OB Collective develops independent verification technology for organisations operating in government, regulated and high-assurance environments.
Our work is concerned with establishing reliable evidence about information and automated actions before organisations rely on them.
We’re building a senior technical team for a six-to-twelve-month engagement on a UK public-sector R&D programme focused on the cybersecurity and operational assurance of AI agents.
We’re looking for a Senior Software Engineer – AI Agent Integrations to develop the integrations connecting agent environments and enterprise systems with our existing verification capability.
As AI agents gain access to organisational data, credentials, tools, APIs and enterprise systems, organisations need stronger ways to understand what agents are permitted to do, what systems they can access and the actions they actually perform.
The programme will extend an existing software capability to collect and verify relevant evidence across different agent environments and integrate the results with organisational security controls.
There is an existing technical capability and a defined research direction.
The challenge is to connect that capability to agent frameworks, identity services, tools and enterprise systems while preserving the provenance, consistency and limitations of the evidence collected.
You will work closely with the Principal Technical Lead and Verification Engineer to build adapters, interfaces and integrations across managed enterprise and open/self-hosted AI environments.
This is a hands-on software engineering role rather than an ML research or model-training position.
We need someone who has actually built and debugged applications involving agents, tool calls, APIs and multiple systems, and understands how behaviour can change across integration boundaries.
The role is UK-based, outside IR35, full-time and primarily remote, with occasional travel for workshops and technical activity.
What you’ll do
- Design and build integrations with AI-agent frameworks and orchestration environments.
- Develop adapters for agent actions, tool calls, APIs, service gateways and enterprise applications.
- Integrate relevant authentication, identity, permission, policy and approval information.
- Work with managed enterprise and open/self-hosted agent environments.
- Implement integrations using MCP and comparable tool-interface patterns where appropriate.
- Establish clear interface contracts and observable evidence boundaries.
- Distinguish independently collected evidence from information asserted by originating platforms.
- Build interfaces through which existing security and policy controls can consume verification results.
- Develop portable integration patterns that minimise unnecessary vendor dependencies.
- Create automated integration tests and investigate behavioural differences between platforms.
- Support controlled failure, bypass, portability and resilience testing.
- Document interface behaviour, evidence limitations, configuration requirements and coverage gaps.
What we’re looking for
- Strong hands-on software engineering experience, particularly Python or backend application development.
- Significant experience integrating applications or platforms through APIs, events, middleware or SDKs.
- Practical experience developing working applications using LLMs, AI agents or tool-calling frameworks.
- Good understanding of authentication, authorisation and enterprise identity concepts.
- Experience designing and debugging integrations across multiple systems and external APIs.
- Strong automated testing and troubleshooting skills.
- Experience with cloud environments and modern software engineering practices.
- Ability to reason about provenance, security boundaries and the reliability of information obtained through integrations.
- Strong interface design and technical documentation skills.
Preferred experience
- LangGraph, LangChain or comparable agent frameworks.
- Microsoft Foundry or other managed enterprise AI platforms.
- Model Context Protocol (MCP).
- Microsoft Entra ID.
- OAuth 2.0 and OpenID Connect.
- Enterprise policy engines or approval workflows.
- Azure, Docker or self-hosted applications.
- Open-weight models and model-provider-independent interfaces.
- Security-control or monitoring integrations.
- Experience training or fine-tuning machine-learning models is not required.