Okta’s Senior posting carries the “#LI-Remote” tag and also asks for “regular presence at customer locations”. Source 1Principal Forward Deployed Engineer - Okta for AI AgentsPublisherOkta (Greenhouse)Source typecompany job postingSource 2Senior Forward Deployed Engineer - Okta for AI AgentsPublisherOkta (Greenhouse)Source typecompany job postingSource 3Senior Forward Deployed Engineer - Okta for AI AgentsPublisherOktaSource typecompany job posting Databricks’ Sr. FDE posting says “FDEs are billable”. Source 4Sr. Forward Deployed Engineer (FDE) - Financial Services (New York City)PublisherDatabricks (careers site / Greenhouse)Source typecompany job posting Neither line is in the title, and each changes the job: how often you are away, and who pays for your hours. Read a posting like a customer’s requirements doc: mark what each line commits the employer to and what it leaves open. By the end you can turn any posting into one page: what the loop is likely to probe, and what to study first.

The posting page

Fill it in before reading forums, so others’ stories don’t set your reading. Seven fields come from earlier lessons: title, team and level from Who hires FDEs; code, after signature and pay from the four tests; and travel from Travel, location and on-site time. This lesson adds six:

New fieldWhat to copy
Customer typeEnterprise accounts, one regulated industry, or developers
Top duty tagsTwo tags, each with the line that earned it
RequirementsEach line binned as a gate, evidence or study
Named stackEvery language, product and protocol, word for word
Likely emphasisWhat the loop may probe, marked as your inference
QuestionsOne for the recruiter, one for the hiring manager

A title suffix can give you the customer type outright, as in OpenAI’s Healthcare FDE postings. Source 5Forward Deployed Engineer (FDE), Healthcare - SF (Ashby)PublisherOpenAI (Ashby job board)Source typecompany job postingSource 6Forward Deployed Engineer (FDE), Healthcare - NYC (Ashby)PublisherOpenAI (Ashby job board)Source typecompany job posting Otherwise the duties name it.

Some fields sit in the job board’s public data:

GH=https://boards-api.greenhouse.io/v1/boards
curl -s "$GH/okta/jobs/7961356" \
  | jq '{title, loc: .location.name,
         dept: [.departments[].name]}'

ASHBY=https://api.ashbyhq.com/posting-api/job-board
curl -s "$ASHBY/openai?includeCompensation=true" \
  | jq '.jobs[]
      | select(.title | test("Forward Deployed"))
      | {title, workplaceType,
         pay: .compensation.compensationTierSummary,
         parts: [.compensation.summaryComponents[]?.compensationType]}'

The first returns Okta’s department field; the second, OpenAI’s FDE pay and its parts. OpenAI’s range shows bare on the page, but the record tags it Salary, so record it as base. Source 7Forward Deployed Software Engineer - SF (Ashby)PublisherOpenAI (Ashby job board)Source typecompany job postingSource 8Forward Deployed Software Engineer - NYC (Ashby)PublisherOpenAI (Ashby job board)Source typecompany job postingSource 9Forward Deployed Software Engineer - Seattle (Ashby)PublisherOpenAI (Ashby job board)Source typecompany job posting

Duties that reveal the real job

Tag each duty line; each tag asks you to show something:

  • Build: a system you shipped and can draw.
  • Scope: a vague request you turned into a plan.
  • Operate: an incident you owned after launch.
  • Sell: a demo that moved a decision.
  • Teach: a session you ran for a customer’s engineers.
  • Product: a product change one customer caused.

