Argo AI

» see argo.ai as it lived, on the wayback machine

~ the last words ~

i tracked argo ai from the tech press, not from any hands-on stake. by late 2022 edge-vision perception was becoming my day-job in ankara. reading argo pdfs on lidar-camera fusion was professional research. i had no ford and no robotaxi anywhere near me.

bryan salesky and peter rander left google's chauffeur project and uber's atg to found argo ai in pittsburgh in november 2016. ford put in a billion dollars in february 2017 as a five-year commitment. the pitch was calm and expensive: build the self-driving stack, ford would build the car, robotaxi service would follow.

volkswagen came in during june 2020 with 2.6 billion, taking the same 42 percent stake as ford. that made argo ai the joint bet of two of the world's biggest automakers. by july 2021 lyft added a small position and the valuation hit 12.4 billion. the company had around 1700 engineers and open test programs across seven cities on two continents.

by 2022 the money was harder. argo tried to raise an outside round. amazon did not bite. ford ran the math on how many more billions level-4 was going to cost before the first paying ride, and jim farley concluded it was not going to happen on a ford quarterly cadence. on 26 october 2022 ford took a 2.7 billion dollar non-cash impairment. vw expensed 1.9 billion euros. staff learned the same day on an all-hands call.

~ the record ~

~ what the shutdown looked like ~

some engineers got picked up by ford's own driver-assist unit and by vw's cariad software subsidiary. the rest were laid off with severance, offices in miami and austin closing inside a week and the seven test fleets rolling to a stop. bloomberg and techcrunch ran the story the same afternoon. the internal shock was less about the shutdown itself than about how fast it arrived after the july 2021 party.

~ what it left behind ~

argo published a fair amount of open work along the way, including the argoverse driving dataset which is still used in academic perception papers. some of the researchers moved to waymo or aurora. the rest ended up spread across the general av-perception job market. by late 2022 cruise still had the industry lead. two years later cruise itself was in restructuring after a san francisco pedestrian incident. the whole robotaxi timeline slid to the right one more time.

~ what it proved ~

that a startup built entirely on two automakers' patience is a startup with a single failure mode. ford could survive writing off 2.7 billion. it could not survive being wrong about that number two more times. jim farley said the quiet part out loud: profitable, fully autonomous vehicles at scale are a long way off. argo ai was the last serious attempt to bet a legacy oem's balance sheet against that timeline.

@}-->--

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

« previous  ·  random  ·  next »