As someone heavily using AI, I understand that this isn’t about depriving competitive advantage, this is about responsible development.

Claude Mythos isn’t a better chatbot, it is something else entirely. It has demonstrated the ability to identify and chain zero-day vulnerabilities in major operating systems. Vulnerabilities that survived in the wild for over 25 years without known exploit.

When a tool reaches that level of capability, you can’t just throw it into the wind. An unrestricted API tomorrow doesn’t just empower small businesses, it hands every malicious actor a pandora’s box of tools capable of causing real harm.

Partnering with organisations in high-stakes sectors gives the maintainers of critical systems a window to respond. Letting experts in systems millions rely on explore the risks limits exposure where security isn’t a feature, it’s a prerequisite.

The argument for democratising AI is valid in principle. But it consistently ignores the gap between offensive capability and defensive infrastructure. Right now the sword is sharper than the shield. People may paraphrase Terry Pratchett’s Death “THAT WILL BE AN IMPORTANT LESSON” but responsible development means building the shield first, in a controlled environment, before distributing the sword. AI capabilities need balance, because often there is no putting the genie back in the bottle.

The technical advantage for SMEs and independent engineers isn’t in having the most dangerous tool first, it remains where it always has, in agility and innovation.

I’ve been revisiting Hannah Fry’s Hello World: Human in the Age of the Machine. Even though it’s nearly a decade old, its exploration of “black box” systems in critical sectors like healthcare and law is more relevant than ever.

An algorithm isn’t a magical arbiter of truth; it is the downstream consumer of a data pipeline. If the historical data feeding that pipeline contains decades of systemic bias, the AI doesn’t solve the problem, it scales it. Hidden assumptions become system rules, automated and propagated at a rate no human ever could. How to sanitise bias remains an open question, there is no one mathmatical model for fiarness.

The Black Box problem is particularly dangerous in the critical industries, like Legal and Medical. In these fields, transparency isn’t just a feature; it’s a requirement of law. If a Senior Data Architect cannot explain why a system reached a specific conclusion, that system is a liability, not an asset.

The real challenge for the next two years isn’t just building faster models. It’s about building accountable, transparent, architectures. We need to find the narrow balance between human intuition and algorithmic efficiency.

As Fry argues, the goal shouldn’t be to replace the human, but to use the machine to highlight our own blind spots. Before we talk about a “revolution,” we need to our focus should be on building systems that support decision-making, not replace it.

Admittedly it’s not perfect. Some of this is down to the level of supervision it needs, with a tendency to repeat earlier mistakes and it is more difficult to exclude it from sensitive files than I would like, global settings work and the .claudeignore package looks promising, but feels like something that should be part of enterprise software by default.

One of my most effective approaches has been directing Claude to generate a persistent memory of my professional style and technical approach. By combining context from across multiple projects the model provides more consistent outputs that match my established patterns. This reduces the cognitive load during review and prevents overlong explanation in responses. Consider asking it to interview you about the stack.

In addition to generation, I’ve integrated Claude into my workflow as a peer reviewer, reducing the burden on the wider team by producing more consistent PRs. When comparing it to other ecosystems, I find the choice is task-dependent. Gemini remains my preferred AI system for broad, knowledge-driven synthesis, but Claude excels within my specific data engineering domain. Its ability to ingest skills and context at execution is a significant benefit. Ultimately, the shift toward agentic tools like Claude Code represents a fundamental change in the data engineering landscape. By offloading the heavy lifting to the agent while maintaining a supervisory role I have been able to focus on more strategic and design decisions. As I continue to refine my local context and push the boundaries of what these agents can handle, it isn’t just about writing code faster, it’s increasing the complexity of the problems I can solve.

After completing the Yorkshire Three Peaks last year, I found myself in a bit of a slump. Without a significant long-term goal, both my physical and mental health had started to decline. When I first heard about the Wolds 20, I was not sure about signing up. There wasn’t long until the event, I was out of shape and hadn’t been training. However, with some encouragement from friends, I decided to give it a try. I wanted to prove to myself that I was still capable of that distance, and I missed the joy of being outdoors with like-minded people.

The day offered ideal walking weather, dry and not too sunny. My kit was still prepared for the Yorkshire Three Peaks so I arrived with a pack that was definitely on the heavy side, noticing a real diversity of kit at the start line. As a last-minute solo entrant, I spent much of the morning among other walkers completing the overlapping 13-mile route. But as the 20-mile path broke away, the crowd thinned. For a while, the only sign I was heading the right way was the occasional runner overtaking me (having started an hour later!).