Baseten’s FDE posting mixes hands-on coding with parts of product management, technical customer success and pre-sales solution engineering: Build, Product and Sell. Source 10Forward Deployed Engineer @ BasetenPublisherBaseten (Ashby)Source typecompany job posting Cognition’s Applied AI Engineer posting has you lead live workshops and pair-programming sessions for enterprise engineering teams: Teach. Source 11Applied AI Engineer @ CognitionPublisherCognition (Ashby job board)Source typecompany job posting

Then rank, don’t count. A line is specific when it names something you could be asked to draw, a measure, or who you answer to. Databricks’ “exceptional customer empathy” names none of these; OpenAI’s “owning integrations, data flows, reliability, observability, and on-call readiness” names systems to draw and a pager to carry. Source 4Sr. Forward Deployed Engineer (FDE) - Financial Services (New York City)PublisherDatabricks (careers site / Greenhouse)Source typecompany job postingSource 12Forward Deployed Engineer (FDE), Financial Services- NYCPublisherOpenAI (Ashby)Source typecompany job posting Tag the two most specific lines. Test your ranking with the recruiter: “Which two of these responsibilities matter most for the person you hire, and which rounds should I prepare them for?”

Requirements you can bridge and ones you cannot

Note each line’s header first: we read “requirements” as the filter, and “preferred” or “you’ll thrive if” as a wish list that needs one prepared sentence.

Gates. No preparation closes them. Salesforce’s Missionforce Associate FDE posting requires an active TS/SCI clearance with polygraph and US citizenship. Source 13Missionforce - Associate Forward Deployed Engineer (FDE), Public SectorPublisherSalesforce (Workday careers site)Source typecompany job posting An on-site city you cannot move to is a gate too. Years sit softer than they look: OpenAI’s Forward Deployed Software Engineer postings put 7+ years of full-stack experience under “You’ll thrive in this role if you”. Source 7Forward Deployed Software Engineer - SF (Ashby)PublisherOpenAI (Ashby job board)Source typecompany job postingSource 8Forward Deployed Software Engineer - NYC (Ashby)PublisherOpenAI (Ashby job board)Source typecompany job posting A years line is a gate only if you are clearly short. Otherwise bin it as evidence, and bring the story whose scope matches the level.

Evidence lines. Only a true story bridges these. Glean’s FDE posting wants shipped AI work, “Not just prototypes.” Source 14Founding Forward Deployed EngineerPublisherGlean (Greenhouse job board)Source typecompany job posting Prepare one production system in depth: who used it, what broke, what you changed. The hardest technical problem question tests that story.

Study lines. Study closes these, and some postings set the depth: Modal’s ML FDE postings say you should be able to go deep on at least one serving or training toolchain, and Baseten’s Training FDE posting says the strongest candidates may “spike in one or two” of its listed strengths. Source 15Forward Deployed Engineer - ML @ ModalPublisherModal (Ashby)Source typecompany job postingSource 16Forward Deployed Engineer - ML @ Modal (Stockholm)PublisherModal (Ashby)Source typecompany job postingSource 17Forward Deployed Engineer (Training) @ BasetenPublisherBaseten (Ashby)Source typecompany job posting Study until you can draw each line and name one way it breaks at a customer; where that would take months, say where your depth stops.

One posting, filled in: Okta’s Senior FDE

A second posting, filled in: Databricks’ Sr. FDE

Take Databricks’ Sr. FDE - Financial Services posting in New York. Source 4Sr. Forward Deployed Engineer (FDE) - Financial Services (New York City)PublisherDatabricks (careers site / Greenhouse)Source typecompany job posting The Databricks guide covers its other FDE postings.

Two AI-lab lines, read in brief

OpenAI’s Financial Services FDE posting in New York states Operate outright: it holds the on-call line quoted above, and asks you to lead discovery and scoping from pre-sales through post-production. Source 12Forward Deployed Engineer (FDE), Financial Services- NYCPublisherOpenAI (Ashby)Source typecompany job posting Tag it Scope and Operate: bring a vague problem you framed, and a live system you kept healthy. The OpenAI guide covers its other FDE postings.

Anthropic’s FDE postings name artifacts to deliver: “MCP servers, sub-agents, and agent skills”. Source 24Forward Deployed Engineer (New York City, NY; San Francisco, CA; Seattle, WA)PublisherAnthropic (Greenhouse job board)Source typecompany job posting Build one before the onsite: a small MCP server with one tool and auth, plus a sentence on how you would check the agent used it correctly.

Turning the posting into prep priorities

Study first whatever your posting names: a protocol, a platform, an artifact, a language. “Strong communicator” fits any posting; a named protocol was chosen for this one.

