Anki

» see anki.com as it lived, on the wayback machine

~ the obituary ~

i first heard about anki the way most people in robotics-adjacent jobs did: through papers and demo videos from the team in the mid-2010s. anki kept coming up because their consumer products were doing real on-device computer vision at a price point nobody else was hitting. cozmo's little face-tracker ran on a tiny embedded soc, did slam-style spatial mapping on a coffee table, and recognized human faces well enough to greet you by name. for a 179-dollar toy in 2016 this was, professionally speaking, a bit absurd.

anki was founded in 2010 in san francisco by three carnegie mellon robotics phds. they raised about 182.5 million dollars across the company's life. andreessen horowitz led the early rounds, with index ventures and two sigma joining later and jp morgan adding debt financing in the last year. the first consumer product was anki drive in 2013, then anki overdrive in 2015: a smartphone-controlled toy car racing track with ai opponents that learned your driving. it sold well at toys r us, won "toy of the year," and convinced the founders to keep going.

cozmo shipped in 2016. it was the size of a forklift but wedge-shaped, with two animated eyes on a small color screen and a personality designed by ex-pixar and ex-dreamworks animators. you could play games with it. it had moods, drooping its eyes on a loss and dancing on a win. vector shipped in 2018 as the always-on home-assistant version of the same robot at 249 dollars, with cloud voice processing built in.

on april 29, 2019 anki told its staff that a critical funding round had fallen through and the company would lay off all 200 employees within a week. cozmo and vector owners woke up to robots whose cloud services were about to expire. digital dream labs bought the assets later that year, kept the cozmo and vector apps running in degraded form, and ships small-scale updates to a much smaller community.

~ the case file ~

~ the funny part ~

anki was not a failed product. it was a successful product at the wrong scale. cozmo and vector both got good reviews. the underlying robotics work was real. the founders had built something nobody had built before at that price. but consumer electronics needs you to sell millions of units a quarter at twenty-percent margins, and a 179-dollar robot toy that costs almost 179 dollars to manufacture is not a business you can scale to where venture investors get their money back. the late-stage funding round did not fall through because anki was failing as a product. it fell through because the math after that round was not going to work for the people putting in money.

~ what carried over ~

the engineering team mostly went on. some of them ended up at waymo or zipline. the on-device-cv work cozmo did at the toy price point is now what a million startups are trying to do on drones and household sensors. doing real computer vision on a small embedded soc with a small power budget is no longer a research problem, it is a product question. anki shipped that thesis a decade early and could not stay open long enough to be right about it.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »