We build software with AI agents. Not as an experiment, and not as a clever autocomplete: by 2025/26 agents were writing all of our code.
That hasn’t made engineering matter less. It has made it matter more. When code arrives this fast, architecture, system design, code quality and security are what stop it turning into a liability. Our engineers design the systems the code lives in, set the standards it has to meet, build the tooling, tests and gates that hold the agents to those standards, and review what comes back.
Anyone can get an AI to write code. Getting code you’d be happy to put in front of a client is an engineering job.
What our agents work with
The model is the least interesting part. What makes the output dependable is the engineering we put around it.
So when an engineer reviews a change, it has already been tested and checked against the platform it’s going to run on.
Why most AI-assisted development disappoints
How it usually goes
Someone prompts a chatbot that knows nothing about the platform. It guesses how the API behaves. A person checks every line by hand, and nothing learned survives to the next job.
What that costs
Code that looks right but fails in production. Rework and growing technical debt. Teams who stop trusting anything AI produces, and nobody able to say who changed what or why.
The answer isn’t less AI. It’s more engineering around it.
Part of everything we deliver
We don’t sell agentic development on its own. It’s how we deliver everything else.
Modernise
Before we change an old system, agents map what’s actually there: the schemas, the dependencies and the workflows built on them. They test our assumptions and show the impact of a change before it goes anywhere near production. They apply classification and governance consistently, and they refactor against the real schema, not a guess at it.
Explore ModernisationMigrate
Moving content or applications is rarely hard because of the copying. It’s hard because nobody’s quite sure what exists, how it’s structured, or how it should behave on the other side.
Agents analyse the source schemas and data, map them to the new platform, flag duplicates and bad data, and rehearse each transformation before running it. Afterwards, they check the result against the target platform’s rules and keep testing how it behaves. You move what you need, cleaned up, rather than carrying an old mess into a new home.
Explore MigrationBuild Faster
Agents produce the platform scaffolding, take on the repetitive and error-prone work, and hold every change to the same patterns. Our engineers spend their time on architecture and on the parts that need judgement.
It’s faster because the waste has gone, not because anything was skipped.
Explore Faster BuildingWhat’s different about how we do it
Built on Your Platform
Agents work directly with the platforms, APIs and systems your software depends on. They know your platform, not just the programming language.
Knowledge That Keeps
Fixes, patterns and lessons are recorded and reused, so each project starts where the last one finished instead of from scratch.
Checked Before Review
Automated tests and checks run against the real schemas and APIs before any person looks at the work.
Engineering First
Architecture, system design, code quality and security are engineering disciplines, and they matter more than ever when code is cheap. Our engineers own them, and it’s the engineers, not the AI, who are accountable for what ships.
What it means for you
- Software sooner, built to the same standard
- Less of the technical debt that AI-generated code usually leaves behind
- A record of every change: what it was meant to do, which model wrote it and which checks it passed
- AI-written code that has been checked before you rely on it
- A project whose knowledge lives in the repository, not in one person’s head
Where you’ll see it
In everything we build: cloud applications and systems integration, firmware for the microcontrollers and sensors that run physical systems, legacy modernisation, governance and compliance automation, and our own products, including Locodium, Transaction Manager and our SharePoint web parts.
The Impact
65% Faster Delivery
Typical reduction in the time from idea to a first release people can use.
45% Less Rework
Less work redone, because output is checked against the platform as it is produced.
80% Faster Onboarding
New people get up to speed faster because the project’s history and decisions are kept with the code.
95% Automated QA
Quality checks run automatically before anyone reviews the work.
Source: our own delivery figures, measured across JFDI projects from 2020 to 2026. Over that period agentic development grew until, by 2025/26, it covered all of our coding.
Working With Us
If you work with us, you’re already getting this, whether we’re modernising an old platform, migrating data or building something new.
We’re open about where and how we use AI, and we’re happy to be questioned on it. Ask, and we’ll show you the tools and walk you through how a change gets from an agent to production.
For the principles behind it, read our Agentic Development Manifesto. For the engineering detail (the tooling, the gates and the way we work), see how we do agentic development.
Talk to Us About Agentic AI Development
Whether you’re curious or ready to start, a pilot lets you see it working on one of your own problems.

