AI-accelerated · embedded developer

Compound · Workplace pensions · Fintech

Adding a senior developer to a pension provider's team, and shipping like three

Compound is a UK workplace pension provider with an in-house engineering team. We embedded a senior developer working AI-first, and over eleven months helped ship the employer portal, a staff admin app, a mobile app and a performance overhaul, inside their process and their tools.

3
apps

employer portal, staff admin, mobile

11
months

embedded in the in-house team

122
call sites

fixed in one performance pass

Every PR
AI-reviewed, then human-reviewed

no exceptions

The problem

Compound runs workplace pensions for employers, their accountants and the people saving into them. That means contribution uploads in the pension industry's standard format, direct debits, opt-ins and opt-outs, transfers between providers, identity checks, and a long list of integrations with payroll, accounting and banking systems. A regulated product with a lot of moving parts and a real engineering team already working on it.

What they needed was not an agency to take it away. They needed more senior capacity inside the team they had, working in their repos, their tickets and their stand-ups, at a pace the roadmap demanded.

What we did

Kai joined the team in September 2025 as an embedded developer, working AI-first under Compound's own conventions: one ticket per commit, branches off their develop line, and an automated AI review on every pull request before a human reviews it. Nothing merges without a person putting their name to it.

The work covered most of the product. On the employer portal: payroll and accounting integrations, direct debit mandate flows, the advisor experience for accountants who manage many companies at once. A new internal admin app for the Compound staff. A large share of the mobile app, including moving onboarding off the device and onto the backend. Batch generation and sending of letters of authority for pension transfers. And in August 2026 a performance overhaul of the GraphQL layer, where encrypted columns had made every wide query expensive: resolvers became selection-aware, an N+1 was removed, and a client-side cache that had been quietly mixing responses across sessions was replaced across all one hundred and twenty-two call sites.

Towards the end of the summer we also rebuilt the admin and portal front ends as greenfield Next.js apps against their new Laravel backend, and wrote the API contract for the move.

What changed

Compound's CEO put it better than we can, below. From our side the numbers that matter are the ones about pace: close to a thousand commits in eleven months from one developer, across three apps and the infrastructure that runs them, with an integration test suite that had been broken for months brought back to green along the way.

It is the clearest example of what the embedded plan is for: a senior person, AI in the loop, inside your team rather than instead of it.

We'd spent two years and a significant budget trying to get our pension app to market. Ikigai came in, assessed the architecture, and essentially rebuilt the product. We launched in three months. The speed and quality blew us away.
- Dan, CEO, Compound

Yours next?

Let's write your case study.

Book a call