DevHub North - September 2025
Welcome
Rosie from MHR welcomed everyone with people from MHR talking about tech with everyone at Dev Hub North. Carl is one of the engineering managers based in Nottingham and has been in engineering for twenty plus years and are up here as they have opened a tech hub here in Newcastle. Steve is an engineering manager in the Newcastle tech hub and has been a Java engineer for twenty years. MHR are a family-owned business that started forty-two years ago based on Nottingham, they have opened a Newcastle Tech Hub to take advantage of the tech scene in Newcastle with a lot of potential to build up talent. The Tech Hub is for learning, and they are looking at AI and building an AI assistant into their product for time management and scheduling which is mainly used when doing business out in the field for remote based rather than office-based staff. MHR has around thirteen people in the North East Tech Hub.
Transforming the World of Work - Our Journey - Adrian Towse
They have a thousand employees with over 250 developers and testers working on their product. Tech always has a high impact in the world of work on employees and organisations, what is changing in HR and companies including embracing change including AI. MHR has always embraced tech and there have been improvements in HR and payroll and their product as from generations of products. They aim to be sustainable and invest in the next generation of products, they don't milk products and let them fall away, they invest in real technology that solves a real problem.
They start with the problem not the technology, Covid changed everything with hybrid working and what has changed is the experience of work with employees driving forwards with learning and what skills they need and will need in the future. Workforce is multigenerational with different experiences and demands and organisations are looking at how they operate to support these changing organisations.
Companies can't just continue to grow head count you can't just add people in, it is more important to understand employees' capabilities and grow existing talent to meet skill needs. People want to work with AI and organisations are changing physically and organisations are moving to value and contribution led but employees need to know how they contribute to the business with a strategic plan on how the organisation looks.
The culture of the business is changing, and employees are driving themselves forwards and productive experience benefits both employee and organisation for self-management on how employees are performing and feedback on value they are providing to the business. Only 30% know how they contribute value to a business so there are huge benefits to business to provide connected culture and organisations are stiving to do more but can't just grow headcount so want to use HR and payroll solutions to help with this.
Opportunities are digital transformation to improve work, using technology to transform service offering to improve things and using AI to leverage knowledge of employees or automate tasks to save time and free people up to do more interesting work that is more valuable to the business. Things are changing so it is critical to be agile and resilient and have the skills to deploy high value work and use data to inform business decisions, reports just don't cut it, need to use data to drive the business. Use AI to supercharge work with AI first with their product.
People First removes the unnecessary complexity of paying people and managing people and puts decision making into the hands of employees with an Azure SaaS model to leverage technology and is solving the problems people are looking for the technology to solve. Some of the things they are talking about is empathetic innovation and nothing scares more is AI so need to look at how to build AI into the product and gain trust so the route has been to create understanding and deeply understand the problem and don't want to replace jobs but reduce mundane tasks to allow people to do the best value oriented work for the best version on an employee.
It is about helping a business transform, AI as a change management project that enhances not complicates work, empowering not replacing jobs with a safe responsible AI adoption. Need to do a good job to take people on the journey with you. AI is in an early phase but is accelerating with acceleration of routine tasks for HR for managers such as check in summarisation for goals and feelings of employees and actions or CV parsing to give time back. How to amplify people to enable people with less experience from skills with those with more knowledge such as sentiment analysis or identify pain points in and organisation to identify the problem areas to focus on and take tangible actions or use natural language to augment to remove technical barriers to searching. Use AI agents for multiple complex tasks using a level of reasoning to get to an outcome such as negotiating a contract with a new starter. See structured data insights, natural language summarisation and AI driven actions.
Accelerate with easier check ins to generate talking points which would have taken some time but can quickly know the last check in with someone. Better recruitment with advert creation, CV parsing and interview question generation to automate tasks may be doing to take away time consuming and repetitive tasks may allow you to see more candidates and even include some wild cards to help people accelerate everyday tasks. Remove technical barriers with natural language queries so may need to know how to create conditions but with natural language can ask how many people have been promoted or have been sick and it can create the conditions.
They have a layer over Databricks for all their storage which they can query but the data is complex inherent to HR is difficult to query and hard to query properly but can now create a natural language query to generate the query and as you write you will get suggestions on what you can query to build a cubeJS query and will get a better idea of how the tool works so don't just do the query but improve the ability of others on how to do it including how to tune values and modify query to then execute and see the data. Empathetic innovation isn't just about making it do something you want to empower a person to do a thing, you don't want to take that power away from them it is a tool.
