ServerlessDays Cardiff 2026, from functions to agents

At this year's ServerlessDays Cardiff,  the conversation went well beyond serverless. Speakers from AWS, Databricks, Microsoft, the DVLA and others had as much to say about generative AI, governance, security and digital sovereignty as about functions and event-driven design. Here’s what stood out for me.

First, serverless…

Serverless is a way of building and running applications without managing servers yourself. The cloud provider handles provisioning, scaling and maintenance, and you pay only for what you actually use. 

Instead of running machines around the clock, you write code (often as small functions) that runs in response to events such as a web request, a file upload or a message arriving on a queue. Despite the name, servers still exist. You just don't have to look after them.

The appeal is simple. It means a lower operational overhead, automatic scaling and faster delivery, so teams spend their time on the product instead of the infrastructure that enables it.

Serverless has moved well beyond short-lived functions answering HTTP requests 

The conversation in Cardiff was about long-running, stateful, AI-driven workflows that still have no servers to manage. The platform is evolving to fit agentic systems.

The shift is visible in the programme itself. Earlier editions of ServerlessDays centred on services like Lambda, while this year's agenda was far broader, with data, AI and agents at the centre.

Lambda is no longer "just a function"

In his opening keynote, "Serverless Compute for Workloads You Didn't Expect", Julian Wood, Principal Developer Advocate at AWS, presented the new Lambda form factors and capacity providers. Where Lambda manages the sandbox, there are Event Functions (the familiar Lambda function, renamed, with durable functions for workflows of up to a year using checkpoint and replay) and Web Functions (a standard HTTP server). Where you manage the sandbox, there are MicroVMs. On the capacity side, workloads can run on-demand, or on Lambda Managed Instances, which suit steady traffic and use instance pricing. 

My takeaway is that Lambda is now a menu of options, and the right choice depends on how much of the sandbox you want to manage and how your traffic behaves.

Durable functions make workflows ordinary code

Gunnar Grosch, Developer Advocate at Sinch, and a former AWS employee and AWS Serverless Hero, showed how AWS Lambda durable functions can be used to build conversational agents in code. Execution is checkpointed and can pause and resume, so waiting on a user, a human approval or an external system no longer needs heavy orchestration. That makes them a strong fit for multi-step and agentic AI workloads.

A practical technique: RISEN prompting

Gunnar also introduced RISEN, a simple structure for writing more reliable prompts: Role, Instructions, Steps, End goal and Narrowing. It's a useful pattern when you're building conversational agents, because it gives the model clear boundaries and a defined outcome.

Managing dependency

Kyra Wulffert, AI Specialist Architect at Databricks, gave a talk titled "From Shipping Models to Operating Agents". She framed dependency as something with eight dimensions, not just a technology choice. As organisations move from shipping models to operating agents, they take on dependencies that need deliberate management. Serverless trades infrastructure management for platform reliance, so it's worth knowing where you accept that trade-off.

Voice AI is moving into the mainstream

In a conversation I had with practitioners at the AWS stand, it became clear that speech is a growing focus. The product to watch is Amazon Nova Sonic, a speech-to-speech model that handles speech recognition and generation together instead of as separate steps. Paired with serverless back-ends, this makes voice-driven applications much easier to build.

AI governance has to keep pace with agents

Samuel Waymouth, Security Specialist Solutions Architect at AWS, spoke on AI governance from foundations to agentic speed. As agents act faster and more autonomously, controls such as auditability, access boundaries and clear ownership need to be designed in from the start, not added afterwards.

AI on both sides of the SOC

Robert Kitching, CTO of PureCyber, gave a lightning talk on the impact of AI on the security operations centre, looking at AI for both attack and defence. Attackers are using AI to become more capable, while defenders can use it to strengthen their security operations. The same technology shapes the threat landscape on both sides, which is another reason to build security in from the start.

Developer friction in the AI age

Grace Briody of Microsoft, in "The Friction Tax: What's Actually Slowing Developers Down in the Age of AI?", argued that AI has not made developers faster overall. Developers may write code ten times faster, but the rest of the pipeline hasn't kept pace. Access to environments, clear documentation, debugging and testing, and approval boards all remain bottlenecks. She also observed that genuine AI investment mostly reaches developers, while few cloud engineers get Copilot licences and few DevOps teams get test environments to build AI agents. 

The question Grace left with the audience was where does the friction in your delivery lifecycle actually lie, in the speed of code generation or the speed to production?

Modern ambition, legacy reality

Tom Brewer, DVLA's Chief Digital and Technology Officer, spoke about the agency's future technology ambitions and the first-hand reality of transforming legacy systems within government. It was a useful counterweight to the AI and agent talk. Most organisations are not starting from a blank page, so the real work is moving forward through the constraints of existing systems.

It’s possible to innovate in financial services without compromising trust

Sarah Jones, CEO of FinTech Wales, moderated a panel with Jen Lothian (CEO and co-founder, Onteric), Charles Radclyffe (CEO, EA Global AI), Oliver Conger and Alex Labonne. 

One standout point came from Charles, who identified back in 2016 the capabilities AI wouldn't, or shouldn't, take over: strategic thinking, human relationships, creativity and accountability. Creativity is now being encroached on, but the other three have held up. 

The panel also spoke up for neuro-symbolic AI and the value of trustworthy AI wherever the reliability of data matters, with good data, good ontology and being outcome-led as recurring themes. A key message was that AI solutions must be customer-centric, and that the best ideas often come from exchanging ideas with other industries. 

Oliver Conger recommended Cameron Herold's book Vivid Vision, which is about describing where you want your business to be in three years.

Risks and things to watch

For all the progress that was on show in Cardiff, the talks were just as clear about the trade-offs. Here’s what I’d watch out for:

• Cost creep as workflows run longer and AI calls multiply
• Latency in real-time and voice use cases
• Observability across event-driven chains
• Lock-in and platform dependency as you build on a single provider
• Governance gaps as agents gain autonomy
• AI-enabled attacks raising the bar for security teams
• Friction beyond code generation, such as environments, approvals and testing, limiting the gains from AI
• Legacy constraints slowing modernisation, especially in regulated and public-sector environments

Opportunities

The opportunities are just as real. The platform now handles much of what used to make ambitious ideas impractical, which opens up new options for teams of all sizes, including those still working with legacy systems. These are what excited me most: 

• Faster prototyping of AI-enabled services
• Less operational overhead, so teams can focus on the product
• New workflow patterns that were once far too complex to build and maintain
• Incremental modernisation, wrapping event-driven patterns around existing systems

ServerlessDays Cardiff showed how far the conversation has travelled, from functions to agents, and from infrastructure to everything that now sits on top of it. 

Serverless hasn’t faded into the background so much as become the foundation that AI, data and agentic systems are built on. But the talks that stayed with me were the ones that kept their feet on the ground. After all, legacy systems still need modernising, governance and security still need designing in, and the bottlenecks to delivery are as often human as technical. 

The technology is moving quickly. But the real work, as ever, is deciding where it genuinely helps and making sure that it is built responsibly. To find out more, visit the ServerlessDays Cardiff website or see the full agenda.

About Bhavana Ananda

Bhavana Ananda is a Senior Software Engineer at Scrumconnect. With more than two decades’ experience in IT delivery, Bhavana’s career has taken her across web applications, AI, semantic web and linked data, ERP and ETL in financial services, systems tools development and embedded software, giving her an unusually broad view of how technology stacks fit together end-to-end.