InstaLILY builds vertical AI agents for the unglamorous half of the economy: distributors, field-service firms, industrial supply. Its product, Lily, which the company bills as “the world’s first AI Forward Deployed Engineer,” generates custom software that automates pricing, logistics, and field-service workflows inside a customer’s systems, and goes live in days rather than months. That positioning is the whole reason the interview looks the way it does: an engineer here does more than ship a feature behind an API and move on. They sit in discovery sessions with a Home Depot or a United Rentals, map a messy real-world workflow, and build the agent that runs it. So the loop tests two muscles at once: can you build an LLM system that works, and can you sit across from a non-technical executive and turn what they say into something shipped.
There’s no public question bank, so the postings are the study guide. The full-stack listing is specific about the stack: TypeScript, React and Next.js on the front end, Python with FastAPI and SQLAlchemy on the back, PostgreSQL for data, Temporal for workflow orchestration, GCP and Docker underneath, and a long tail of enterprise integrations. The closest siblings are other applied-AI startups that embed engineers with customers: the Decagon interview guide and the Harvey AI interview guide cover the same build-for-a-named-customer rhythm, and the broader AI-native company interview guides hub collects the rest. Where InstaLILY differs is the vertical: its customers run warehouses and parts catalogs, not law firms or support queues, so domain messiness is the product.
What InstaLILY is, and why the funding tempo sets the bar
InstaLILY was founded in 2023 by Amit Shah and Sumantro Das, is headquartered in New York, and has offices in San Francisco, London, and Toronto. In July 2026 it closed a $60 million Series B led by Energize Capital, with Insight Partners and strategic investors Home Depot Ventures and United Rentals participating, per the funding announcement coverage. That brings total funding to roughly $100 million. The company says revenue grew fivefold over the prior year, and it lists customers including The Home Depot, United Rentals, PartsTown, Henry Schein, Radwell International, PartsSource, and Kedrion Biopharma. It also claims one national distributor saw over $200 million in new annual sales from the system, which is the company’s own number, not audited.
This matters across the interview table because the strategic investors are also customers: Home Depot Ventures and United Rentals both fund InstaLILY and run on it. A mediocre deployment costs a customer and sours an investor at once, so the loop will probe whether you can be trusted in front of those accounts.
The interview loop, as best anyone can infer it
One Glassdoor report describes an HR screen, then a 30-minute technical phone screen, then a final on-site that ran about six hours and was described as mostly general and conversational, heavy on background and past projects rather than whiteboard puzzles. A separate 1point3acres write-up describes phone calls, a take-home project, and on-site coding, and includes the candidate’s notes on company culture and compensation. Those two don’t fully agree: one mentions a take-home, the other doesn’t, which probably means the exact sequence depends on the role and the month. So confirm the round count, the take-home, and the on-site format on your first recruiter call rather than trusting either report as current.
The one logistics detail worth pinning down early is in-office expectations, because the sources conflict. A third-party listing claimed five days a week in the New York or San Francisco office for the first six months, then four days in person with one remote. The company’s own Greenhouse posting states no remote or hybrid policy and only says the role involves regular in-person customer work. Ask directly, and given the embedded-with-customers model, plan for a heavily in-person job whatever the written policy says. The postings don’t state a visa-sponsorship policy either, so if you need sponsorship, raise it on the first recruiter call, especially with roles split across New York, San Francisco, London, and Toronto.
InstaLILY engineer interview loop at a glance
| Stage (inferred from two candidate reports and Greenhouse postings, Oct 2026 — not an InstaLILY-published loop) | What the reports and postings suggest | How to prep |
|---|---|---|
| HR / recruiter screen | Background, motivation, logistics including in-office expectations and location | A specific reason you want applied AI for physical-goods firms, not generic “I like LLMs” |
| Technical phone screen (~30 min) | Neither report describes the content; at applied-AI startups it’s usually a short practical problem or a walk through past LLM work | Be able to narrate an agent or RAG system you built end to end, with the hard parts |
| Take-home project (role/vintage dependent) | Reported by one candidate, not the other; likely a small full-stack or agent build | Ship something clean and runnable with a short README; treat it as production code |
| On-site (reported ~6 hours) | Mostly conversational: past projects, a coding portion, architecture and tradeoffs | One project you can defend end to end; practice explaining it to a non-engineer |
The coding bar: full-stack craft, not puzzle speed
Both reports describe practical work over algorithm trivia, which fits the stack. The full-stack role asks for three to five years shipping production systems, so the coding portion is likely closer to “build or extend a small feature” than “invert this tree under time pressure.” Still, don’t skip fundamentals. Keep the coding patterns catalog sharp enough that hash maps, sets, and two pointers are reflexes, and keep your Big-O analysis automatic, because even in a practical problem the interviewer is watching whether you pick the structure with the right worst-case behavior. If there’s a take-home, treat it as the real signal: it compiles, it runs, it has a test or two, and the README explains the tradeoffs you made. For a company that ships into customer systems in days, clean and legible beats clever.