The middle stretch took me through long uphill sections and into the beautiful dry valleys. By the time I reached the Thixendale checkpoint for lunch, I was over halfway and found a group moving at a similar pace. By the final manned checkpoint, the fatigue was setting in. It was incredibly tempting to call it quits, but with only five miles to go—and a significant slab of Victoria sponge under my belt I dug in and carried on.

After the challenging wolds near Huggate seeing Millington finally hove into view in the valley below was the best sight of the day. My feet were suffering and the clock was ticking. I dumped the last of my water to shed some weight and pushed through the final mile.

Ending the day with a cup of tea and a bowl of Thai chicken stew at the village hall was the perfect conclusion. It wasn’t my fastest walk, but I would do it again, it was a real boost to know that I can do what I set out to and to meet so many people enjoying the outdoors.

I managed to keep up with the front group for most of the day, finishing under the target time of 12 hours. It was a brutal test of physical fitness and willpower, but a fantastic experience.

It was an early start, leaving Leeds before 6am and starting for Pen-y-ghent from Horton in Ribblesdale at 7. A fast scramble to the top saw us at the summit by 8:20. Next the route takes you across to Whernside, over twelve miles of hiking and the iconic Ribblehead Viaduct along the way.

We stopped near the viaduct to wolf down a few sandwiches before heading off on what I found to be the toughest part of the challenge, the ascent of Whernside.

Whernside is the highest of the three peaks at 736 metres (2,415 ft). The climb up Whernside is not steep, but it is long and was the closest I came to giving up completely. About halfway up I felt exhausted, everything was aching and my thinking was muddy and slow. I had not done anything like this in the months of training, and I really hit the wall. Either the continuous climb, or not enough food and water or both took a toll on me and I was done about mid-way to the summit, but thanks to the encouragement of my colleagues and eating a few of the snacks I was carrying, I managed to make it to the top by 13:13.

A change of socks and another sandwich and I was ready for the next leg to Ingleborough. The decent from Whernside to Chapel-le-Dale was steep and challenging. A colleague who had managed to keep the pace despite only joining a few days before the challenge dropped out when we met the support team at Chapel-le-Dale. It was no small achievement to get as far as he did and I was soon to get a taste of the same blisters and exhaustion.

Chapel-le-Dale to Ingleborough is boggy, you follow tracks that are slowly leading you towards a steep accent. As someone who is not a fan of heights, it was intimidating seeing the near vertical route and false summit we were about to attempt. I finally fell behind the front runners as I had to stop and gather myself before attempting the climb. Like most tasks, once started it was not as bad as I first thought. The false summit was something I expected but still took a mental effort not to stop at that point. There was a pinch point that felt dangerous and then the final trig point was in view. With a surface like something from another world, the broken top of Ingleborough was not easy to walk on, but I made it. Some other friendly walkers took a picture for me and that was peak three at 16:34.

The path down to Horton in Ribblesdale should have been an easy few miles after the day, but broken ground, exhaustion and walking alone at this point dragged the leg out into a tunnel of putting one boot in front of the other. I finally arrived at Horton in Ribblesdale at 18:34 and finished the day with a pint and a meal at the Golden Lion, catching up with the front runners and waiting for the rest of the groups who all made it in over the next hour and a half.

As a challenge, it was something I had never thought I would be capable of doing. It was a real test of will and endurance. As someone who expected to be near the back, I was really pleased not only to make it in under the cutoff, but also to be one of the faster people on the day. I am not in a hurry to repeat it, but like many of the people I met along the way on their 2nd Yorkshire Three Peaks, 3rd or more I would be up for doing it again sometime.

One of the biggest announcements from Google I/O 2023 was the release of PaLM2, a new AI language model that is a step forwards for AI and a rival to GTP4 in the LLM space. While technically very interesting it was overshadowed by the excitement of the new search experience Google is experimenting with in a closed Beta available in the US only. Their new AI driven search allows users to pose more complex questions than previously possible and cites sources in its response.

It was also a big year for hardware, with the release of the Google Pixel 7a. It already has mixed reviews, but is launched with a good price point and a free pair of pixel buds for a limited time, which makes it quite attractive. Google is very excited about their first folding phone, the Google Fold, but I prefer their new hybrid tablet/home hub which seems to solve the issue of having a monotask home hub which has limited use for individuals and the tendency of tablets to languish until needed by combining their new tablet with a beefed up dock which charges and lets the tablet also act as a hub with some nifty account switching UI features letting users personalise their tablet experience.

Beyond this, there are a huge number of releases, many of them incorporating AI to accelerate the creative process. I am still digging through the vast amount of content, but it is safe to say that it is going to be an exciting few months ahead.

After the kickoff, there were a number of breakout sessions where attendees could learn more about specific AI technologies, including open source examples provided by the University of Bradford. I attended an ideation session with two of their Doctorates who specialised in AI.