People First HR assistant can answer questions such as a career development questions but if you know your summaries of check ins and career development roadmap you can start to look at how to fill those gaps. HR assistant is smart but not by itself, treat the AI like an actor, AI agents are great tools but by themselves they are dumb, you can give it a role to play, let it know its boundaries, let it know it has a box of tools it can use, provide missing context where needed. You will have your system prompts with the overarching personality, the user message for what you want and the guardrails for your agent such as don't talk about violent things or unlawful things but tuning the prompt is a big piece of work but if you get it wrong it can allow all sorts of things so there is a lot of work to get this right so it doesn't do anything wrong.
MCP is another AI server with a box of tools that an AI agent can call to do things it can't do. AI is good at natural language and MCP allows an AI to call functions in code to perform an action to do things, it is an API to return a list of tools with a schema on how to call them from a list of numbers to a name of a person and then can use this to return raw data as you need it rather than English prose to then perform an action on you could call APIs or run other tools. They are looking at this across their platform not just in their tool. They are also looking at anomaly detection so if can spot things early then you can take action and can bring things together with a management lens which can extend things to summarise not just for an individual but across a whole department to see what are the key things for different divisions within the business to highlight recruitment candidates, skills gaps or all sorts of things.
The augmented coworker is that every user can be augmented with knowledge acceleration and skill amplification to create a high value employee to bring more value to the business. Companies that are clued up on skills they have that they aren't using or skills they might need in the future or using this to do planning or adjacent skills that may be relevant or could be repurposed. There are so many areas where having skills insights would be important and their aspiration is to have a strategic partner for the organisation, not just paying people and HR but really contributing to the business. When you understand skills employees have vs what they need then fill that gap and then going into area of strategic workforce planning. They want to put agentic AI on top of things, People First product or Teams to allow AI anywhere in the context of their people and 52% of businesses now use AI in some form and it is predicted to grow and is an area they want to focus on with agentic AI to create assistive workers to augment people.
AI everywhere, when looking at perspective of a processing system so if have many AI assistants may be prompted have daily tasks but have a job advert already approved and may be able to arrange to see these candidates or put out an advert to get more than an AI assistant could do and others could put together interview questions, transcribe interviews or help make decisions. Need to keep the right balance between AI and humans to accelerate the mundane things and keep key human interactions. Using machine learning could tailor learning paths to individual employees and tailor that to business needs and then spend the time that is free to do more interesting or valuable things and deliver self-serving things that deliver a lot of value. These are exciting times and being left behind with AI is a problem and not moving forwards is always a problem and need to evolve our knowledge, train our teams and take them with you.
How to you protect against bias when using these automated systems? You have to be careful to build the right prompts and keep the openness in the system, it is a balance and an evolution of a process and keep things that help and ditch the ones that don't. Like anything it is iteration and finding the process and make sure there is a human in the middle to make that final call and can't be relying on the AI.
Designing for everyone : Why accessibility can't be an afterthought - Sean Lynch & Ed Wheldon
Making technology usable for everyone and is suitable for people with disabilities and is usable by everyone. There is 1.6 billion people with a disability and at some point, 50% will have a disability with a temporary or permanent disability. There is auditory, cognitive, physical, speech and hearing disabilities. They follow the WGAC guidelines from W3C which are recognised for accessibility the goal to remove barriers for people with disabilities so people can use websites or use software or make them more accessible and usable. People can use tools to hear a website or use other assistive technology to make something more operable or make things understandable, so they are intuitive and easy to use and robust for a range of assistive technologies, browsers and more.
Disability is mismatched human interactions where this could be looking at a menu in a restaurant or maybe in a noisy environment or even driving. When we think about disabilities we sometimes think of it as a characteristic of someone but this is the medical model but what is more true is the social model where a person is disabled by design. If you went to a library and found a book just in braille you would be at a disadvantage or at a play that was signed, so by design can be disabling, when something is flexible based on user needs it becomes more accessible, when working with software we can make someone's life a little bit better. Differences in user needs where people with different types of low vision need opposite things so may need smaller text to focus on content or larger text to be able to see this where accessibility is about supporting flexibility for different user needs.
The cost of not caring about accessibility - the US, UK and Europe require equal access to digital content ensuring accessibility for all users which is monitored for WCAG 2.2, and US companies have been sued on website accessibility such as Netflix, Amazon, Burger King and more. Also, others like ADP, SAP, Glassdoor and DocuSign have had lawsuits and fines so need to show and embed accessibility into products. Risks of non-compliance includes potential legal action from customers with disabilities, fines, damage to reputation, lost sales opportunities from customers who cannot access the website along with negative publicity and brand image.
