Sierra’s AI-native interview: plan, build, review, and what Sierra says it looks for

By FDE Interview Prep, a SourceStrongAI company

In this post11 sections

1. What Sierra changed, in its own words
2. Where the agent engineer fits: an FDE role under another title
3. Plan: you drive the ideation
4. Build: the interviewer leaves and you build with AI
5. Review: the demo, the code and how you used AI
6. What Sierra says it looks for
7. What candidates report, and what is unknown
8. A practice routine for a plan, build and review session
9. Questions people ask
10. Keep reading
11. More from the blog

You have a Sierra onsite on the calendar, or you are deciding whether to apply, and you have heard that Sierra is the company that dropped LeetCode. That is close to what Sierra itself says: in an April 2026 post titled “The AI-native interview”, by Vijay Iyengar, Arya Asemanfar and Angie Wang, Sierra says it removed coding and algorithms interviews from its engineering process and replaced them with an AI-native onsite in three parts: Plan, Build and Review. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog In the build, the interviewer steps out and you have 2 hours with the AI tooling and frameworks of your choice; in the review, you demo, walk through the code and explain how you used AI. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog For every other round and company, start with the forward deployed engineer interview guide.

This post covers each stage of the AI-native interview, what Sierra says it looks for, what candidates report, and a routine to practice the whole session at home.

One caution: the post covers Sierra’s engineering interviews in general, not one role’s loop, so confirm yours with your recruiter.

What Sierra changed, in its own words

Sierra says its old engineering process was “fairly standard”: two coding interviews plus interviews for algorithms, system design and culture fit, followed by reference checks. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog

It says the old interviews produced signal mostly about mechanics, and that without clear interview signals, hiring managers leaned more on referrals and prior experience. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog

Here is what changed, stage by stage, as the post describes it. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog

The post does not say whether the culture-fit interview or reference checks remain; ask your recruiter.

On the pilot, Sierra says candidates improve a colleague’s draft PR in a medium-sized codebase using coding agents, and that the level of AI allowed there is “still TBD”, because new models can zero-shot many fixes. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog If your loop includes it, ask what tools you may use.

So your first technical conversation with Sierra is now a design round. Sierra builds customer-service agents, so the agentic system design interview is the closest practice: tools, permissions and hand-offs to a person. For how Sierra compares with employers that still run algorithm rounds, see do FDE interviews have LeetCode?

Where the agent engineer fits: an FDE role under another title

In September 2026, no job title on Sierra’s main Ashby board used “Forward Deployed Engineer”. ^[2]Source 2Software Engineer, AgentPublisherSierra (Ashby)Source typecompany job posting Its customer-facing role for building agents is titled Software Engineer, Agent, in the Engineering department and the Agent Engineering team. ^[2]Source 2Software Engineer, AgentPublisherSierra (Ashby)Source typecompany job posting The words do appear, though: Sierra’s Deployed Infrastructure Engineer posting calls the role a Forward Deployed Infrastructure Engineer, and a Tokyo listing uses that title. ^{[3, 4, 5]}Source 3Deployed Infrastructure Engineer (San Francisco)PublisherSierra (Ashby)Source typecompany job postingSource 4Sierra Japan job boardPublisherSierra (Ashby)Source typecompany job postingSource 5Deployed Infrastructure Engineer (London)PublisherSierra (Ashby)Source typecompany job posting

Our reading: Software Engineer, Agent is an FDE role in all but name. Sierra defines agent engineers as people who work with its customers to design, build and ship agents on its platform, ^[6]Source 6Meet the AI agent engineer (Natalie Meurer)PublisherSierraSource typecompany blog and its own people talk about the role that way. Natalie Meurer, Sierra’s Head of Agent Engineering, said at the AI Engineer World’s Fair in 2026 that she had been hiring for “these deployed roles” for about two to two and a half years. ^[7]Source 7The Dirty Secret of Forward Deployed Engineering (Natalie Meurer, Head of Agent Engineering, Sierra; AI Engineer World's Fair 2026)PublisherAI EngineerSource typerecorded talk or interview In the same talk, she said the one constant across the many versions of the FDE role is that every forward-deployed engineer is accountable to the customer. ^[7]Source 7The Dirty Secret of Forward Deployed Engineering (Natalie Meurer, Head of Agent Engineering, Sierra; AI Engineer World's Fair 2026)PublisherAI EngineerSource typerecorded talk or interview

In a July 2024 Sierra post, she wrote that engineering expertise, customer obsession and business curiosity define some of the most successful agent engineers. ^[6]Source 6Meet the AI agent engineer (Natalie Meurer)PublisherSierraSource typecompany blog

Search only “forward deployed engineer” and you will miss Sierra; search “Software Engineer, Agent” too. Sierra’s careers page also says interviews take place at one of its offices because the company works in person, and that the process evaluates both capabilities and alignment with its values. ^[8]Source 8Careers | SierraPublisherSierraSource typecompany hiring page Plan to be in the room.

