I was picked by the recruiter for the product omni team, working on Airtable's vibe coding agent. Airtable recently acquired DeepSky to build HyperAgent, and there are plans to connect it with Airtable itself, so it's basically multi-agent development (HyperAgent -> Omni).
Recruiter call - 30 minutes, phone screen: why Airtable, what project I wanted to work on, my relevant past experience.
Coding - 60 minutes on HackerRank, implementing Airtable's undo/redo feature. Most of the code was already written, including the tests — I just had to add the undo and redo logic. Kind of like the LeetCode design-browser-history problem. Had to decide whether to store the full state, partial state, or just the sequence of commands. Lots of follow-ups after that.
OOD - How do you avoid changing the undo/redo code's logic every time you need to support a new kind of operation? Originally it only changed a single cell's value — how do you extend it to support deleting a whole row or a whole column?
How do you efficiently delete a row or a column without touching the entire row's or column's data?
Which cells can't be undone? There was some product-level discussion around this, but in the end I still had to write some code for it.
Handling undo/redo for cell updates that have dependencies on other cells. Kind of like Harvey's spreadsheet.
1-2 hour take-home, plus an HM round - 60 minutes, pure behavioral, leadership, and past projects.
The 1-2 hour take-home was to use HyperAgent to build an agent you'd actually use yourself — something like an agent that reads your email or checks your calendar — and then record a 10-15 minute video walking through your experience using it and the product you built. The stated evaluation criteria were the product itself plus how you used HyperAgent. Airtable gives you $1000 in credit to build on their own platform for this — basically like Lovable, but for building agents. There's a lot you can do with it; it can connect to other tools and systems.
The HM interview was separate, roughly 15 traditional behavioral STAR-style questions: the most challenging and complex project I'd worked on, an example where I personally changed a product's direction, how I lead and mentor people, handling unexpected events, handling situations that didn't turn out the way I expected, what I use AI for besides code generation to help me build things, what technical security concerns (prompt injection) I think about and handle, and what metrics I use to judge whether an AI product is ready to ship.
Discussion
Loading comments…