AI: What does it mean at Buzz?
There’s no doubt that AI is having an impact across the software development landscape. We caught up with Buzz Technical Director, Paul Axten, to find out what that really looks like in 2026.
Q: Can you sum up what AI means in the world of software development?
AI is a catch-all term that incorporates machine learning, generative models, large language chat bots and a whole raft of other things besides. Prior to about half way through 2024, in software development, when we talked about AI, it was usually from the perspective of incorporating AI features into whatever it was we were building. Text and image recognition, sentiment analysis, trend detection and the occasional automated chat.
Rapid change has occurred since then, and now when we talk about AI we’re more often than not talking about the tools we use to make the software we’re building. The same technology used to generate those crazy images of people with 12 fingers is being used to generate crazy code, too!
Q: How is Buzz using AI in its day-to-day project work?
Well, it turns out in the right hands, the code isn’t that crazy at all. In fact, with careful guidance and a strict set of standards, AI code can be downright beautiful.
You may have heard the term ‘Vibe Coding’. It’s a little like jazz music. You jam with the AI, feeding it prompts and riffing of the responses trying to create something harmonious until you have software that feels like what you want. This is an approach non-programmers often take. It’s fun and can lead to surprising outcomes, but consistency and reliability are a challenge.
Buzz really value consistency and reliability, so we take a different approach.
First of all, we have a frame of reference. In exactly the same way we establish best practice, code style, technologies, testing and acceptance criteria with our dev teams, we brief the AI on our expectations. We show it our code, let it learn our methods and ask it to follow those standards. That even includes brand guidelines and design.
Secondly, we build an AI team. Much like we wouldn’t expect a single developer to do everything, we assign tasks to different AI agents so that each can focus on what it’s good at. This is important because AI is very easily distracted and prone to drift. The limited scope is like putting blinkers on a horse to prevent distractions.
Finally, we test and validate. Sometimes automatically, sometimes manually; but always in a manner where we know what the expected results are and can be sure everything is behaving as planned.
With the AI team assembled and quality control in place, we can get working prototypes up and running more rapidly than we could previously have created visual mockups. We then share working ideas with clients, refine and iterate quickly. This lowers cost and, more importantly, lowers risk. The prototypes evolve into features of the final live system, and all of the code is written in the shape and style we’ve developed over the years, meaning any of us can dive in and tweak when needed.
Beyond that, the AI will also document the code, describe pull requests (chunks of code ready for quality control review) eloquently and help in writing user manuals.
We’re the same sized team, doing the same skilled work to the same or better quality levels, just more quickly and effectively.
What do you think some of the most impressive changes (brought about by AI) have been?
Did you ever play the game Theme Park as a kid? The one where you could put the hot dog stand close to the big rollercoaster to deliberately make people sick? Well, the lead programmer on that game was Demis Hassabis. He went on to found Google DeepMind, which, in 2024, won the Nobel prize for Chemistry with AlphaFold. AlphaFold AI predicts protein structures and has been used to revolutionise the discovery of new medicines.
I love this story. Someone who used early AI to entertain by making people sick in a game has created AI that makes people well in real life! Amongst all the negativity about AI taking jobs and plagiarising art, this story stands out as a massive beacon of hope. AI is being used right now to benefit the health of humanity globally.
More closely related to software development, in the right hands, the quality bar has also been raised - yet, at the same time, the barriers to entry have been lowered. We can now automate basic programming tasks in a way that massively speeds up the initial phases of a project. Clients see results more quickly, and that means lower costs. Because of the rigour we apply to all projects, we’ve traditionally had a reasonably high minimum cost, so being able to reduce this has had a direct impact.
What do you think we can expect to see in 2026?
Six months in 2025 moved AI generated software from not viable to viable. We’re seeing further improvements every few months now. If we crossed the viability line in 2025, I expect quality to improve to the point in 2026 where AI-generated software is simply accepted as the new standard and becomes normalised.
That said, experts are predicting we’ll find the ceiling of Large Language Models (LLMs) that drive this technology. The models are already massively resource hungry; their use is affecting the price of hardware and the need to keep the hardware cool is a massive drain on resources.
That said, efficiency gains are being made, and we can see that with the AI processors included in phones and laptops (so, smaller models can be run in the palm of your hand). I think the critical thing is that models don’t need to be much better than they are today in order to simply become business as usual. We may see the financial bubble around AI burst, but the tech is good enough today to already be making a massive difference.
In summary? AI software development will be normalised, developers will adapt their methods and the software produced will be more readily available, more cheaply, to more people.
How should businesses be preparing to implement AI themselves? Is it worth it?
This is an interesting question, and possibly a little misleading. Not all businesses will have a need to implement AI directly. But, all businesses will be effected by AI one way or another, and preparation will determine how positive that effect will be.
I’m sure the majority of businesses have already sent an email that had a little assistance from an AI writing tool. They will certainly have received one. And if any staff have searched for something in the web, the results they saw will have been summarised by AI. Recognising that individuals within the business will already be using AI in their own way is a good first step. Establishing some simple guidelines so that everyone in the organisation knows your minimum requirements is the next step. Provide guidance on fact-checking AI answers and proof reading AI text to help promote an awareness that AI is fallible and requires as much care and attention as any other tool.
If you’re considering AI for software development, or for a technology integration, treat it like you would a new hire or contractor. Establish what your expectations are, determine the outcomes you want and how you’ll measure them. Know what you’re willing to invest. Each time you get a new version to review (and that could be multiple times an hour with AI!) validate it against your expectations, and have a framework in place for keeping the good stuff and amending the bad. Ideally, have someone on your team that can gauge both code quality and complexity. Complexity is still the enemy of long term code maintainability, even if you’ve got AI helping out.
Most businesses will have a dark corner that hides a dreaded high effort task or issue that’s been procrastinated on. Many business owners will say they’d pay anything to shine a light into that corner and make the problems go away. Often, a fresh pair of eyes will help, and healthy discussion can lead to clarity. From there, options present themselves, solutions start to become apparent and AI assistance may be part of that. We’ve already produced proof of concept demos with the help of AI that provide a realistic view of the tools your business could be using in the near future. Those demos provide a concrete talking point, promote buy-in across the business and offer a vision of what positive change could look like.
Is it worth it? You’re in the driving seat, you get to decide. The great thing is, AI makes it quicker and easier to make that decision than ever before.
What do you think the biggest challenge in AI is?
It would be wrong not to mention the environmental impact of AI data centres. That’s a massive challenge that needs to be addressed — both in terms of increased efficiency and improved sources of green power, cooling and manufacturing.
As a B Corp, we know it’s important to weigh up the value of using AI with the adjoining impact, and look at our use on a long-term basis. For businesses with similar concerns, this is where a set of guidelines for general staff use can really help.
There’s also the fear that AI will take jobs, just like other machines have before it. For the first time though, the machines are taking a bite out of workers in ‘the knowledge economy’ who, just a few short years ago, thought there was zero competition from mechanisation. That’s a big shock to a lot of people who thought they were irreplaceable! The key here is to learn it, not shy away from it, in order to keep your skills relevant.
In reality, the pace of change will be the biggest challenge to most, though. Soon enough, there will be no catching up. Just like fast internet blindsided Blockbusters (but catapulted Netflix), AI driven events will occur that divide those on either side of them. We’re all in for a rollercoaster ride, so we’d better choose our lunch carefully.
