Forward Deployed Engineer: Foundations
The Forward Deployed Engineer Foundations certification is a 50-question, 45-minute exam that proves you can ship working features with coding agents (Claude Code, Cursor, OpenCode, Codex) the way a senior engineer does, not by chat-completion guessing.
How the credential happens
- 01
Pay $25.00
Pay once via Paddle. No subscription.
- 02
50 questions · 90 min
Server-timed. Randomized question + option order per attempt.
- 03
Pass at 40 / 50
Server scores instantly. No partial-credit games.
- 04
Retry for full price
Available 2 hours after submit. Fresh question subset on each retry.
- 05
Public certificate
Verify URL with issuance date, recipient, live status.
See a sample verify page →
Who this cert is for
If you are early in your engineering career
You probably built your first agent-assisted feature by trial and error and got something working. The cert is the structured version: it tests whether you actually understand why the working version worked, so you can ship the next ten without re-discovering it. Hiring managers ask this in interviews now; pass the exam and you have the answer on paper.
If you have been shipping code for years
You have an opinion about coding agents and a small graveyard of half-finished side projects to back it up. The cert is the version your principal engineer respects: server-scored, no token-padding tricks, no vendor PR. Prove you can be the senior on the team people ping when the agent runs off the rails. Status that is earned, not announced.
What you’ll prove
- Configure a coding agent with the right tools, scopes, and guardrails for the repo you are working in.
- Set context deliberately so the agent reads what it needs and ignores what would mislead it.
- Decompose a feature into the size and shape of unit an agent reliably finishes in one shot.
- Give the agent instructions clear enough that the resulting code is reviewable, testable, and shippable on the first PR.
- Organize the work so the human reviewer can actually catch the wrong-but-plausible code the model produces.
What the exam covers
- Agent configuration and setup (10 questions)
- Context-setting. What to feed, what to hide (10 questions)
- Splitting and decomposition. Right-sized units of work (10 questions)
- Out-of-distribution with clear instructions (10 questions)
- Organizing for testing and review (10 questions)
Sample questions
Real questions from the bank, in the same shape you’ll see in the exam. Correct answers are not shown.
- Q01Context-setting
You're asking a coding agent to add a new endpoint to a 30-file Go service. The agent has a tool to read files. Which approach is most likely to produce a reviewable PR on the first attempt?
- Tell the agent to read all 30 files first, then add the endpoint.
- Paste the contents of two adjacent endpoint handlers into the prompt and point the agent at the route table file.
- Give the agent a vague high-level description of the service and let it explore the codebase.
- Run the agent in a loop until tests pass, regardless of what it changed.
- Q02Splitting and decomposition
A teammate asks the coding agent to 'add user authentication, password reset, and admin user management.' The agent produces 1,200 lines of new code across 14 files. Which decomposition gives the highest probability of correct, reviewable output?
- Same prompt, but ask for tests in addition.
- Three separate prompts: one for auth, one for password reset, one for admin management, each finishing on a green test before moving on.
- One prompt that lists detailed acceptance criteria for all three features.
- Let the agent do all three at once but use a more capable reasoning model.
- Q03Organizing for testing
Your coding agent has been asked to refactor a complex pricing calculator. It produces code that passes the existing test suite. What is the highest-leverage next step before merging?
- Merge: passing the test suite is the contract.
- Read the diff and have the agent add tests for the specific edge cases the old code clearly cared about.
- Run the agent again on the same prompt and merge if it produces the same output twice.
- Ask the agent to explain its changes in plain English and merge if the explanation reads well.
Honest tradeoffs
- · Not a generalist AI literacy test. We assume you already know what a model is.
- · Not a vendor cert. There is no Anthropic-specific or Cursor-specific section.
- · Not a sub-agent or workflow architecture exam. That is the Systems Foundations cert.
- · Not a guarantee any specific employer will accept it. Treat it as a strong public signal.
It is proof you can ship code with a coding agent the way a senior engineer ships it: deliberately, reviewable, with the failure modes named ahead of time.
Frequently asked
- I have been using Claude Code or Cursor for a year. Do I need this?
- Use and structure are different things. The cert tests whether you can deliberately configure, decompose, and test agent-driven work. Most heavy users have learned by trial and error and discover during practice that their intuition has holes the cert names explicitly.
- How is this different from a vendor certification (Anthropic, Cursor, etc.)?
- Vendor certs prove you know one product. This is an exam on the underlying engineering discipline: configuration, context, decomposition, instructions, testability. Same answers apply whether you use Claude Code, Cursor, OpenCode, Codex, or whatever ships next quarter.
- Will employers recognize this credential?
- Each certificate has a public verify URL at academy.juscode.co/c/<id>. Recruiters click once and see the issue date, your name, and live status (valid, expired, revoked). Paste it into LinkedIn Licenses and Certifications, an email signature, or a resume PDF.
- Is the exam vendor-neutral?
- Yes. Questions describe the situation (a coding agent with file access, a tool budget, a repo with N files) without naming a specific product. If a question implies a product feature, that feature is one every mainstream agent platform offers today.
- What if I fail?
- Pay $2 to retry once after a two-hour gate. The gate is automatic and protects against brute-force guessing. The retry uses a fresh question subset.
- How long is the cert valid?
- One year from issue. After that the public verify page shows the cert as expired and you can sit a renewal exam.
Before you start
- · You’ll capture the name on your certificate before the exam starts. It cannot be edited after publishing.
- · 1.5 h hard timer. The server auto-submits whatever’s answered when it expires.
- · One attempt per payment. Next attempt available 2 hours after submit.
- · Read the honor code. AI-assisted attempts may be reviewed and certificates revoked.
What this changes for you
You ship features with a coding agent and most of the time it works, but you cannot quite explain why this prompt worked and yesterday's did not. The senior engineer reviewing your PRs is starting to ask harder questions.
You name the context budget, the decomposition, the instruction shape, and the test plan before you start. Your PRs land reviewable on the first pass and you have a credential that says a senior engineer would respect the discipline behind the work.
Exam integrity & refund policy
This is a proctored exam. By purchasing and beginning it, you agree to the honor code and to automated, in-browser integrity monitoring of your session. Conduct that breaches the integrity policy, including leaving the exam window, switching tabs or applications, using external help or AI tools, or copying, cutting, or pasting exam content, is recorded. Repeated violations result in immediate forfeiture of your exam attempt.
Exam fees are non-refundable once an attempt has begun, including attempts ended early for integrity violations. jusCode Academy may withhold, suspend, or revoke any certificate obtained through conduct that violates this policy, with or without prior notice.
One attempt per payment. Next attempt available 2 hours after submit.