Paul Withers' Blog

Is “AGENTS.md Engineering” The Next Optimisation Approach?

Mon, 23 Feb 2026 00:00:00 +0000

Is “AGENTS.md Engineering” The Next Optimisation Approach?

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”.

Lessons in Effecting Change - How We Got A Domino Non-Production Server License

Sun, 22 Feb 2026 00:00:00 +0000

Lessons in Effecting Change - How We Got A Domino Non-Production Server License

More than ten years ago, Christian Guedemann and I worked with various IBMers to try to get a free Domino server entitlement for developers. It took time - not an unacceptable amount of time, in my personal opinion - but we got more than we hoped. The dev and test server license is no longer available, because licensing for customers changed. Some are critical of that decision. But none, as far as I’m aware, have demonstrated taken the high-agency mindset that Christian and I demonstrated and which is crucial to effect change.

Transforming Software with AI

Sat, 21 Feb 2026 00:00:00 +0000

Transforming Software with AI

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.

AI and The High-Agency Mindset

Thu, 19 Feb 2026 00:00:00 +0000

AI and The High-Agency Mindset

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.

AI and Marketing Content

Sun, 01 Feb 2026 00:00:00 +0000

AI and Marketing Content

The Microsoft Revolution

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.

AI, Tailwind, and The Future of Media

Mon, 12 Jan 2026 00:00:00 +0000

AI, Tailwind, and The Future of Media

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?

AI Coding - Thoughts About The Future of Development

Sun, 14 Dec 2025 00:00:00 +0000

AI Coding - Thoughts About The Future of Development

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.

Effective AI Usage: Managing Hallucination

Sun, 26 Oct 2025 00:00:00 +0000

Effective AI Usage: Managing Hallucination

One of the biggest challenges when working with AI is hallucination. I've encountered and fought against it. It's worth discussing what can and can't be done to solve the problem.

Effective AI Usage: Understanding Brains

Sun, 05 Oct 2025 00:00:00 +0000

Effective AI Usage: Understanding Brains

I've talked about the (current) moving parts of AI and AI-fu. But a fundamental aspect of AI-fu is being aware of how we think and how that's different to how LLMs "think". It's probably true that most people are not consciously aware of how they think or aware of how colleagues thinking works differently. So it's well worth raising that topic, because it's crucial to the quality of results.

Effective AI Usage: AI-fu

Mon, 15 Sep 2025 00:00:00 +0000

Effective AI Usage: AI-fu

In my last blog post I talked about the many aspects of modern AI and the importance of understanding them all. But even more important than this is what I am terming "AI-fu". So what is AI-fu?

XPages App to Web App: Part 20: Custom CSP Settings

Sun, 14 Sep 2025 00:00:00 +0000

XPages App to Web App: Part 20: Custom CSP Settings

A good web server will enforce Content Security Policy settings. If you are using something like Express as the web server, the endpoints will set that Content Security Policy. In the case of Single Page Applications hosted on Domino REST API, since release 1.15 by default a strict CSP is applied. But it is possible to change the CSP settings per application.

Effective AI Usage Part One - What is AI?

Mon, 08 Sep 2025 00:00:00 +0000

Effective AI Usage Part One - What is AI?

In my last blog post I talked about some lessons I've learned from using AI. I talked about a follow-up article talking about AI use at a higher level. Recent experience has reinforced my thinking on this. In this blog post we're going to focus on what AI is, the initial interaction, and training.

More AI Lessons

Sat, 16 Aug 2025 00:00:00 +0000

More AI Lessons

A little while ago I blogged on developing at speed. The obvious omission from all aspects was AI. But AI – like an IDE – is just a tool. Unless you understand what it can and can't provide, unless you use it intelligently, you will not reap the benefits. But unlike an IDE, AI doesn't come with a set of menus that hint at what it can and can't do. AI doesn't come with a marketplace of extensions that provide functionality shared by the community. And it's so new that we're all working it out. So what are my thoughts?

Supercharging Input to Domino REST API Agents

Tue, 05 Aug 2025 00:00:00 +0000

Supercharging Input to Domino REST API Agents

One of the things I learned when building HCL Volt MX LotusScript Toolkit was that calling a web agent with ?OpenAgent URL populates the NotesSession.DocumentContext with various fields containing useful information from the request. So when I was building agent processing functionality into the POC that became Domino REST API, I utilised the same approach to provide opportunities to pass contextual information across to an agent.

Developing at Speed

Mon, 30 Jun 2025 00:00:00 +0000

Developing at Speed

One of the main outputs of research development is the proof of concept. An early lesson I picked up when I joined HCL Labs was to deliver working code, not slides. And the key when building a proof of concept is speed. In some cases, it may end up proving why an approach won't work. In many scenarios, it may end up being put on a shelf indefinitely. Even if the concept proves appealing, the implementation choices may not be the preferred option for the final solution. So speed is of the essence: spending a couple of weeks building something that goes nowhere is acceptable; spending a couple of months is not. So the ability to get maximum results in the minimum time is key.

But how do you do that?

Adventures in Rust

Sun, 15 Jun 2025 00:00:00 +0000

Adventures in Rust

One of my core principles for IT research development can be summed up by a saying more than 2500 years old, attributed to Solon and appearing twice in Plutarch’s Life of Solon: “I grow old always learning many things”. In many ways, to stop learning is to stop living.

2025 05 18 engage 2025

Sun, 18 May 2025 00:00:00 +0000

This week, once again, I will be speaking at Engage in Den Haag. As usual, it promises to be a great conference. And as usual, I will be busy.

Shu-Ha-Ri

Thu, 15 May 2025 00:00:00 +0000

Shu-Ha-Ri

Some months ago Stephan Wissel pointed me to an article on Shu-Ha-Ri, which refers to Alistair Cockburn's Agile Software Development. It's well worth software developers being aware of the concept, because it is crucial to the developer's life journey.

Domino REST API, CORS and Regex

Thu, 01 May 2025 00:00:00 +0000

Domino REST API, CORS and Regex

Release 1.1.3.1 of Domino REST API introduces a breaking change in CORS handling. This makes configuration less straightforward, but as the documentation states, it increases the flexibility and probably makes things a lot easier for larger environments. And though regex is not something Domino developers work with regularly, there are tools close to home that can help.

Using AI

Sat, 26 Apr 2025 00:00:00 +0000

Using AI

As a researcher, I'm always looking to learn, to expand the toolbag I have, and find innovative ways to improve outcomes. Even when AI is not at the heart of the project I'm working on, I'm constantly looking for ways it can make my life easier and life easier for developers using what I create. And the fact that we targeted a standard IDE means the effort required to integrate AI is reduced. But as with any new tool, it's important to learn what it can do and what it can't. And work this week has shown me that this requires a diligent approach.