Plan: you drive the ideation

Sierra describes Plan as a working session to define a product to build: the candidate drives ideation, and interviewers ask questions to strengthen it. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog

You propose the spec, out loud, while people push on it. That is where your scope gets decided, and scope is the judgment Sierra says it looks for (more on that below). Your scope also decides whether the build finishes.

What to say in the first few minutes

This structure is ours, not Sierra’s. It gets a scoped plan on the table fast.

1. The user and the moment. “This is for a support lead at the end of a shift, trying to see which conversations went wrong.”
2. The one flow you will build end to end. “The flow I’ll build is: pick a conversation, see why it escalated, flag it for review.”
3. What you will skip, and why. “I’ll skip login and account management, since they don’t show anything about the problem.”
4. How you will know it works. “If I can take a real-looking transcript and get the right reason back, the core works.”
5. Your fallback. “If the classification part fights me, I’ll hard-code the reasons and keep the flow working.”

Sierra’s post says it shares evaluation criteria and advice before the onsite, and gives two examples: it is fine to cut scope as you build, and to skip boilerplate such as CRUD and auth to focus on what is unique. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog The skip line above is you taking that advice in public. If you have not seen the criteria, ask your recruiter.

When an interviewer pushes

Say an interviewer asks, “What about a support lead who manages several teams?” A weak answer adds a feature. A strong one decides:

“Good case. It changes the data model, since a conversation now belongs to a team. I’ll add a team field so the model is right, but the UI stays single-team for today. That keeps the build inside the time.”

The product-sense prompts lesson drills this move: users, flows, data and one metric. For the spoken version, practice coding with vague requirements.

Common mistakes in the plan

• Pitching a platform. “An AI ops suite for support teams” cannot be built and demoed in one sitting.
• Letting the interviewer write the spec. Their questions strengthen your idea; they don’t replace it.
• No fallback. When a hard part stalls, you want a decision already made.

Build: the interviewer leaves and you build with AI

Sierra’s post says the interviewer steps out and the candidate brings the idea to life over 2 hours, using the AI tooling and frameworks of their choice. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog

The interviewer is out of the room, so your process only shows up in the review if you leave a record of it. And “of their choice” means the choice is part of your preparation.

Before the day

• Pick your stack and your AI tool now. Use a framework you have shipped with, and have a blank project that runs, with a place for seed data.

During the build

This is our method for a timed build with an AI assistant, not a rubric Sierra publishes. Here is how we would spend the clock on the support-lead idea from the plan.

Our two-hour budget, not Sierra’s

• Get the thinnest end-to-end version working first. One screen, one call, real-looking data. The walking skeleton lesson shows how to cut to it.
• Keep a decision log. A plain NOTES.md with one line per decision: what you cut, what you chose over what, and where the AI got something wrong and how you caught it. It becomes your review script. A few lines is enough:

Cut: login, multi-team UI (team field kept in model)
Chose SQLite over Postgres: zero setup
AI bug: retry loop swallowed errors; caught when
  a bad record vanished; added a test
1:02 pivot: hard-coded reasons, classifier later

• Check at the halfway mark. If the core flow does not work end to end, cut until it does. A pivot you logged is the agency Sierra says it looks for.
• Read what the assistant writes. The review looks at your code for technical judgment, with data model, abstractions and extensibility as the examples. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog Code you cannot explain is a liability there.

For narrating AI use with an interviewer in the room, see AI-assisted coding interviews.

Review: the demo, the code and how you used AI

Sierra’s post says the candidate demos what they built, then the team debates the key product flows and choices, reviews the code for technical judgment, discusses the path to production, and digs into how the candidate used AI. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog

A way to open each part, in our words:

Words for the hardest part: how you used AI

A specific answer sounds like this:

“I used the assistant for scaffolding and the parsing code. I wrote the data model myself, because everything else depends on it. It generated a retry loop that swallowed errors, and I caught it when a bad record vanished from the list instead of showing a failure. I rewrote that part and added a test for it.”