System design for agents that run inside someone else’s systems
This is where the applied-AI part earns its keep, and the posting hands you the material: agents, RAG pipelines, and evaluation harnesses, wired to a pile of third-party enterprise integrations. A realistic prompt, shaped from those responsibilities, might be: walk me through how you’d build and ship an agent that answers pricing questions off a distributor’s product catalog, and how you’d know it’s right. A strong answer treats the model as one component in a larger system design, not the whole thing. Start by splitting the data; that split is what a real ML interviewer is listening for. A product catalog is structured, so the right tool is text-to-SQL or a typed tool call against the catalog database, not a vector store, and database indexing on the lookup keys is what keeps reads fast on rows that get updated all day. Reserve RAG for the unstructured material around the catalog: spec sheets, return policies, service manuals. Pitch “chunk and index everything” and expect pushback. Talk about evals: how you measure whether the agent is accurate before it touches a customer’s real pricing, because “it looked good in the demo” is how you lose a United Rentals. Talk about the integration seam: enterprise systems are slow and flaky, so you want idempotent writes, retries, and a workflow engine (InstaLILY uses Temporal) to keep a multi-step automation from half-completing and corrupting a customer’s state. If you can name where the agent should refuse and hand off to a human, you’re showing the judgment the embedded model actually needs.
Candidates underweight the human-in-the-loop story. InstaLILY’s pitch is that the system runs the business with people in control, so a design that can’t explain what it did, can’t be audited, and can’t be overridden is a bad one here no matter how accurate. Bring that up before the interviewer has to.
What they screen for beyond code
The conversational on-site carries most of the signal. “Forward deployed engineer” is InstaLILY’s marketing name for Lily the product, not your job title (the openings are Software Engineer I, II, and Full Stack), but the full-stack posting still describes a client-embedded role: running discovery with customers, embedding during rollouts, presenting to executives. So the behavioral and past-project rounds are checking whether you can run a technical discovery with a non-technical stakeholder, translate a vague operational pain into a scoped build, and present architecture and tradeoffs to an executive without hiding behind jargon. Prepare one project you can narrate end to end, with a real decision and a real tradeoff in it, and keep it tight with the STAR method. Then practice telling that story to someone non-technical, out loud, until the business impact leads and the implementation follows. That reordering is close to the entire job.
Comp, with the usual caveats
InstaLILY’s job postings carry real base bands, and they vary by level. As of postings dated mid-July 2026, Software Engineer I in New York or San Francisco listed at $100,000 to $120,000, Software Engineer II at $120,000 to $140,000, and the more senior Full Stack Software Engineer role at $150,000 to $190,000, all base, with stock options on top and refreshers described for top performers. A London posting listed £54,000 to £62,000, a different market that doesn’t convert cleanly to the US bands. Those bands top out at the Full Stack role; anything more senior isn’t published, so a staff-plus candidate negotiates the top from the outside. Anchor on those numbers: there’s no InstaLILY-specific data on Levels.fyi yet, and the general New York medians there measure total compensation, not base, so they aren’t comparable. The equity is the live variable: one Series B deep with an undisclosed valuation, the option grant is both the upside and the hardest piece to price, so get the strike, the share count, and the latest preferred price, then model it with the total comp calculator. Treat the grant, not the base, as the center of any salary negotiation. Since the posted bands skew modest for New York, you’re weighing early equity and unusually direct customer exposure against a higher cash number somewhere more established.
InstaLILY interviews like what it is: a young applied-AI company whose engineers are expected to be in the room with the customer. Build something real with LLMs that you can defend end to end, get comfortable explaining technical tradeoffs to someone who doesn’t write code, and have a sharp answer for why automating a parts distributor’s workflow is the problem you want.
Practice the behavioral round:
