By FDE Interview Prep, a SourceStrongAI company

In this post11 sections
  1. What you already bring from solutions architecture
  2. The gap: owning production code in the customer’s systems
  3. The coding bar, in employers’ own words
  4. A plan to close the gap
  5. Rewriting architect stories as delivery stories
  6. Answering ‘why leave solutions architecture?’
  7. Roles in between: applied AI and forward deployed architects
  8. Where to start this week
  9. Questions people ask
  10. Keep reading
  11. More from the blog

You already run the discovery call, draw the target architecture and build the demo that wins the room. The loop asks something your job rarely makes you prove: that you can write production code inside a customer’s systems and own it after it ships. The move from solutions architect to forward deployed engineer is realistic, and your customer skills are a real head start. The gap is code you owned, and our plan closes it in three parts: one production-grade build, timed coding and debugging practice, and interview stories in which the shipped code was yours. For the full side-by-side of the two roles, start with the FDE vs solutions architect guide. It applies whether you’re aiming at an AI lab or at an enterprise software company.

What you already bring from solutions architecture

Some FDE postings name your background outright: Adobe’s Experience Platform FDE postings list solution architecture among the typical backgrounds, next to software engineering and technical consulting. Source 1Senior Forward Deployed Engineer (Bucharest)PublisherAdobe (Workday careers site)Source typecompany job posting

Most of an FDE posting will sound familiar. OpenAI’s Forward Deployed Engineer posting says FDEs own discovery, technical scoping, system design, build and production rollout with customers, and asks for people who write and review production-grade code. Source 2Forward Deployed Engineer (FDE) - SFPublisherOpenAI (Ashby)Source typecompany job posting You do the first three of those five things every week.

That matters, because at least one hiring lead says the business half is the easier one to learn on the job, and you already have it. Jia Wu, a deployed engineering lead at Cognition, said in a talk at the AI Engineer World’s Fair 2026, posted in July 2026, that business sense can be learned, but being the technical expert in the room is very hard to teach. Source 3How Forward Deployed Engineering is done at Cognition (Jia Wu; AI Engineer World's Fair 2026)PublisherAI EngineerSource typerecorded talk or interview

Kevin Bai, of Anthropic’s applied AI team, gave his general view of the role in a talk published by AI Engineer in July 2026: an FDE is someone you would hire as a software engineer and would also trust in front of a customer. Source 4Forward Deployed Engineering 101 (Kevin Bai, Member of Technical Staff, Anthropic; AI Engineer World's Fair 2026)PublisherAI EngineerSource typerecorded talk or interview As an SA, you have been proving the second half for years. The work is the first half. The lesson on the double bar helps you find out, with evidence, which side you are weaker on.

What transfers directly:

  • Discovery: finding the real problem behind the stated one.
  • Scoping: saying no to the part of the request that won’t fit.
  • Stakeholders: a skeptical CISO and an impatient VP in the same week.
  • Architecture judgment: which integration patterns survive a security review.

The gap: owning production code in the customer’s systems

The Pragmatic Engineer reported, in August 2025, that at OpenAI the line is who writes the code: FDEs write it directly on customer infrastructure with customer tooling, while solutions architects are more advisory, rarely write code on customers’ infrastructure and usually build proofs of concept on anonymized or offline cuts of data. Source 5What are Forward Deployed Engineers, and why are they so in demand? (Gergely Orosz)PublisherThe Pragmatic EngineerSource typenews report

Baseten’s Solutions Architect posting puts the gap in one employer’s own words: it asks for technical depth “without needing to write production code”, and says the SA pulls in FDE support for deeper technical work. Source 6Solutions Architect @ BasetenPublisherBaseten (Ashby)Source typecompany job posting

Employers draw the same line in their postings:

  • Google Cloud’s GenAI FDE postings set the role against “traditional advisory roles” and describe someone who codes, debugs and jointly ships agentic solutions inside the customer’s environment. Source 7Forward Deployed Engineer IV, GenAI, Google Cloud — Google CareersPublisherGoogleSource typecompany job posting
  • Cursor’s FDE postings say FDEs embed with customer engineering teams to ship production-grade workflows, and put it bluntly: “This is not a demo role.” Source 8Forward Deployed EngineerPublisherCursor (Ashby job board)Source typecompany job board
  • Anthropic’s Forward Deployed Engineer posting asks FDEs to work within customer systems to build production applications with Claude models. Source 9Job Application for Forward Deployed Engineer at AnthropicPublisherAnthropic (Greenhouse)Source typecompany job posting
  • OpenAI’s Financial Services FDE posting goes further and asks the FDE to own on-call readiness for the production AI systems it ships. Source 10Forward Deployed Engineer (FDE), Financial Services- NYCPublisherOpenAI (Ashby)Source typecompany job posting

How the habits differ, in our words:

SA habitWhat the FDE role adds
Demo on clean dataRuns on the live feed
Hand off to their teamYou build and roll out
Sample codeTested, logged, retried
Done at the winDone when it works in production

A small example: the offline cut versus the live feed

Say you built a pilot report on an export the customer’s analyst cleaned for you. This is proof-of-concept code, and on the pilot data it works:

from decimal import Decimal

def total_by_region(rows):
    totals = {}
    for r in rows:
        region = r["region"]
        amount = Decimal(r["amount"])
        prev = totals.get(region, 0)
        totals[region] = prev + amount
    return totals

Now point it at the live feed. One row says "1,200.00", one has an empty amount, one has "West " with a trailing space, and one has no region at all. We ran it: the comma row and the empty amount raise InvalidOperation, the row with no region raises KeyError, and any one of them kills the whole job. The trailing space fails quietly: it files that row under a second region called "West ".

The owned version expects the mess and reports it:

from decimal import Decimal, InvalidOperation

def parse_amount(s):
    s = (s or "").replace(",", "")
    try:
        d = Decimal(s)
    except InvalidOperation:
        return None
    return d if d.is_finite() else None

def total_by_region(rows):
    totals, rejects = {}, []
    for n, r in enumerate(rows, start=1):
        region = r.get("region") or ""
        region = region.strip()
        amount = parse_amount(r.get("amount"))
        if not region or amount is None:
            rejects.append(n)
            continue
        prev = totals.get(region, 0)
        totals[region] = prev + amount
    return totals, rejects

On the same live rows it returns West’s total, with the trailing space trimmed and the comma removed, plus the row numbers it could not read. The is_finite() check matters: Python’s Decimal accepts "NaN" and "Infinity", and one "NaN" row would silently turn the whole total into NaN. It also assumes commas are thousands separators. That is a question for the customer, not a guess: which locale does the feed use? What changed is the question behind it: what happens on the day the data is wrong, and who finds out?

That question is also the one to say out loud in the room: “Should a bad row stop the run, or should we finish and send the rejects to someone? Who is that someone?” An SA asks it in a design review. An FDE writes the answer into the code.

The coding bar, in employers’ own words

Some employers say plainly that their FDEs clear the engineering bar:

Each is about one employer, not a rule for every loop. Candidate reports add texture, each from one person:

  • One poster on Blind reported, in April 2024, that a CodeSignal screen for Anthropic’s Solutions Architect/Forward Deployed Engineer role had LeetCode-style questions, “maybe leetcode easy level”, and that they could not finish the second in time. Source 13Anthropic Solutions Architect/Forward Deployed Engineer CodesignalPublisherBlind (teamblind.com)Source typecandidate report on Blind
  • One poster on Blind reported, in July 2026, that the recruiter on their Databricks Resident Solutions Architect process said the RSA roles were being converted into FDE roles, and asked them to go through more interview rounds. Source 14Databricks Interview ExperiencePublisherBlind (teamblind.com)Source typecandidate report on Blind

The second report is the move in miniature: same company, same candidate, more rounds once the title changed.

No employer publishes how its rounds are weighted, and no one publishes a pass rate. Plan for coding in some shape: an algorithm problem, a practical build, someone else’s code to debug, or an AI-assisted build. The lesson on what coding rounds test sorts the evidence by company so you can practice the shape your targets use.

A plan to close the gap

This plan is ours. It is how we would prepare an SA for the loop, not what any interviewer uses to score you.