It shows delegation, verification and a real catch. Practice the general version with how you use AI tools.

When someone questions a choice, name the trade-off and what would change your mind: “I stored the reason as free text to move fast. If leads need to filter by reason, I’d make it a fixed list.” Defend a design choice drills this.

For the path to production, rank the gaps rather than listing them. For the support-lead app:

“First, the classifier is wrong on transcripts it has never seen, so I’d build an eval set of fifty labeled escalations before launch. Second, there’s no auth, so real transcripts can’t go in. Third, there’s no monitoring of flag rate, so I’d alert if it drops to zero.”

The order is the point: risk to the customer comes first.

What Sierra says it looks for

• Agency and judgment. It says the format makes it easier to gauge agency (do they pivot when they get stuck?) and judgment (how do they scope what to build within the time?). ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog
• Initiative, ownership, judgment, system understanding and product thinking. It says the process aims to capture these. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog
• Strengths, not only a lack of weakness. It says Sierra is hiring for strengths, and that its debriefs have shifted from “should we hire this person?” to “where would this person thrive, and how do we support them?” ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog
• Criteria that don’t depend on what you build. It says the evaluation criteria are agnostic to what the candidate builds, and that interviews run in pairs to improve calibration. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog

So pick an idea you can scope well, not the most impressive-sounding one. And decide which strength you want the debrief to remember, then make the plan, the build log and the review all show it. What hiring leaders say they look for maps each trait to the round where you show it.

What candidates report, and what is unknown

We found a few Blind posts and one prep-blog write-up about Sierra’s process, and none of them clearly describes the AI-native onsite.

• An FDE process “very similar to swe”. A Blind poster wrote in April 2026 that they had done Sierra and Harvey FDE processes, and that the “process was very similar to swe if not same, maybe some more practical stuff like parsing jsons.” ^[9]Source 9Anthropic Applied AI / FDEPublisherBlind (teamblind.com)Source typecandidate report on Blind The poster does not say which stages they sat.
• Coding interviews, role not named. Three posters on Blind wrote about Sierra coding interviews in 2026 without naming the role. ^{[10, 11, 12]}Source 10Sierra AI interview process?PublisherBlind (teamblind.com)Source typecandidate report on BlindSource 11Weird Sierra interview experiencePublisherBlind (teamblind.com)Source typecandidate report on BlindSource 12Got a coding interview in an hour for SierraPublisherBlind (teamblind.com)Source typecandidate report on Blind Two wrote before Sierra’s post: one poster wrote in January 2026 about a screening that meant calling a mock API and building a linked list, and another wrote in March 2026 about a 45-minute phone screen with 30 minutes of coding. ^{[10, 11, 12]}Source 10Sierra AI interview process?PublisherBlind (teamblind.com)Source typecandidate report on BlindSource 11Weird Sierra interview experiencePublisherBlind (teamblind.com)Source typecandidate report on BlindSource 12Got a coding interview in an hour for SierraPublisherBlind (teamblind.com)Source typecandidate report on Blind A third poster wrote in August 2026 about a coding interview, stage not stated, whose first part meant reading existing code quickly: “imagine leetcode, but in a real codebase.” ^{[10, 11, 12]}Source 10Sierra AI interview process?PublisherBlind (teamblind.com)Source typecandidate report on BlindSource 11Weird Sierra interview experiencePublisherBlind (teamblind.com)Source typecandidate report on BlindSource 12Got a coding interview in an hour for SierraPublisherBlind (teamblind.com)Source typecandidate report on Blind Of the three, it is the only one dated after Sierra’s April post, and it still involved coding, so be ready to read an unfamiliar codebase fast. Practice that with working in an existing codebase with AI.
• A prep-product write-up. An anonymous March 2026 write-up on gaijineer.co, a blog that sells interview prep, describes a process for Software Engineer, Agent with a debugging round on a 4-5 file agent codebase and an agent-building take-home presented in a 60-minute session. ^[13]Source 13Sierra Software Engineer, Agent Interview ExperiencePublishergaijineer.co (a blog that promotes the furustack interview-prep product)Source typeprep-product blog (anonymous, may be compiled) The prep site says the presentation brings many “why” questions, about the architecture and the approach to tool calling. ^[13]Source 13Sierra Software Engineer, Agent Interview ExperiencePublishergaijineer.co (a blog that promotes the furustack interview-prep product)Source typeprep-product blog (anonymous, may be compiled) It predates Sierra’s post and may be compiled: treat it as unverified.

