I’ve been using AI for over two years, progressing from clever auto-completion through vibe coding to agentic engineering. Recently I iterated for a couple of weeks over a prototype application. The biggest problems came down to a single word, a term I’m found myself using more than any other over the last few weeks. And it’s not exclusive to a single project. It’s a term that’s been used across coding work, research work, everything.
When I was at school in the late 1980s, our school got a computer. Just one, in the library. I remember using it for some document I wrote, I can’t remember what, on a floppy (yes, actually “floppy”) disk. At university we had computer labs, but I still wrote my undergraduate dissertation on an electronic typewriter. My MA was written on the computer and during my PhD I published two articles on an “electronic journal”. But computers had not had a significant impact on education.
Fast forward to now, and most schools have Google accounts for their students, even at primary school. During lockdown lessons were delivered on Microsoft Teams. Online applications are not only used for teaching Computer Studies, learning Python, HTML and CSS. They’re also used for maths homework and times tables, as well as pointing to videos on BBC website and others.
But computers and mobile devices have just replaced text books and written homework. They have changed where learning happens, not how learning happens. But I think AI should create a monumental change in the steps through education. The question is how long it takes academics to realise it.
Last week I spoke at Engage conference in KAA Arena Ghent. I’ve been attending Engage since 2010 and only missed one of the annual conferences since then. It was my first conference, so it’s always held a dear place in my heart. Although it’s many years since I developed for Domino REST API, my work over recent months gave me experience to build innovative proof of concepts using Domino REST API and experience of various models and AI agents to integrate the API with various interfaces.
AI usage has evolved over the years as the power and limitations of LLMs and agents have progressed. The community has learned that they needed ways to provide additional abilities. Coding clients have adapted to provide those capabilities. Thankfully the IT ecosystem has changed dramatically since the birth of enterprise IT, so open source and standards have become the default mechanism instead of a last resort after community pushback. But we’re starting to see yet another approach, one I’ll call Token Engineering.
AGENTS.md has become the de facto standard for directing your agent. Claude Code had CLAUDE.md but with AGENTS.md becoming part of the Agentic AI Foundation last year, even Claude Code now recognises and uses AGENTS.md. But the benefits of AGENTS.md are now being discussed and challenged in academic circles. However, maybe what’s being discussed is just the first iteration. And just as the AI world went through Prompt Engineering and then Context Engineering, maybe the next phase (or a next phase) is “AGENTS.md Engineering”.
The ability of AI to generate software applications has taken gigantic leaps forward in just the last few months. A team of agents using Claude Opus 4.6 wrote a C compiler with minimal human interaction. In the press release for Opus 4.6 Anthropic build Claude with Claude. I personally use coding agents more and more, and I’ve seen significant rapid application development with AI.
It was only late last year that I heard the term "high-agency mindset", even though it's nearly a decade old. But after learning about it, I started to recognise signs to some extent in myself and some others I know. It became understandable why high-agency individuals surround themselves with other high-agency individuals, and why they can become frustrated with low-agency individuals. Where others see blockers and wait for (or expect) someone else to clear them, high-agency mindset individuals can see solutions or expect people to take responsibility for resolving them - or else they just go around them. It’s a mindset that aligns well with the concept of Directly Responsible Individuals or DRIs.
But it also became apparent that AI could have a big impact on the future of the two high-agency and lower-agency individuals.
About 20 years ago there was a major shift in document processing which had profound repercussions for the AI world, years before it started. With Office 2007, Microsoft made a major change in their document formats. Up until that point, Microsoft had preferred closed, proprietary, customer lock-in of content into silos of information. But during the 2000s the rise of Apache OpenOffice and its open document formats, along with governmental demands for interoperability, put pressure on Microsoft to introduce OOXML. Of course it had added benefits - competition alone rarely encourages change. The new format was up to 75% smaller, improved corruption recovery and improved security.
Recently there was an announcement that Tailwind was laying off 75% (3 of 4) of its engineering team because of revenue collapse. The initial hint came from a comment on a PR and was followed up with a podcast on X. This has caused a lot of discussion in the IT world. But I want to take a different approach. What does this mean for the media industry and the future of how we find out news?
Vibe coding is probably the term of the year. Since being coined by Andrej Karpathy in a tweet in February the term has gained widespread adoption. My job is research, so I'm not one to accept code without review. But I absolutely need to be aware of new technologies and approaches, and evaluate their usefulness. And all technologies improve over time. So AI-assisted coding has been a regular part of my work for 18 months. Over recent weeks I've used it more and more, for a wide variety of purposes. And research is not just about trying things, it's about extrapolating and anticipating future usage.