How I’d prep for Affirm’s Software Engineer I, Continuous Delivery role in 30 days
Updated Nov 16, 2026
The job
Software Engineer I, Backend (Continuous Delivery), Affirm
Remote (US) · $130,000 – $180,000 base in CA, WA, NY, NJ and CT; $115,000 – $165,000 in other US states; plus equity · Posted Nov 12, 2026
View the posting →What this team actually does
Affirm lets people pay over time at checkout. Every one of those checkouts runs on code that Affirm’s engineers ship every day, and the Continuous Delivery team builds the system that ships it. The team sits in Developer Productivity, which the posting says “owns much of the path between a developer’s laptop and production”: developer environments, CI/CD pipelines, test automation, dependency management and release engineering.
The team’s job is to let other teams “release independently through strong automation, standard patterns, predictable rollouts, resilient recovery mechanisms, and less manual toil.” The posting also says why it matters right now: the software development lifecycle is “increasingly AI-enhanced”. When AI helps everyone write code faster, there’s more code to ship, and shipping it safely becomes the hard part.
It’s an entry-level role, it’s fully remote, and the posting counts “work or internship experience”.
| Area | What to know |
|---|---|
| Language | Python or Kotlin (the posting’s words). I’d pick Python |
| Runtime | Kubernetes on AWS. Affirm runs “90+ EKS clusters across environments and regions” |
| Data | MySQL |
| CI | Buildkite, with code on GitHub and build artifacts in Artifactory |
| Repo | A 10GB monorepo, with caching so builds and checkouts stay fast |
| Infrastructure as code | Terraform (run through Spacelift), on the platform team the deploys land on |
| Testing | Pytest and Playwright on the sister Test Infra team |
How I found the stack (do this for any job)
- Read the posting closely. It says “Python or Kotlin” and “AWS, MySQL and Kubernetes”. That’s a start, but it doesn’t say what the pipeline runs on.
- Read the other postings from the same org. A Staff Software Engineer, Continuous Integration posting from Developer Productivity names the build tools: “GitHub, Buildkite, Artifactory”. A Software Engineer II, Test Infra posting names Pytest, Playwright, Docker and AWS.
- Read the platform team’s postings too, since that’s where your deploys land. Affirm’s Compute Platform posting says they operate “90+ EKS clusters” and use Terraform and Spacelift.
- Look for talks by the company’s engineers. Buildkite hosted a webinar with a senior staff software engineer at Affirm on the technical architecture behind Affirm’s Buildkite setup. It covers how they cut cold repository checkout times “from 10+ minutes to under 45 seconds” with nightly snapshots and S3 caching, and “the trade-offs between monorepo benefits (auditability, compliance, unified workflows) and performance challenges at scale.” Watch it before your interview. Those trade-offs are exactly the kind of opinion question this team can ask.
So: Python, Buildkite and GitHub, shipping to Kubernetes on AWS, with MySQL. That’s where I’d spend the 30 days.
Understand the product first
This team has two kinds of users: Affirm’s customers, who feel it when a bad deploy breaks checkout, and Affirm’s engineers, who use the pipeline every day. Spend an hour as both.
- Open the Affirm app or how it works and walk through getting a pay-over-time offer at checkout. Picture what a broken deploy does to that screen in the middle of a sale.
- Install kind and run a small web app on Kubernetes on your laptop. Change the code, build a new image and roll it out. Watch
kubectl rollout statusas pods are replaced one by one. - Ship a broken version on purpose (make it return errors) and roll it back with
kubectl rollout undo. Time how long the bad version was live. - Add a readiness probe and ship the broken version again. Notice how Kubernetes now refuses to send traffic to pods that aren’t ready.
Then ask yourself the questions this team answers every day: how do you know a deploy is bad before most users see it? How fast can you undo it? What happens when the old and new versions run at the same time and read the same database?
System design questions
- Design a deployment system for hundreds of services across many Kubernetes clusters. How does a team ship without waiting on anyone else?
- Design a canary release: send 10% of traffic to a new version, decide automatically whether it’s healthy, then promote it or roll it back.
- A deploy goes out and checkout errors go up 2%. Walk through how your system notices and what it does next.
- Blue/green vs canary vs rolling updates. When would you pick each, and what does each one cost?
- Your CI takes 40 minutes on a big monorepo. How would you make it fast? (Caching, only testing what changed, splitting the work across machines.)
- Monorepo or many repos? Argue both sides for a company like Affirm.
- How do you rename a column in MySQL while the old and new versions of the code are both live?
- An emergency fix has to reach production in minutes. What can you skip safely, and what can you never skip?
Concepts they’ll probe
- CI vs CD vs continuous deployment, and what each stage of a pipeline is for.
- Deployment strategies: rolling updates, blue/green, canary, and feature flags (separating deploy from release).
- Rollbacks: what makes them fast, and why some changes (like database migrations) can’t simply be rolled back.
- Health signals: error rate, latency, readiness and liveness probes, and how to compare a canary against the old version.
- Kubernetes: Deployments, ReplicaSets, Services, probes,
maxSurgeandmaxUnavailable. - Compatibility: old and new code running side by side, backward and forward compatible data, expand/contract migrations.
- Build speed: caching, incremental builds, running only the tests a change affects, monorepo trade-offs.
- Reliability of the pipeline itself: retries, flaky tests, idempotent deploy steps, audit logs.
- DORA metrics: deployment frequency, lead time, change failure rate and time to restore.
- Python: writing clean, tested code, calling APIs, and handling errors and timeouts.
Unlock the rest
- The week-by-week 30-day plan, with daily practice on Devmaxx
- The exact book chapters and docs for each week
- The project: a mini deployment pipeline in Python, with GitHub, Buildkite, canary releases on Kubernetes, automatic rollbacks and a MySQL deploy log
- The entry-level bar: what they expect from someone with internship or early work experience
- The resource list with every link
Get the full guide, free
Enter your email and the guide opens right here. You’ll also get the occasional Devmaxx email. Unsubscribe anytime.