What nobody publishes:

• whether the Software Engineer, Agent loop uses the AI-native onsite;
• what build prompts look like, or whether you pick the idea from scratch;
• how long Plan and Review run;
• how the debugging pilot is scored.

Our lesson on what FDE coding rounds test sorts the other formats you may meet, including reading someone else’s code.

A practice routine for a plan, build and review session

This routine is ours. Run it with a friend, or alone with a recorder.

Pick a prompt you have not seen. Write a few on cards and draw one. Fictional ones that work:

• a front desk that needs to see which voicemails need a callback today;
• a small online shop that wants a returns assistant;
• a support lead who wants to know why chats get escalated.

Plan out loud. Have your friend play the interviewers and push on the idea. Use the opening above, and stop when you have one flow, a skip list and a fallback.

Build for two hours, the length Sierra’s post gives. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog Use your chosen AI tool, follow the budget, keep the decision log, and check at halfway.

Review every part in the table above. Record it and listen back once.

Run it again with a new prompt and fix only the weakest item. The plan stage is the hardest one to practice alone, because it needs someone pushing back on your idea. The free practice case puts you in front of an AI customer with a vague problem on a 10-minute clock, so you practice turning a loose request into a scoped first version, with a score that quotes your own words back to you. It is free with a sign-in: run it once before you draw your first card.

Questions people ask

Does Sierra still ask LeetCode questions, and what have candidates reported?

Sierra said in April 2026 that it removed coding and algorithms interviews from its engineering process and replaced its no-AI coding phone screen with a system design interview. Three posters on Blind, who did not name their role, wrote about Sierra coding interviews in 2026, including one in August 2026 that involved reading existing code.^{[1, 10, 11, 12]}Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blogSource 10Sierra AI interview process?PublisherBlind (teamblind.com)Source typecandidate report on BlindSource 11Weird Sierra interview experiencePublisherBlind (teamblind.com)Source typecandidate report on BlindSource 12Got a coding interview in an hour for SierraPublisherBlind (teamblind.com)Source typecandidate report on Blind

Can you use AI in Sierra’s onsite?

Yes, in the build. Sierra’s post says the candidate builds over 2 hours using the AI tooling and frameworks of their choice, and the review digs into how they used AI along the way.^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog

Is Sierra’s agent engineer the same as a forward deployed engineer?

As of September 2026, no title on Sierra’s main job board uses “Forward Deployed Engineer”, though its Deployed Infrastructure Engineer posting calls the role a Forward Deployed Infrastructure Engineer. Its customer-facing role for building agents is titled Software Engineer, Agent, and Sierra’s head of agent engineering has spoken about hiring for these deployed roles.^{[2, 3, 7]}Source 2Software Engineer, AgentPublisherSierra (Ashby)Source typecompany job postingSource 3Deployed Infrastructure Engineer (San Francisco)PublisherSierra (Ashby)Source typecompany job postingSource 7The Dirty Secret of Forward Deployed Engineering (Natalie Meurer, Head of Agent Engineering, Sierra; AI Engineer World's Fair 2026)PublisherAI EngineerSource typerecorded talk or interview

Keep reading

Guides

• The forward deployed engineer interview, round by round

Lessons

• Product-sense prompts: users, flows, data and one metric
• The walking skeleton: a functioning first version
• What FDE coding rounds test, by the evidence
• What hiring leaders say they look for

Questions

• I will describe a feature vaguely. Ask what you need, then implement it.
• Extend a feature in a Python codebase you have never seen, using an AI assistant, and explain every change it made.
• In your take-home you chose one approach over the obvious alternative. Defend that choice as if I were the customer’s architect.
• How do you use AI coding tools in your daily work, and where do you not trust them?

More from the blog

1. Company loops

   C3 AI’s AI screening interview and the coding rounds candidates describe

   Two candidates report C3 AI’s FDE process opening with an AI interview, then coding and system design. What it asks and how to prepare for each step.

   10 min readPublished September 2026

2. Company loops

   Cognition’s deployed engineer take-home: what candidates built with the Devin API

   What candidates built for the Cognition take-home, the two Devin API calls you need, and how to scope, guard and present the integration.

   10 min readPublished September 2026

3. Company loops

   Databricks AI FDE coding round: preparing for applied data science and machine learning

   What Databricks publishes, what one candidate was told about the AI FDE coding round, and a prep plan: a baseline model, leakage checks and trade-offs.

   10 min readPublished September 2026

More in Company loops, page 2