They are proactive to issues rather than reactive from the design process and detect 40% of issues before entering development phase and have automated testing to capture issues such as contrast or missing alt tags and semi-automated tasks to detect any issues or can have manual audits with screen readers and professional services that can pick up on things don't think of or edge cases or nuances that couldn't be detected, there is no single approach need a multi-layered approach. Leaving users behind can be fails with just a box ticking exercises and aren't real solutions to accessibility.
When accessibility is an afterthought, it is no good, need to shift it before design which is better but what it should be is considered at every stage of analysis, design, development, testing, release and beyond as the internet and guidelines constantly change such as from WCAG 2.0 to 2.1 was resize up to 200% to 400%. How to you make things accessible by following the WCAG levels with A, AA and AAA which is for government websites with most restrictions but they aim at AA but need to make it a rounded experience and have champions in each teams for accessibility from dev, test and design and be clear to everyone what their ownership of areas of accessibility will be.
You have to bring people together from plan, design, build, test and release and need to design with accessibility in mind and will spec up a design and make things really clear and use tools in Figma to recommend accessibility elements and plugins can do a lot of the work for you or use Google Lighthouse to pick up basic things like missing title text on buttons, ordering page in a logical way and others can be detected and can get a report to see how are getting on from Google Lighthouse with a combination of automated and manual tests. They also work with a company called Level Access who are a consultancy who will come in and check your work with an independent report that can be fed back into their backlog.
MHR also do user testing to test software with charities and see how small changes can make a big difference so recommend working with charities to do this. Think about the testing you have and have someone check your work. Are we at a tipping point in digital accessibility but it hasn't been fully baked into processes, digital experiences are becoming more complex and many struggle with accessibility at scale. For example, across a million pages there were an average of 51 errors per page such as contrast, missing labels etc which automated tests would pick up are still out there. Potentially things are getting more complex where site elements have increased 61% in the last six years, progress is fragile so keep the focus and bring accessibility as forward as possible, make it continuous, blend methods and make it a team sport across design, test and QA.
The future of development in an AI world - Ricardo Elías, Ludmila Gavriliuc & Chris Key
Building AI into the development experience but AI is a big topic so will look at the issues that actually matter to developers. Is there fear or fascination by AI? Does anyone fear AI or are fascinated by AI, do they fear it will replace their jobs, but you can be fascinated and fear AI. How AI is changing the developer role, there were the days when you had to write code line by line and now this has changed, they are doing design and API management and contracts between microservices and talk to AI agents so as a principal isn't doing much code but a senior is reviewing code and AI is becoming part of building day-to-day software and prompt engineering is one of the future jobs according to Ludmila.
Engineers are system conductors of different AI tools, but all developers are different, there are differences in the way they operator. For some people writing code is artistry and they don't want to hand this off to an AI. Developers' evolution in the AI era, programming languages evolve so you need to learn about new features otherwise you are outdated, such as Java has things to make cleaner code or have records, but Java has a lot of boilerplate code so are improving that along with safety and reliability of code. Frameworks such as Spring help Java to be one of the main choices for back-end development and for cross cutting concerns.
Looking at software development lifecycle of stories, TD and clean architecture might have no ambiguity but often developers will put work in but will make assumptions on how to implement this and it might not be how it was imagined so could put stories through AI to challenge assumptions to see if this is what they meant or can change the words. When building systems to interact between them in any language there are frameworks to help such as test-driven development to ensure that code written does what is needed, can test the code based on expected outcomes but could take the Git diff and put this through an LLM to generate tests. Clean code and clean architecture helps and will interact with AI coding agents better, they have millions of lines of code in their services, so how do you build context for your systems, can you describe your services and how do you give context for an AI to generate code that makes sense without overloading it, how to tune this balance can be difficult, often the agent makes the choice on whether it needs to include something.
MHR use AI to help with testing and test generation by writing a spec to generate tests or write specs to generate code which can then run the tests and refactor. When you have a user story that can fit into a unit test this can be fed into an agent, and the test can be written which fails as a red test and then can ask to write the code to fulfil the unit test to become a green test. Developers need to have a good understand that code is reliable and does what is expected of it. Tooling and productivity include IntelliJ, VS Code and tools to help write code faster and for testing code such as Junit to build things quicker and can use GitHub Copilot or ChatGPT. AI is also helping to do AI code reviews and summarise what code changes are or use static code analysis but instead of raising your own PR you can get the AI to fix it for you, and everything is improving with that.