In the afternoon, we had the opportunity to work on our own AI projects. I worked on a project with direct application to the data pipelines that I create at work. I had a lot of fun working through ideas on this project, and I learned a bit more machine learning.

The event ended with a session where attendees talked through their ideas. I was really impressed by the creativity and ingenuity of the use cases suggested by the teams.

The week after the hackathon, I had the opportunity to present my project at The AI Show, a follow up event run by AI-tech UK. I was really nervous about my presentation, but it ended up going really well. I was able to share my project with a panel of experts, and I received a lot of positive feedback. I didn’t win the ongoing mentoring, but the two projects that did were both projects for social good and very deserving.

The AI Show was a good chance to meet new people including developers from the Leeds area and beyond and I came away with some interesting conversations and great new connections.

I’m excited to continue learning about AI, and I’m looking forward to seeing what the future holds for this amazing technology.

I am pretty sure I don’t. Nore am I a unicorn or a jedi, I am more of a bolder with plenty of inertia and a tendency to start crushing things when I start moving in the right direction, but that is a whole other article.

Productivity Ninja by Graham Allcott is a series of actions and processes to help you improve your productivity. It’s not fantastic going when you keep putting the book down as you don’t have time to do whatever action is advocated ‘right now’ which I find a little frustrating. It is also very much advocating a very narrow system of actions, rather than more general approaches that can be adapted as I have seen in different books. Admittedly there are a lot of caveats about adapting the exercises and I do like systems, so I have persevered.

Throughout the book it has been interesting to hear familiar ideas reiterated: the Prato principal is mentioned repeatedly, but two other ideas that I have started to hear mentioned are also covered, Parkingson’s Law; work expands to fill the time available and Hofstader’s Law; work takes twice as long as you originally anticipated, even when taking into account Hofstader’s Law. Both of these ideas are interesting takes on your capacity to do work, but are used as illustrations in the wider theme of the Productivity Ninja, separating planning from doing.

Eat that Frog is very much about planning, as are the 10 habits of highly successful people, so overall it is nothing new here as a meta idea. Do the systems it suggests work for me? I am not sure they do. I do like the idea presented of getting all of my inputs down to nothing, no email, empty in-tray, no more nagging thoughts about getting the guttering cleaned out. The challenge is that it requires what the author describes as a ruthless Ninja mindset to process it all; splitting out actions and information, capturing everything in the master to-do list and filing or discarding all the scraps of information you are holding onto, just in case.

It also requires you to be ruthless about staying on top of these, carving out time to manage the constant flow of information, which is easier said than done.

On the whole Productivity Ninja brought nothing outstanding, with a fairly prescriptive view of what is right in productivity and a series of quick Ninja hacks that may or may not work for you. For me it is one to pass on.

One of the key things that I had underestimated was the sheer amount of enthusiasm I would come back with. Spending time with many different people applying python to different problems and disciplines really shook up my viewpoint of what I am doing, making what I have been working on day-to-day feel a little less mundane. It seems to be an almost universal truth that no matter how good what you are doing is, without some context it is easy to lose sight of what makes it special.

Another is that while many of the talks did not directly impact what I did, they all had learnings that I could take away from them, be it deep insights into the inner workings of the language, or high level ideas such as developing personal knowledge graphs.

Inclusivity was a big deal for the community, with the organisers making as many accommodations as they could including subtitling talks in the main assembly hall and providing a free crèche. They really went out of their way to make everyone welcome and that was reflected in a warm and welcoming community.

This year was a short format event, running for only 3 days due to the ongoing uncertainty of COVID19 when planning. 2023 will be back to a 5 day event and something to watch out for.

Doing a little, every day, leads to progressive change. It doesn’t sound like a new idea, but it is a great one to drive home and to consider more pro-active thought about our small everyday actions. One of the key learnings that I have take from this is about setting myself up to be more successful. By making things I want to do less often harder like putting the crisps on the top shelf at the back of the cupboard and fruit on the side next to where I make coffee, I make it simpler for my brain to make good choices when all I am really thinking is that I want a snack.

Combining this with things like planning the next day from Eat That Frog should in theory mean that every day is spent in a way that Sun Tzu described as fighting downhill, with the inertia working in my favour or as James Clear mentions in his book and the developer Oswald Nuckols is quoted as saying; being “proactively lazy”.

One of things from this that I have not been able to capture is a list of my habits. I am not used to thinking about how I act day to day. I guess that is partly the point of trying to describe your daily habits. Perhaps one of my habits is the lack of a habitual routine.

Reality is a little messy though, so while every day might not be perfect, this book gives me another set of strategies I can use to help me do more and achieve more of my goals.