AI Platform Engineer, Agentic Interfaces

Build and operate the interface, semantic layer and controls that let AI agents work inside IFS software safely.
This role is hybrid or remote.
The role:
You will own the web client of an AI agent product for our enterprise customers. It is a React and TypeScript single-page application, and almost everything on its screen arrives as a stream from a language-model agent.
What you will do:
Build an interface that people can trust while the content on it is still arriving, changing and occasionally wrong.
Turn what the agent produces into a usable interface: prose, generated views and actions the user can take.
Make sure the agent acts only with the user's explicit consent, and that people can always see what it did and why.
Own the front-end architecture, the component system and the contracts with the teams behind the agent.
Help move the interface towards Lit and Web Components, so that what we build is standards-based and not tied to a single framework or application.
Treat accessibility, performance, testing and observability as part of the product rather than a late pass.
Work across the boundary into the services your interface depends on, then ship what you build and run it in production.
Why it is interesting:
Most front-end work assumes that a page loads and is then correct. Here the content arrives in pieces, gets revised, sometimes fails partway and is sometimes wrong. And some of it is code a model wrote, which the browser should not trust. The protocols involved are young and still changing (AG-UI, A2UI, MCP Apps). You will help shape them, not just consume them.
What we are looking for:
Substantial production single-page applications, built and operated, with real depth in TypeScript and React.
Asynchronous state you can reason about.
Browser security as a working concern: iframe sandboxing, postMessage, Content Security Policy, sanitising content you did not write.
Component and design-system work in a shared codebase, including web components used inside React.
Accessibility as a practice. You know how to keep it tested.
Willing to write Go and to read Python. Prior experience in either is not required.
Test-first habits and continuous delivery.
Clear writing. Decisions are recorded as short architecture decision records (ADRs), and protocols are documented before they change.
Comfortable working with AI coding agents as part of how you build.
Also valuable:
A product surface built on a language model in production: streaming, partial output, retries, and showing uncertainty.
Hands-on knowledge of the Model Context Protocol (MCP) and MCP Apps, AG-UI or A2UI.
Experience building with Lit and Web Components, including shadow DOM and custom elements used from other frameworks.
Expression languages or compilers.
A platform where behaviour is configured through data or metadata rather than written in code.
Data visualisation.
Familiarity with Docker, Kubernetes and Helm.
Knowledge of an enterprise domain: asset management, field service, manufacturing, supply chain.
Strong software engineering first. Everything else is applied on top of that.
Production experience building and operating enterprise systems, with real depth in distributed systems, cloud-native architectures, API and schema design, event-driven systems, security, observability and CI/CD.
Strong programming in a modern backend language. See the note on language under “Open calls” before advertising.
Experience delivering AI systems built on large language models, retrieval-augmented generation (RAG), agentic workflows and orchestration frameworks, including tool use, function calling, workflow orchestration and autonomous or multi-agent architectures, with the judgement to know where they fail.
Evaluation as a discipline: experimentation, benchmarking, prompt engineering, tracing, quality measurement and agent tuning, improving an agent against evidence rather than impression.
Ability to design solutions that integrate enterprise applications, business processes, workflows and data platforms.
Depth in at least one of the following:
Tool-surface and agent-runtime engineering. MCP servers, tool ecosystems, capability modelling, discoverability, governance, versioning, backward compatibility, multi-tenancy isolation.
Knowledge graphs and semantic modelling. Ontology design, context engineering, embeddings, vector databases, retrieval and search technologies, memory architectures, grounding strategies.
Enterprise platform depth. Oracle PL/SQL, OData, and comfort working inside large metadata-driven systems where behaviour is configured rather than coded.
Desirable
Experience with agent frameworks such as Semantic Kernel, Microsoft Agent Framework, LangGraph, AutoGen, PydanticAI, the OpenAI Agents SDK or CrewAI.
Experience building reusable AI platforms, MCP ecosystems or shared engineering capabilities used across multiple products and teams.
Containerised platforms and infrastructure automation: Docker, Kubernetes.
Experience with Azure, AWS, GCP or another hyperscale cloud platform.
Reverse-engineering or interpreter work.
Token-efficient agent design.
Enterprise software domains: enterprise asset management, service management, manufacturing, supply chain, aerospace and defence, energy, telecommunications, construction, industrial AI.
Contributions to open-source projects, technical communities, conferences, publications or standards.

Employment type
Full-time

Work arrangement
No