MHR have used AI to create a bash script to add retry logic to have configurable and back off interval to solve the problem they had or another example which was taking a page built in Angular and told an expert to rebuild it in React using AI, took a screenshot of the page and took the Swagger definitions and took the prompt including coding standards, functionality and what it is for and it did do it and the code was really well structured. People may look at this and see that AI can do everything in React for them but need to look at how to keep control and how things work properly. They had an issue with noisy neighbours with too many messages being sent to a queue and worked with ChatGPT to produce a solution in C# to allocate message to different queues, but if go to management and say have a solution but use AI to generate animation to explain an intended solution to visualise something complex. They had a component in Angular and have used GitHub Copilot and background agent which they have taught to replace a component so they feed in lots of issues to resolve them with these components and will be far quicker and something that no-one wants to deal with and solve a big headache that allows them to move onto other things.
Career reflections, what does it mean with interaction with AI, Ricardo things AI won't replace them, and Ludmila mentioned it will work alongside developers. They are always trying to keep learning, and it is a new tool and need to embrace it and the future belongs to developers who embrace change. AI is a tool, and AI does is helpful to developers but from a development point of view it is not mature and is still early stages, but it will improve. Big changes take time, when cars appeared horses and cars were mostly side by side but now there are many cars with just a few horses on the road. Developers have to have a willingness to adapt and AI can help with reliability and scalability with a human centred approach to work together with AI, it doesn't matter what language the code is written from it can be a user story, it can be reviewed for security etc, it doesn't matter if it is Java, .NET or even assembly, using the right tools for the right job to solve a problem.
Q&A
Question to Sean and Ed about developers starting out on what to keep in mind for developers and how do you improve culture of company to think about accessibility first. Can have a look at user needs first before looking at the guidelines, just understand what different user needs are and what requirements people have and can learn a lot online with a tonne of training videos and there are videos on YouTube about accessibility. Changing the culture of the business was something they had to do at MHR and a lot of their products were old and have to go to the top to make sure that it is important, it can bubble up from the bottom but have to raise it to a high level to build resources and time to build accessibility into your product.
Accessibility from an AI point of view what is the challenges from these? Software will get a lot better with AI, scanning software will be better to find issues, they are embedding chat with a whole new interface on how screen readers interact with this and how do you handle micro-UI such as graphs to a screen reader.
For converting Angular to React do they use GitHub Copilot project wide context? They did their exercise in Cursor but have a massive variety of IDEs so have project files to tell the AI what to do and have variations for IntelliJ or Visual Studio Code and make it so the developers can use whatever IDE they want. They have been putting their code into GitHub to get more context and for AI assistants with MCP are using GitHub Copilot for background agent to raise issues, describe changes and then assign this to a GitHub Action to perform in the changes so can create issues ahead of time along with tests and can assign these to the agent while working on other things and the code changes will be assigned back for review and if it is not exactly what you need can comment and the agent can make those changes. You treat Copilot like a junior developer that you can assign a task to, but every single time it needs to regain that context and will make interesting assumptions depending on the tasks you give it, you can encourage it to not make the same mistakes again in the future.
Development of AI, how do you check the AI prompt, do you have a developer checking every line of code? They still have pull requests, you still need human interaction to review changes, the AI will not commit the changes directly to production, the idea is still to have a human approve a code change performed by an AI. When using Copilot's background agent, it co-authors you, so you get commit messages and at the end of the day you get ownership over these code changes.
Someone heard that software developers are just prompt engineers and are just solving problems differently not to put anything into production without having a broad idea of what it is doing so need to have human interaction on this and when context switching AI can help especially when switching between different languages. They mentioned no one is using AI fully for development as need to have reliable code, AI is learning from us and not all code going into AI was written by a senior or knowledgeable person, AI is very good at writing tests but still need to have that interaction with a senior developer not only to review AI code but code written by junior developers. Simple scripts or small amounts of code can be reviewed from AI, AI will not innovate, who knows what it has been trained on, to some extent you can reason about something, but you don't want to be releasing or merging something you don't understand.
Question about MHR, what was attraction about moving to Newcastle? MHR has different teams across the world including Lisbon and in Poland, they have lots of feature teams and around this have architecture and cloud platform so rather than expand teams in Poland they looked at expanding in the UK, they looked at different cities and Newcastle came out on top. They embarked on this and the opened an office recently and have a couple of teams with a great team of guys but may look at expanding elsewhere but don't want to fragment and want to build on the success they have had in Newcastle, there are plenty of big tech firms with plenty of people here.