Your Stuut interview is Friday, Nov. 20, from 11:30 a.m. to 12:15 p.m. Pacific on Zoom. Steve Hu’s email names the role Member of the Technical Staff — Applied AI, Fullstack. The invite confirms the time and Zoom link. The interviewer and format are still unconfirmed. Calendar event · Steve’s email
If you only prep two things tonight, rehearse a five-minute teardown of your strongest production project and walk through the payment-matching case below. Those cover the biggest overlap between the actual job description and what past candidates reported.
What Stuut does and why they’re hiring
Stuut started by helping businesses collect overdue invoices. It’s expanding into the whole order-to-cash process, from validating orders to resolving disputes and matching incoming payments to invoices. The insight behind that expansion is that a late payment often starts upstream with a missing purchase order, a pricing mismatch, or an invoice sent to the wrong person. Stuut raised a $52.5M Series B led by Insight Partners in October and cited Honeywell running Stuut on legacy SAP. It announced an EY US partnership on Nov. 17. Stuut Stuut
I read the full posting for your role and the separate FDE posting. Yours emphasizes building the agents, the infrastructure that controls their actions, and the finance team’s UI. That includes durable state, safe ERP writes, human approvals, audit trails, and evaluations tied to customer outcomes. The posting explicitly welcomes workflow and distributed-systems experience and says you can learn Pydantic AI and the finance domain on the job. The FDE posting emphasizes customer onboarding, ERP integrations, and turning deployment work into reusable tooling. The overlap is real, especially around messy data and customer-specific rules. Second Summit Indeed
Questions I’d rehearse, in order
An anonymous MTS candidate described conversational deep dives on architecture and agent systems. A separate FDE candidate reported a 45-minute second round focused on what they’d built and was asked about their most complex integration. That makes a project deep dive my best guess for Friday’s round. Glassdoor
| Likely question | What to cover |
|---|---|
| Walk me through the most complex system or integration you built. | Draw the data flow. Explain what you personally owned, one decision with a real trade-off, what broke in production, and the result. Have one number ready if you can substantiate it. |
| How would you make an agent safe to write to an ERP? | Let the model interpret messy inputs and propose an action. Validate against current records and tenant policy before writing. Explain how you’d recover if the write succeeded but the response timed out. |
| How would a workflow pause for human approval and resume? | Persist the proposed action and its evidence. Show reviewers what’s ambiguous and let them approve, edit, or reject. Recheck the invoice state before acting on an approval that might be hours old. |
| How would you evaluate the agent before turning on automation? | Replay historical cases, run in shadow mode, and track incorrect actions alongside automation coverage and reviewer workload. Turn each failure into a regression case. |
| A customer has a workflow nobody else uses. What do you build? | Find out whether it’s a hard business control or a preference. Use configuration for recurring differences, isolate customer-specific integration code, and bring repeated needs back to the product team. |
Practice case
A manufacturer gets one $25,000 payment covering several invoices. The remittance email is missing an invoice number, and SAP times out during posting. What happens?
Start by recording the payment and extracting candidate invoice references into structured fields. Check the customer, currency, open balances, and totals against SAP. Route ambiguous matches to a finance reviewer with the supporting evidence. Give the posting a stable transaction reference. If SAP times out, look up that reference and reconcile the outcome before retrying. Record the proposed match, approval, and ERP result so finance can reconstruct what happened. Start rollout in shadow mode and measure incorrect postings and how many payments can be matched safely without review.
What to say and what to ask
“Why Stuut?”
What caught my attention was the point that an overdue invoice often starts with a missing PO or a pricing mismatch. Once the software can investigate and act on that, the engineering gets interesting. You have to keep state across systems, give finance a way to review decisions, and recover cleanly when a tool call fails. I’d like to work on that end to end.
Ask them:
- “Can you walk me through a recent production failure that changed how you designed the agent harness?”
- “How do Applied AI engineers and FDEs split ownership when a customer’s workflow exposes a gap in the core product?”
- “What workflow would you want me to own in my first 90 days, and how would you measure whether it’s working?”