Before studying, read the company’s other FDE postings beside yours: shared lines describe the unit, changed lines your role. All three Okta FDE postings share one identity-protocol line, so it belongs to the unit, whatever the level. Source 1Principal Forward Deployed Engineer - Okta for AI AgentsPublisherOkta (Greenhouse)Source typecompany job postingSource 2Senior Forward Deployed Engineer - Okta for AI AgentsPublisherOkta (Greenhouse)Source typecompany job postingSource 19Principal Forward Deployed Engineer (Singapore)PublisherOkta (Greenhouse)Source typecompany job posting The pay line changes: as of September 2026, the Senior posting states a base range and the US Principal posting an OTE range, base plus incentive compensation. Source 1Principal Forward Deployed Engineer - Okta for AI AgentsPublisherOkta (Greenhouse)Source typecompany job postingSource 2Senior Forward Deployed Engineer - Okta for AI AgentsPublisherOkta (Greenhouse)Source typecompany job posting

Identity lines

That shared line lists 2.0, , , , RFC 8693 token exchange, act claims, CIMD and DCR, and DPoP. Source 1Principal Forward Deployed Engineer - Okta for AI AgentsPublisherOkta (Greenhouse)Source typecompany job postingSource 2Senior Forward Deployed Engineer - Okta for AI AgentsPublisherOkta (Greenhouse)Source typecompany job postingSource 19Principal Forward Deployed Engineer (Singapore)PublisherOkta (Greenhouse)Source typecompany job posting Token exchange is RFC 8693, whose section 4.1 defines the act claim; DPoP is RFC 9449; and CIMD, the Client ID Metadata Document, is still an IETF Internet-Draft. Draw an agent acting for a user; it asks for a token exchange:

POST /token   (form fields)
grant_type=urn:ietf:params:oauth:grant-type:token-exchange
subject_token=<user's access token>
subject_token_type=urn:ietf:params:oauth:token-type:access_token
actor_token=<agent's token>
actor_token_type=urn:ietf:params:oauth:token-type:access_token
audience=https://api.customer.example

The access token it gets back, decoded, names both parties:

{
  "sub": "user-123",
  "act": {
    "sub": "agent-7"
  },
  "aud": "https://api.customer.example"
}

The API sees who the request is for (sub) and who is making it (act). Say what it logs and refuses in one breath: “The API logs agent-7 acting for user-123 on every call, and refuses a token whose aud is not this API, a scope the user never delegated, an act claim on an action no agent may take, such as changing an MFA factor, and, with DPoP, a proof whose key does not match the token’s cnf.jkt.”

Measurement lines

Snowflake’s Senior/Staff Applied AI FDE posting asks for quality metrics, evaluation frameworks and golden datasets built from customer goals. Source 25Senior/Staff Forward Deployed Engineer, Applied AI @ SnowflakePublisherSnowflake (Ashby job board)Source typecompany job posting OpenAI’s Healthcare FDE posting asks for that measure quality against customer-specific acceptance thresholds. Source 5Forward Deployed Engineer (FDE), Healthcare - SF (Ashby)PublisherOpenAI (Ashby job board)Source typecompany job posting For either, prepare a page like this, for a fictional refund agent:

goal: fewer refunds issued in error
golden_set: past tickets, labeled by
  the customer's support leads
metric: wrong-refund rate on the
  golden set
launch_bar: agreed with the customer
  before the build
watch: weekly sample of live refunds,
  reviewed by a support lead

Then say: “The bar was the customer’s, set before we built, and here is what I watch after launch.” Practice it on designing an evaluation harness.

Data-platform lines

For a posting like Databricks’, prepare how a shuffle moves data between partitions and why one skewed key leaves a single straggler task holding up a stage. Find the hot key first:

from pyspark.sql import functions as F

(events.groupBy("account_id").count()
    .orderBy(F.desc("count"))
    .show(5))

spark.conf.get("spark.sql.adaptive.enabled")
spark.conf.get("spark.sql.adaptive.skewJoin.enabled")

Then say: “If the hot key hits a sort-merge join, adaptive query execution splits the skewed partition when spark.sql.adaptive.enabled and spark.sql.adaptive.skewJoin.enabled are both on. A count or sum by key is partly computed inside each task before the shuffle, so its skew costs little. If skew survives, I salt the key: a random suffix on the big side, and the matching rows copied on the small side.”

