AI Staff Engineer in Norwich

Location: Norwich Office or Sofia office
Reports to: SVP Engineering
Type: Permanent, full-time
Level: Staff IC
Why we're hiring
We've got an AI strategy with two pillars: making our own teams faster, and shipping AI into
our platform. Adoption is moving. Funding is in place. OpenAI, Cursor, Claude Code, and
Bedrock are all live in some form.
What we don't have is a single person who owns the platform underneath all of it. That's this
role.
You'll build and run the shared platform we use for AI: model access, cost governance,
evaluation, safety. When a squad says "we should use AI for that", they shouldn't have to
start from scratch.
What you'll own
The AI gateway. A paved way for engineers and product squads to use the AI tools we've
picked. Consistent auth, logging, fallbacks, and cost attribution.
Bedrock and AgentCore. Lead our adoption. We're evaluating AgentCore for agentic
workloads now. You'll take it through to production: architecture, cost model, integration with
the rest of our AWS estate.
Cost governance. Per-tribe visibility. Alerts before the bill, not after. Tied to where the
spend is paying off and where it isn't.
Evaluation. A standard way to test AI tools and features, and to catch regressions when
models change underneath us.
Safety. Prompt injection, PII, output filtering, audit trails. Pragmatic, proportionate to the
risk, not bureaucratic.
Adoption. Building the platform isn't enough on its own. You'll work with EMs and Staff
engineers across all five tribes to make sure it gets used, and the patterns we learn get
spread.
The AI Guild. A cross-tribe group that decides what we adopt, what we retire, and what's
worth experimenting with next. You'll run it.
Success metrics. Define what good looks like for internal AI tooling (cycle time, defect rate,
time saved) and for product AI features (quality, latency, cost per request, customer
outcome).
What success looks like
By six months
● AI gateway in production, used by at least one internal tool and one product feature.
● Cost dashboard in production. EMs can see what their tribe is spending.
● AgentCore and Bedrock evaluation done. A clear go / no-go with production evidence
behind it.
● First evaluation suite running against real AI features.
● AI Guild meeting regularly with people from all five tribes turning up.
By twelve months
● All product AI features go through the gateway. No squad is rolling its own.
● Every team shipping AI uses the standard eval pattern.
● AI spend is predictable and tied to value. Not necessarily lower; governed.
● Measurable cycle-time gains on at least two engineering workflows we can attribute
to internal AI tooling.
● RapidAI use cases shipping through the platform.
By two years
● AI is a normal engineering capability, not a special programme. New features take
days to wire up, not weeks.
● We can swap models without rewriting product features.
● AI cost, latency, and eval data show up in engineering decisions the same way DB
performance does today.
What we want from you
We care about how you think and what you've shipped. That said:
● You ship. You write code, dashboards, and runbooks that other engineers use.
You're not someone who'll spend three months on a strategy deck.
● You think in platforms. You build the version that works for everyone, not a
bespoke solution for each squad.
● You can hold a room. Staff engineers in the morning, a VP in the afternoon. You can
explain the same trade-off to both without losing either.
● You've changed your mind about AI before, based on evidence. You can tell us
about a use case where AI didn't pay off.
● You know the unit economics. You can tell the difference between "AI is
expensive" and "this pattern is expensive, here's a cheaper one".
● You understand the benefits and the risks of an AI first approach running at scale.
Tradeoffs between public models and self hosted solutions
● You know Bedrock in production. We're an AWS shop and Bedrock is our strategic
substrate. You should already have the IAM, VPC, throughput, and observability
scars. AgentCore experience is a big plus given where we're going.
Useful, not required
● AgentCore in production, or a comparable agent runtime (LangGraph Platform,
Vercel AI SDK, in-house)
● Built or operated an LLM gateway
● Built or run an eval framework in production
● Owned cost governance on a meaningful AI workload
● Shipped customer-facing AI and handled the security and legal conversations that
come with it
● Run a Cursor or Copilot rollout and know what made adoption stick
● Background in Platform, DevEx, ML Platform, or Applied AI. We're open.
How we work
● 5 engineering tribes (Money, POS, Business, Data, Platform), ~120 engineers.
● Offices in Norwich and Sofia.
● AWS-native. GitLab. Slack-first.
● OpenAI, Cursor, Claude Code, are in real use. AWS RapidAI funding is unlocking
customer-facing AI work.
● UK fintech SaaS scale-up. Sales-led, cashflow-conscious, willing to invest where the
upside is real.
● You'll report directly to me. Clear remit, exec sponsorship, the air cover to make
decisions stick.