Career path
Agent Engineer
An agent engineer builds the software that lets a model take actions: the tools it can call, the permissions around them, the state that survives a failure and the evaluations that show the agent did the right thing. Employers rarely use the exact title, so read what a posting owns. Use this path to understand the work, build evidence for it with the site's agent course and prepare for the interview.
A 10-guide reading path
- Agent Engineer: A Practical Career and Portfolio Guide
Understand the layers of the work and the recoverable agent exercise.
- Agent Engineer vs AI Engineer vs ML Engineer
Tell an agent role from an AI or ML engineering role before applying.
- How to Become an Agent Engineer
Choose the lessons that close your gap and the three portfolio artifacts to build.
- Run a Bounded Tool-Calling Loop in Python
Separate what the model proposes from what the application authorizes.
- Build an MCP Server and Client for a Ticket-Label Tool
Expose a tool over MCP and test its schema and protocol failures.
- Human Approval for Agents: Store and Check the Exact Decision
Bind an approval to the exact change and reject stale or repeated requests.
- Evaluate a Ticket Agent: Dataset, Scoring Code and Failure Report
Score an agent with failures kept in the denominator.
- Agent Design Patterns: Run Four Control Flows and Their Failures
Decide when a fixed workflow beats a tool loop, and see how each fails.
- Build a Ticket Workflow with Signed Delivery and Approval
Connect delivery, retrieval and approval, then verify restart and replay.
- Agent Engineer Interview Guide
Prepare the agent design, evaluation, debugging and coding rounds.
Compensation in current employer postings
Checked September 26, 2026. Annual ranges from 6 specific employer postings. These are not market averages, comparable levels or guaranteed offers. Read the linked posting for the current range and package.
| Employer and role | Location | Posted range | What the range covers |
|---|---|---|---|
| Scale AI Frontier Agents Engineer (Applied AI) | San Francisco / New York | $180,000 - $225,000 USD | Base salary that may span several career levels; equity and benefits are separate parts of the package. |
| Elastic Agentic AI Engineer | United States (distributed) | $94,300—$149,200 USD | Typical starting base salary with no variable compensation; select locations including the San Francisco Bay Area and New York City metro use $113,300—$179,200 USD. Eligible for the stock program. |
| Sierra Software Engineer, Agent | San Francisco / Atlanta / New York (on-site) | $180K – $390K | The posting's compensation field, plus equity. It does not say whether the range is base salary. |
| Decagon Senior Software Engineer, Agent Product | San Francisco (on-site) | $200K – $400K | Base salary, plus equity. |
| Harvey Senior Software Engineer, Agents | San Francisco (hybrid) | $193,400 - $290,000 USD | Compensation range for hiring across mid, senior and staff levels; equity and a bonus are offered in addition. It does not say whether the range is base salary. |
| WRITER Software engineer, agents | New York City / San Francisco / Seattle (hybrid) | $146K – $240K | Base compensation in San Francisco and New York City, open to multiple levels; other US locations are $132K – $216K. Equity may be offered separately. |
Interview practice bank
Original practice prompts, not leaked questions or a claim about any employer’s interview loop.
An agent's tool call changed a record, but the worker crashed before saving the result. What happens next?
Draw the failure window, use a stable identifier or idempotency key to find out whether the action happened, and explain why a blind retry can apply it twice.
The model passes a correctly shaped ID for a record the user cannot access. Where is that stopped?
Trace identity from the request to the tool, check ownership in the application for reads and writes, and explain why a prompt instruction is not access control.
Design the tools for an agent that updates customer accounts.
Prefer narrow tools with explicit arguments, name the actions that need a person's approval, bind approval to the exact change and give every tool a visible error state.
The evaluation score improved, but complaints about the agent increased. What do you check?
Compare the dataset with real tasks, check final state rather than the agent's summary, test whether the judge accepts confident wrong answers and add the missed cases.
An agent keeps repeating the same failing tool call. How do you fix it?
Find why the stopping rule did not fire, make the tool's error informative, add a step limit and a visible failure state, and show a regression case for the loop.
Which parts of this task should not be an agent?
Keep steps that never change in ordinary code, name the one decision that needs a model and the evidence that the model improves it, and compare against a fixed workflow.
Other AI engineering roles
12-guide path
Forward Deployed Engineer
Customer discovery, implementation in the customer's environment, and ownership of the outcome.
10-guide path
AI Engineer
Model-powered applications, retrieval, evaluations and operations.
5-guide path
GTM Engineer
Revenue systems, integrations and commercial experiments.