AI Applications Architect - Agent-First Development

We are looking for a curious, technically capable builder who wants to develop quickly through real work at the frontier of applied AI.

At Nationwide Utilities, our CEO is already directing specialist agent teams across a connected business-software portfolio. The work spans customer operations, email and documents, supplier portals, bill validation, markets, procurement and other business applications. We are also developing the coordination tools that help manage the development itself.

These applications are at different stages of build, integration and release. Your opportunity is to help connect the pieces and turn ambitious development into dependable software people actually use.

This is not a conventional coding role with an AI assistant added. Agents are central to implementation. Your contribution is the requirements, judgement, coordination and hands-on engineering that make the result work. We value practical ability, learning speed and follow-through over a long CV or a senior title alone.

How we work

The CEO is hands-on: defining outcomes, directing teams, reviewing screens and making decisions. The development environment uses specialist planner, builder, checker and task-management sessions, with shared coordination across projects.

Clear “done means” goals, versioned instructions, evidence of results and controlled releases matter. So does the ability to spot when a busy agent is not moving the actual goal forward. You will help improve this method, not simply add more activity to it.

Your contribution

• Translate business goals into clear requirements, guidelines, reusable agent instructions and testable acceptance criteria.
• Break work into manageable stages, map dependencies and keep several agent-led workstreams moving without losing the critical path.
• Use Claude, Codex and appropriate local tools for implementation, testing, review and documentation.
• Read code, debug problems and integrate changes. Know when to direct an agent, when to intervene and when to ask for experienced review.
• Distinguish built, tested, integrated and live. Challenge sample-data demos and confident agent summaries until the actual user journey is proven.
• Help make MVPs production-ready: data integrity, permissions, security, testing, useful interfaces, deployment, monitoring and rollback.
• Coordinate shared testing and deployment resources; resolve stale instructions, unclear ownership and failed hand-offs.
• Reduce avoidable supervision by giving the CEO clear progress, concise decisions and early warning of genuine blockers.
• Our development environment includes four Blackwell GPUs, lightweight local computers and multiple premium Claude/Codex accounts. The goal is more useful, verified software, not simply more sessions, tokens or GPU activity.

• An insatiable, practical interest in AI: you experiment, learn and can explain what did and did not work.
• Software foundations: you can work with code, APIs, databases and version control, and diagnose a problem rather than blindly retry prompts.
• Pace and prioritisation: you can juggle workstreams, preserve context and focus on the next thing that will genuinely unblock delivery.
• Clear written thinking: an ambiguous idea becomes an executable instruction and an outcome a person can check.
• Finishing discipline: you care about usability, integration and reliability as much as producing the first working demonstration.
• Initiative, honest communication and a strong appetite for a demanding development opportunity.
• Experience with MCP, local models, GPU workloads, containers, CI/CD, Python or TypeScript is useful; you do not need to arrive as an expert in everything. Energy-sector experience is welcome but not essential.

Relevant employment, independent projects and substantial learning projects can all demonstrate your ability. This is learning through hands-on contribution alongside the CEO, not a promise of a formal training programme. The scope you own and your production-release authority must match demonstrated capability and agreed review arrangements. One developing hire is not expected to replace every engineering speciality.

This is a permanent, full-time role, with ordinary working days 100% office-based in Kensington, London W14. It is not a hybrid role.

The agentic development environment is intended to operate around the clock. You would be expected to remain available for remote intervention at any time outside office hours, including nights and weekends, using a mobile device or laptop. Active intervention is needed only when required to keep agent work progressing; this is not a statement that you would be working continuously.

Specific ordinary working hours, monitoring frequency, response expectations, cover, rest and the treatment of out-of-hours work within or beyond the stated salary will be discussed and documented before appointment.

Apply

Apply through Indeed with your CV and a short account of something you built: the problem, your own contribution, how you directed agents and what you checked before calling it complete. We are interested in the difficult parts and your reasoning, not just a polished demo.

AI-assisted applications are welcome, but you should be able to explain your own decisions. Share only material you have permission to disclose; anonymised examples are welcome. Applicants of all ages and backgrounds are welcome. Let us know if you need an adjustment to the recruitment process.

Pay: £35,000.00 per year

Ability to commute/relocate:

• London W14: reliably commute or plan to relocate before starting work (required)

Work authorisation:

• United Kingdom (required)