Software Engineering Recruitment for AI-Native Teams

A senior software engineer hired into Podium, part of PA Media Group, with no AI signals on the CV and no professional experience of the stack being built, who has since become the profile the client calibrates every later hire against.

On paper this was a C#/.NET backend engineer with nothing on the CV about AI at all. No LLMs, no agents, no tooling. Any keyword search would have discarded the application in seconds, and several probably did. The brief made it harder still: a TypeScript and Node platform being built from scratch, a hybrid pattern requiring regular time in the office, a budget below the market rate for the experience being asked for, and an explicit instruction to avoid engineers who over-engineer. That last one rules out a great many people who interview well.

Two things decided it. The technical assessment chose Server-Sent Events over WebSockets for a one-way data feed and justified it on protocol overhead rather than novelty, picked a framework on benchmarks and native type support, decoupled ingestion from delivery with an internal event bus, and set out what scaling it would actually require. Restraint, argued for. The second was a way of reviewing other people’s code by severity, sorting review comments into boulders, pebbles and dust so a senior engineer never becomes the bottleneck on somebody else’s preferences. Placed as a Senior Software Engineer and now the benchmark the client uses when describing who they want next. When a client starts saying "someone like that one" instead of restating the job spec, the calibration has landed.

Podium, part of PA Media Group, hired a software engineer with no professional TypeScript on the CV at all, on the strength of learning velocity and genuine AI fluency rather than stack tenure.

The specification asked for years of professional TypeScript, a hybrid pattern that ruled out remote-first candidates, and a commitment to an on-call rotation once the team settled. Each of those narrows a pool. Together they close it. The strongest person we found had built a career in PHP and Laravel and had none of the TypeScript tenure the brief asked for, which on a conventional read is where the conversation ends.

The technical assessment answered the stack question outright: a clean pivot into TypeScript, React and Node, full type safety with no escape hatches, test coverage across the suite, real-time updates over websockets and schema validation on every request. It came with a documented AI-native workflow, development logs showing where the model had been directed and where it had been overruled, and internally built Model Context Protocol servers written to make the work more accurate. Placed alongside a senior engineer into the same new team.

StreamAMG, part of PA Media Group, hired a software engineer whose CV read junior and whose technical assessment was the most production-shaped, and best documented, we had reviewed for that role.

On experience alone this candidate read junior, and on a conventional sift would have been filtered out on years rather than assessed on work. The hiring manager’s own reaction after the first-stage interview was that the experience looked junior but the person did not. That gap is invisible on a CV and it is the whole reason this kind of hire gets missed.

The technical assessment settled it. A production-shaped submission rather than a toy one: interface-first abstractions, dependency injection, test coverage across every layer including a test that listeners were cleaned up on disconnect, and forty-plus atomic commits. The AI workflow documentation was the strongest we had seen on an assessment of that type, showing where the candidate had pushed back on the model rather than accepted it. Offer made five weeks after the brief opened.

PE Limited hired a senior software engineer into its AI team, not to build models but to connect them to a C# and C++ product suite that customers already depend on.

Not an AI engineer, and not hired to be one. PE’s core products are built in C# and C++, and an AI team producing models and prototypes is worth very little until somebody wires that work into the software customers actually use. That job needs deep conventional engineering in the languages the estate is written in, alongside enough understanding of what the AI team is building to be a useful counterpart rather than a downstream ticket-taker. It is a combination almost nobody advertises for, and Guildford, five days a week, narrowed it further.

Sits inside the AI function, connecting the models and AI products it builds to an existing suite of engineering software. The bridge between an AI team and a real product line, which is where a lot of AI programmes quietly stall.