Linear vs Height

Height was founded in 2018 by Michaël Villar, previously at Stripe, and raised around $18 million including a Series A led by Redpoint. For most of its life it was a well-designed, fast issue tracker in the same competitive set as Linear, with a chat-first interface that made task discussion feel less like filing and more like talking.

In October 2024 it relaunched as Height 2.0 and made a much larger bet, describing itself as an autonomous project collaboration tool. The pitch was that project management is mostly repetitive judgement: triaging incoming bugs, spotting duplicates, chasing stale tasks, keeping specs aligned with what the code actually does. Height put a reasoning engine behind those jobs and gave every plan, including the free one, access to it.

It was a genuinely interesting product and it did not find a market fast enough. That is worth saying plainly rather than glossing, because the failure carries information: teams liked the idea of an agent that tidies the backlog, and not enough of them would change tools for it. Tidying is not the expensive part of the work.

Height's thesis was that a project tool should do project management. The part it chose to automate was the housekeeping: triage, duplicates, stale specs. That is real work, and it is not the work anybody is short of. Nobody has ever missed a deadline because the backlog had duplicates in it.

Polaris makes the same general bet in a different place. Instead of automating the maintenance of the record, it puts AI workers on the roster as teammates and gives them the work itself. Humans and agents are rows in the same members table, so assignment is identical for both. Assign a task to an AI worker and a real cloud machine wakes for it, runs a live tool loop with web search, ticks the acceptance criteria on the task, and posts the finished output as a comment with any files it produced. The machine never closes the task; a person does, and rates it.

Worth saying given the subject of this page: an autonomous project tool has already failed in this market once, and Polaris is in free public beta with no customers, no revenue and no case studies. What exists is a working product, recorded demos including an unstaged machine session, and pricing that charges nothing for the software and roughly two dollars per human-equivalent hour of work actually delivered, itemised on a work log you can challenge. Judge it on that rather than on the category.