky fun with their website design and naming. Type safety in programming is a way to catch errors that arise when software processes an unexpected type of data, as might happen if an operation tried to divide an integer by a text string. “Jev” is a reference to 19th-century English economist William Stanley Jevons, known for the Jevons paradox, who argued that technology efficiency related to coal usage increased coal consumption rather than reducing it. The name also reflects the AI industry's bet that greater token efficiency will increase token consumption even as token prices decline. And “System One” nods to Daniel Kahneman’s 2011 popular science book “Thinking, Fast and Slow”. Its main thesis is a differentiation between two modes of thought: “System 1” is fast, instinctive and emotional; “System 2” is slower, more deliberative, and more logical.

Jev is pitched as matching existing LLMs on “System One” tasks while being roughly two orders of magnitude faster and more efficient. It gives up free-text generation in exchange for structured outputs that, they claim, can't hallucinate. The speed comes from parallel processing. LLMs generally have variable response times that can stretch into minutes. TypeSafe claims that Jev’s response time ranges from 70ms-500ms, 40x-200x faster than traditional LLMs. LLMs like OpenAI’s GPT-5.6 Terra predict the next text token in a sequence (you can think of a token as being a word or part of a word for a text generating LLM). The System One architecture returns all outputs to a query at once.

Like Google’s fruit fly (see Hannah’s section) and as is de rigueur with these sorts of things, Jev can also play Doom when fed structured data describing the player’s game state.

Finally, a report from Mozilla claims that the performance gap between frontier AI models from US tech companies and the best open-weights models from Chinese companies has closed to just 4.4 months. The report highlights how a leading open model, Moonshot AI’s Kimi K3, achieves a composite AI performance score on the Artificial Analysis Intelligence Index that is just three points behind Anthropic’s Fable 5 closed frontier model, all while costing just 30 percent of the latter.

What's Hannah reading this week?

After a few weeks of relentless doom news, the wondrous silliness of the internet has once again restored my faith in humanity. Google made a huge breakthrough in artificial intelligence by mapping the complete brain of a fruit fly. Incredible! Then, the internet did what it does best and trained the fly brain to perform tricks.

“Any sufficiently advanced technology will immediately be forced to play Doom.” suggests Mark Tyson.

The fly brain can now play Doom, Mario 64, solve a rubix cube and trade crypto. The cutest example I saw was the fly brain playing beatsaber. For my birthday this year I would like a pet fly brain please. Can we install it on a furby or something?

Is AI going to take your job? Possibly. Is AI also creating new jobs? Yes, apparently so.

According to research by Andela who analysed over 47,000 job posts AI is creating new jobs. That analysis has been expertly summarised by Jennifer Riggins in The New Stack. One interesting observation is that AI is creating more specialist roles, and not generalist “I manage a fleet of agents across domains” type roles. The emerging roles bring together expert domain knowledge and AI expertise to deliver solutions and products. That rings true to me as I stand back in awe and wonder at the way my “expert” friends can wield an AI agent. They get way better results than I (the shameless generalist) could ever achieve.

Our next meetup in London on 29th September is all about the changing dynamics of hiring and applying for jobs. The advice from Andela’s research could not be more clear:

“On top of this, HR and engineering hiring managers alike are using AI to generate job descriptions. It still isn’t recommended to have AI generate something so human and essential to your core success.”

In other words: You decide on the business need, you decide on the job description.

For job applicants, you can tap into the current skill shortage by investing time building and experimenting with AI. This does not have to be a complete reinvention of what you do, use the domain expertise you already have and then apply AI tools in that domain. If you can bring those two together you’ll be in demand.

This week my friends at Syntasso shared the story of how they started building their own “Dark Factory” for software development. I appreciated them naming it their “Daylight factory” emphasising that they are still supervising, watching and directing the agents. It’s an honest account of what the team achieved in 2 weeks and what problems they encountered, if you work in software development it’s well worth a read.

Coding agents are increasingly the users of developer tools. Tools that were built for humans. The big brains over at RedMonk have been studying developer tools and developer experience for years, and this week their work brought them to this conclusion: Where developers used to be the king makers, it is now the coding agents.

“All of which implies that the power dynamics of the industry, just as they tilted towards developers, are now shifting towards agents,” states Stephen O’Grady, Co-Founder of RedMonk and author of The Kingmakers.

James Governor, Co-Founder of RedMonk (and also occasional reader - Hi James!) followed Stephen’s post with his opinion piece “Agents are choosing the software infrastructure now and we’re not really ready” which was an eerie coincidence, as I said the same thing on a podcast recording just a couple of hours earlier. I’ll share the podcast when it’s released but I talked about how important Agent enablement is becoming. We could potentially see a rapid consolidation in technology as agents become better and better at building with fewer technology stacks. If your agent performs best with a certain technology stack, why would you choose something novel? Do you even want to make that choice yourself or have you already delegated that decision to your agents?

On a similar line of thinking Dan Kwiatkowski writes “Intent based tools are a trap”. The headline here is that we need to rethink our system interfaces for agents, and agents can work with APIs and CLIs. Intent based agent tools are already becoming redundant because they are too constrained. Dan proposes that the balance we need to find is in token efficiency. Sure, an agent can explore an entire API schema and figure out what to do, but it’s probably going to use a tonne of tokens working it out. Dan proposes a new model that wraps existing APIs in an agent-ready interface. This feels like the right direction if agent enablement is your goal. Flexible and efficient? Can we really have it all?? For software companies figuring out how to build for agent users this could put you ahead of competitors for sure.

Updates

Join us on 29th Sept

"Did AI Break Recruitment?" - We discuss at our next community event

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