For each other named tool, have one line ready: “I used X for Y in production; at a customer, the first thing I’d check is Z.”

Keep one page per target and reread it before each round

Each field is one line you can reread in the lobby. A filled example for a fictional employer:

company: Examplecorp
title: Forward Deployed Engineer,
  Healthcare
team: Applied AI
level: not stated
read_on: September 2026
code: production apps inside
  customer systems
after_signature: owns launch criteria
  and monitoring
pay:
  figure: not stated
  basis: unstated
travel:
  - "'Travel may be required.'"
  - no unit, no ceiling
  - customer sites
customer_type: hospital systems;
  clinical and IT buyers
top_tags:
  build: ship production apps
    inside customer systems
  operate: own launch criteria
    and monitoring
requirements:
  gates:
    - "hybrid, 3 office days: met"
  evidence:
    - line: production AI in a
        regulated setting
      story: discharge-summary
        drafting, audited by nurses
  study:
    - FHIR
    - evals
    - OIDC
named_stack:
  - Python
  - TypeScript
  - FHIR
likely_emphasis: build and eval
  design (our inference)
questions:
  recruiter:
    - Which two duties matter most,
      and in which rounds?
    - What is the range, and is it
      base or OTE?
  hiring_manager:
    - Who carries the pager once a
      customer deployment is live?

Reread it before each round: pay and gates before the recruiter, study lines before technical rounds, duties before the hiring manager. A line that commits you without saying how becomes a hiring-manager question, like the pager one above.

Some applications ask about these rows: Palantir’s US Government FDSE application form asks whether you support the advertised travel percentage and are comfortable with the advertised pay range. Source 26Apply: Forward Deployed Software Engineer - US Government (Washington, D.C.) - Palantir on LeverPublisherPalantir Technologies (Lever)Source typecompany job posting Write your Travel and Pay rows as a yes with a condition, not a shrug; the Palantir guide covers that form.

Practice your two top tags next: for Build, the SSO design question; for Operate, the production incident you led; for Scope, your first week with a vague problem; for Product, a custom build or a product change; and for a data-platform posting, the case Moving analytics to a new data platform.

Now annotate one posting, then answer What customer-facing experience do you have? out loud for its customer type.

Keep going. That question is free with an illustrative answer, the Build question with a model answer, and all 180 practice questions show their framework free. Pro opens the other stories’ answers, and the case with its scorecard.

GlossaryForward deployed engineerA software engineer who builds and ships production systems inside a customer’s problem and environment, accountable to that customer’s outcome.More on Forward deployed engineerGlossaryForward deployed software engineerPalantir’s title for its FDE role, called Delta internally; OpenAI and EY also post FDSE titles, each with its own duties.More on Forward deployed software engineerGlossaryAgentA system in which a model chooses steps and tool calls to complete a task, within limits the design sets.More on AgentGlossaryModel Context ProtocolAn open protocol for exposing tools and data to models through servers, so agents can call a customer’s systems in a standard way.More on Model Context ProtocolGlossaryOn-target earningsBase pay plus variable pay if targets are met; a posting that states pay as OTE includes incentive pay in the figure.More on On-target earningsGlossaryOAuthA standard for granting an application limited, revocable access to resources on a user’s behalf without sharing passwords.More on OAuthGlossaryOpenID ConnectAn identity layer on top of OAuth that lets an application verify who a user is and get basic profile claims.More on OpenID ConnectGlossarySAMLAn XML-based standard for exchanging authentication assertions between an identity provider and an application.More on SAMLGlossarySCIMA standard protocol for provisioning and deprovisioning users and groups from an identity provider into applications.More on SCIMGlossaryLaunch criteriaThresholds agreed with a customer before building that decide whether a system goes live.More on Launch criteria