Part one: a production-grade build

Build one small thing a real user would run, and take it all the way. The target is not a clever model but the parts a demo leaves to the customer’s team:

What makes a build production-grade

  • A named user and a messy input you did not clean by hand
  • A walking skeleton deployed end to end before any feature work
  • Tests, including one for a failure you hit and fixed
  • Retries and timeouts on every external call
  • Logs that tell you which record failed and why
  • A README that says what you left out, and why

The portfolio project post covers choosing the project and presenting it.

Part two: timed coding practice

If most of your code has lived in notebooks and demo repos, written alone with no clock, the round will feel different: a timer, someone watching, and a prompt you have not seen. Practice in those conditions.

Start with the practical builds closest to the job: parsing messy files, and calling an API that pages, throttles and fails. The free messy CSV question is the live-feed example above at interview size. The flaky API integration question, in Pro, covers the API half, and the API integration coding round post works it in Python: retries with backoff, cursor pagination and idempotency keys.

If your targets report algorithm screens, add easy and medium problems too. Don’t skip them because the job “isn’t LeetCode”: the older Blind report above, from April 2024, came from a screen for a role with “Solutions Architect” in its title. Source 13Anthropic Solutions Architect/Forward Deployed Engineer CodesignalPublisherBlind (teamblind.com)Source typecandidate report on Blind The coding rounds lesson linked above has more current evidence by company.

Part three: debugging someone else’s code

In our view this is the most FDE-shaped skill, and the one a demo-building job practices least. On the job, the bug is in code you didn’t write, on a system you can’t restart, with a customer waiting.

Practice the method, not the answer:

  1. Reproduce the failure before you touch anything.
  2. Write a failing test that captures it.
  3. Form one hypothesis at a time, and say it out loud.
  4. Fix the smallest thing, then rerun the test.
  5. Say what would stop it happening again.

Work through the unfamiliar bug question, then write a failing test that reproduces a reported bug. Both reward the narration an SA is already good at, pointed at code.

Rewriting architect stories as delivery stories

Your stories are the other half of the gap. An SA’s best stories often end at the handoff, and a behavioral round will find that edge.

Leo Mehr said on a podcast published in July 2026 that he assesses ownership by asking candidates about the decisions they made and what they were responsible for. Source 15Leo Mehr - Ramp's $44B Bet on Services (YouTube auto-generated English captions)PublisherBasil Chatha, YouTubeSource typerecorded talk or interview If an interviewer probes the way he does, expect the question on every project you name: which part was yours?

Here is a fictional before and after.

Before: “I designed the reference architecture for the claims pipeline, ran the workshops, and the customer’s team built it on schedule.”

The follow-up writes itself: “What code did you write?” If none, the story shows an advisor.

After: “The customer’s team was short-staffed for the pilot, so I wrote the ingestion service myself. I chose a nightly batch over streaming because their source system locked tables during the day. When the first run dropped rows, I found the encoding bug, added a reconciliation check, and handed over a runbook. Their team took it over after go-live.”

The second version has decisions, a failure, a fix and a handover, all in the first person. Run an ownership audit over each story you plan to tell: every decision gets a named owner, and anything the customer’s team did is credited to them.

If the audit leaves you with no code you owned, the fix is a different story, not better pronouns. Two ways to get one:

  • At work, before you apply: volunteer to own one integration on a pilot through go-live, including the part after the demo.
  • On your own: the build from part one becomes the story.

Then practice the two questions that test this directly: a prototype that ended up in production, which rewards SAs whose proof of concept outlived its plan, and a project you owned end to end.

Answering ‘why leave solutions architecture?’

Expect some version of this question, from a recruiter or a hiring manager. Weak answers leave the interviewer worried:

  • “I want to be more technical.” It invites “then why not software engineering?”
  • “I’m tired of the sales side.” FDE work is customer-facing too, so the interviewer wonders whether you will tire of that as well.
  • “It’s the hot role right now.” It says nothing about you.

A stronger answer connects your past to the work, with evidence. Here is the shape, with fictional details:

“For several years I designed integrations that customers’ teams then built. On one pilot I ended up writing the ingestion service myself and ran it through go-live, and that was the part of the job I wanted more of: owning what ships, not only recommending it. As an FDE, the customer work and the production code are the same job, and I already have the customer half.”

It is short, has a moment of proof, and doesn’t criticize SA work or sales. The lesson on the bridge story builds this answer from your own history and tests it against the follow-ups.

