AI Accelerator: Nicholas Brien, Lead Software Engineer

In this installment, we talk to Nicholas Brien, Lead Software Engineer, who’s building AI into the products our clients use, while also leveraging it for a variety of tasks. Here’s where he thinks AI is genuinely changing his craft, and where he’s still skeptical.

Q: How did you first start using AI? Was there a tool or project that really got you excited?

A: I played around with DeepMind’s image recognition tutorials back in 2016 and found them really interesting, but then I kind of forgot about AI for a while. When ChatGPT was released, though, it really got my attention and made me realise how quickly the technology was advancing.

I think coding tools like Copilot and Claude Code have opened up endless possibilities for what I can create as a developer. I have a few pet projects that run on microcomputers which I’d always wanted to develop, and now I can simply prompt a coding agent and have something up and running in a fraction of the time it would have taken before.

Q: How are you using AI typically right now – work? Side projects? Both? What is AI helping you to do?

A: I use coding harnesses and AI-powered development tools extensively for project work and day-to-day tasks. We’ve also used AI to help automatically identify and fix security vulnerabilities in our code.

On the Revolution team at Confluence, we’ve developed an AI Knowledge Agent that uses large language models and retrieval-based search to help both our internal client services teams and our clients themselves make sense of what’s happening inside the platform. One of the more practical applications is around import session summaries — when a client runs a data import, the system can automatically analyse the session and surface a plain-English explanation of what happened, flagging any issues caused by bad data or misconfiguration. Previously, diagnosing those kinds of failures meant manually working through logs or escalating to a specialist. Now clients and support staff can get straight to the answer, which makes a real difference when you’re trying to onboard at scale or resolve something under time pressure. The broader goal is helping clients and client services teams become more self-sufficient — spending less time in back-and-forth support cycles and more time actually getting value from the system.

More recently, we’ve been working on an MCP server for Revolution — taking what started as a proof of concept and making it production-ready. The idea is that instead of logging in and navigating through the product to run a report, a user can simply ask a question in plain English from wherever they already work. We’ve been prototyping it in Excel — where most Revolution users live — and the early results are genuinely encouraging. You can ask for portfolio performance versus a benchmark, break it down by sector, generate charts, and pull together a one-page client report, all through a conversation. There’s still work to do to get it to a place where it’s robust and secure enough for client-facing use, but it’s one of the more exciting things I’ve worked on.

Outside of work, I also have a few pet projects that use AI to automate some of the more repetitive tasks around the house.

Q: How has AI changed your profession, both for you, and future entrants?

A: I still think AI cannot replicate the more complex parts of my day-to-day work, and I still enjoy being challenged by difficult problems.

I produce more output than ever before, which means there’s also more administration, more QA interactions, and far more meetings with product managers because we can deliver features much faster.

I feel like I can handle more tasks and take on trickier technical challenges that may previously have been too complex or time-consuming to analyse and fit into a typical sprint.

It’s also become easier to hand work over to other developers, knowing they can use AI to help navigate parts of the system that previously required a significant amount of oversight and knowledge transfer.

We’re also developing entirely new AI-driven tools and features, which is creating new opportunities and challenges for everyone involved. It’s an exciting shift because many of the products and processes we’re building simply didn’t exist a few years ago.

The Revolution MCP server is a good example of that. What would previously have taken back-and-forth between a client, an implementation team, and a product specialist to get the right data out of Revolution in the right format could, once this is production-ready, happen in seconds through a conversation. It also requires thinking carefully about data security, lineage, and how you surface the right context alongside the numbers. That’s genuinely new territory, and it keeps the work interesting.

If anything has decreased my enjoyment, it’s reviewing poor-quality AI-generated code. AI can accelerate development, but it can also generate a lot of technical debt if it isn’t used carefully.

I think it’s going to become harder for junior developers to enter the market when senior developers can increasingly distil their knowledge into coding and testing agents at a similar cost.

For new developers, I think success will come from focusing on deep domain knowledge first — whether that’s finance, healthcare, data science, robotics, or design — and treating coding as one of several valuable skills rather than the primary differentiator.

Q: Where do you see the current limitations of AI? Do you have any skepticism?

A: I’ve never really been skeptical of AI, but I’ve always understood its limitations. My main skepticism today is around how AI is portrayed in the news. A lot of the coverage focuses on AI disrupting humanity, whereas the reality is usually much more nuanced.

AI isn’t particularly creative. It still needs a human to provide direction, context, and judgement. We’re also still very early in the journey — many AI tools feel like we’re in the technology demonstration phase rather than the fully mature product phase.

A frustration many share is that AI can sometimes create the illusion of progress. It can generate large amounts of code, documentation, or analysis very quickly, but that doesn’t necessarily mean the output is correct, maintainable, or valuable. Human oversight is still essential.

AI can sometimes create the illusion of progress… Human oversight is still essential.

Nicholas Brien, Lead Software Engineer

I would add the long-term cost of AI is something that could come back to haunt organisations once it becomes fully integrated into every workflow. Many businesses are becoming dependent on tools that may become significantly more expensive over time.

With that said, I am personally and professionally very excited by how fast this technology is changing, and applying it in the solutions I build, and the way in which I build them.

Q: What is the AI development you are watching right now – and why?

A: Small, focused models, fine-tuning, and edge LLMs are particularly interesting to me. I think we’re moving towards a future where purpose-built, specialised models run directly in your browser, phone, or local devices, trained on your own data and tailored to your specific needs.

The potential applications in areas such as healthcare are especially exciting, where highly specialised models could provide assistance without requiring data to leave the device.That same principle — connecting a specialised model directly to trusted, structured data — is exactly what we’re working towards with the Revolution MCP server. The model doesn’t need to know everything; it just needs a secure, well-defined line to the right data. It’s the same thinking behind the Knowledge Agent too — rather than a general-purpose model trying to guess at context, you give it a curated, structured knowledge base and let it work from there. Seeing that approach come together in practice, and knowing it’s the direction the industry is heading, makes it a particularly interesting time to be building in this space.