[Event] JManc 2026
Yesterday I was so delighted to attend JManc 2026, hosted at AutoTrader's fantastic new office at No. 3 Circle Square.
I started the day with breakfast on the rooftop alongside Lorenzo and Bruce. We had great conversations, enjoyed the beautiful views over Manchester, and it was the perfect way to begin an inspiring day.
Keynote
The event opened with an inspiring keynote by Clare Sudbery, "The Power of Craft: Staying in the Driver's Seat of AI-Assisted Development" She focused on how software engineers should work with AI while continuing to apply professional judgement and craftsmanship.
One statement particularly resonated with me: AI has a planetary brain, but it lacks real-world wisdom.
LLMs have effectively absorbed an enormous amount of human knowledge. They can perform many tasks remarkably well, especially when given the right prompts. However, real wisdom comes from experience and context, something today's AI still cannot truly possess or connect together. It made me think about an important question: How can we tame this titan? It's a question that is worth each of us seriously considering.
Session 1 - How Does AI Impact Software Engineers and Our Future?
Nobody questioned that AI has become remarkably capable. Given clear requirements and well-crafted prompts, AI can produce code with impressive speed and quality.
However, software engineering has never been just about writing code. Requirements analysis, system architecture, integration, production support, communication with stakeholders, operational concerns, and long-term maintenance all remain essential.
If AI reduces the time spent writing code, perhaps developers can spend more time focusing on the bigger picture:
- designing better architectures
- understanding enterprise environments
- improving system reliability
- collaborating with business stakeholders
- solving complex production issues
One interesting discussion centred around agile estimation. If AI enables developers to implement features much faster, should story points decrease accordingly? Many of us argued that coding has rarely been the true bottleneck in software delivery. The real bottlenecks often lie in:
- requirement refinement
- dependency management
- integration testing
- deployment
- production issue investigation
AI accelerates implementation, but it doesn't eliminate these challenges.
Session 2 - Is Spring Still Relevant?
With newer frameworks such as Quarkus and Micronaut gaining popularity, it's a fair question.
Certainly, Quarkus and Micronaut offer advantages, particularly around startup time and memory footprint. In reality, several factors continue to make Spring Boot the dominant choice:
- enormous community support
- a mature ecosystem
- abundant learning resources
- easier recruitment of experienced developers
- years of production experience
Trust also plays a huge role. Many organisations already rely heavily on the Spring ecosystem, including Spring Web, Spring Data, Spring Security, and many other projects.
In particular, Spring Data often becomes deeply embedded within business applications. Once an organisation builds its domain logic around Spring Data, especially across multiple data sources such as relational databases, NoSQL databases, and search technologies like Apache Solr, moving away from the Spring ecosystem becomes extremely difficult.
One exception discussed was ultra-low-latency systems, such as front-office trading platforms, where every microsecond matters. In those specialist environments, lighter frameworks or different technologies may be more appropriate.
Apparently, for the overwhelming majority of enterprise applications, Spring remains highly relevant.
Session 3 - Running Local LLMs for AI Development
Seb shared his experience of using a local Qwen model as an AI coding assistant.
His workflow was particularly interesting. He first uses a proprietary model to generate a development plan. Once the overall plan is ready, he hands the implementation work over to a locally hosted Qwen model, which can work overnight.
This dramatically reduces token costs while keeping source code private. It was a practical example of combining the strengths of proprietary and local LLMs.
Session 4 - Future Java Language Features and Building AI Applications
Desired Future Java Features
Several language improvements generated enthusiastic discussion:
- non-nullable types to reduce NullPointerException
- named parameters to simplify object construction and reduce reliance on Builder patterns
- improved support for lazy constants
- better type projections
- richer ways to define operations directly on types
- easier copying or cloning of records
- greater flexibility to mix different programming paradigms within the same source file
Building AI Applications in Java
Several Java AI frameworks were discussed, including:
On the other hand, AI applications also introduce additional concerns, including:
- hallucinations
- changing model behaviour over time
- prompt regressions
- model degradation
Reflections
AI is now at the centre of conversations about the future of software engineering, and it was interesting to see how it naturally became part of almost every discussion throughout the day.
It was also wonderful to see more volunteers helping to organise the Manchester Java Community. Growing communities rely on passionate people who are willing to contribute, and it's encouraging to see the community continuing to grow.
A huge thank you to the Manchester Java Community for organising another excellent event, and to AutoTrader for hosting us in their impressive new office. I really enjoyed the day and I'm already looking forward to the next JManc.
Comments