Prepare for three follow-ups:

  • “Why not a software engineering role?” Your answer is the customer half. The why FDE and not software engineering question has a model answer.
  • “What is the largest thing you shipped yourself?” Name it, its users and its failure modes.
  • “How do you feel about owning code in someone else’s environment?” Say what you’d ask for on day one: access, logs, a test environment and a named contact on their side.

Roles in between: applied AI and forward deployed architects

If the jump feels too far today, some roles sit in between, and their postings ask for code.

  • Applied AI architect. Anthropic’s Applied AI Architect posting calls the role a pre-sales architect, pairs it with account executives, and still asks it to ship working code as prototypes, proofs of concept and near-production examples. Source 16Job Application for Applied AI Architect, Commercial at AnthropicPublisherAnthropic (Greenhouse)Source typecompany job posting That makes it a way to build the evidence an FDE loop asks for.
  • Forward deployed architect. Notion’s Forward Deployed Architect postings ask for 5+ years in tech services or consulting, solutions architecture, implementation, customer engineering or a similar customer-facing role. Source 17Forward Deployed Architect @ NotionPublisherNotion (Ashby job board)Source typecompany job posting Solutions architecture is on that list by name.

Whichever title you aim at, ask the recruiter early how the role is leveled.

Titles blur in both directions, so read the posting, not the title. The lesson on telling FDE, SA, sales engineer and SWE roles apart shows the lines that decide it, and the post on spotting an SA role in disguise applies them to FDE postings.

Where to start this week

Do one move from each part before you apply anywhere:

This week

  • Choose the one build, and deploy its
  • Solve one practical coding question on a timer, talking out loud
  • Debug one piece of unfamiliar code, test first
  • Run the on your two best stories
  • Say your “why leave SA” answer aloud, and time it

Start tonight with the messy CSV question, which is free: set a timer and talk out loud. Then check your answer against the model answer. The question bank holds 180 interview questions with model answers, including the coding, debugging and ownership questions linked above.

When you want the customer half tested too, the free practice case puts you with an AI customer who needs something built and hasn’t told you everything yet (free, with a sign-in).

The full question bank and the rest of the lessons come with Pro, which starts with a 7-day free trial. The plans are on the pricing page.

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 engineerGlossarySolutions architectA customer-facing technical role that designs, advises and often builds proofs of concept; postings place it before the sale, after it, or both.More on Solutions architectGlossaryWalking skeletonThe thinnest end-to-end version of a system that performs one small real function across its main components, built first and then extended.More on Walking skeletonGlossaryOwnership auditRewriting a story so every decision has a named owner: what you decided and did, and what others did, credited to them.More on Ownership audit

Questions people ask

Can a solutions architect become a forward deployed engineer?

Yes, and some FDE postings name it: Adobe’s Experience Platform FDE postings list solution architecture among the typical backgrounds. The harder part is the coding bar. Ramp’s Director of Engineering has said FDE candidates there must pass the software engineering interview loop.Source 1Senior Forward Deployed Engineer (Bucharest)PublisherAdobe (Workday careers site)Source typecompany job postingSource 11What is a Forward Deployed Engineer? (FDE Explained) feat. Leo Mehr of Ramp (YouTube auto-generated English captions)PublisherdearCC (Clara Shih), YouTubeSource typerecorded talk or interview

What is the main difference between a solutions architect and an FDE?

Who writes and owns the production code. Google Cloud’s GenAI FDE postings contrast the role with traditional advisory roles and describe someone who codes, debugs and jointly ships solutions inside the customer’s environment. Cursor’s FDE postings put it more bluntly: “This is not a demo role.”Source 7Forward Deployed Engineer IV, GenAI, Google Cloud — Google CareersPublisherGoogleSource typecompany job postingSource 8Forward Deployed EngineerPublisherCursor (Ashby job board)Source typecompany job board

Should I apply for applied AI architect roles first?

They can be a bridge. Anthropic’s Applied AI Architect posting calls the role a pre-sales architect and still asks it to ship working code as prototypes and near-production examples. Aim at FDE roles once you can show production code you owned.Source 16Job Application for Applied AI Architect, Commercial at AnthropicPublisherAnthropic (Greenhouse)Source typecompany job posting

Keep reading

Guides

Lessons

Questions

More in The FDE role