The autonomous drones are based on deep learning.

Judged as of 2018-01-01. The criteria apply to the specific autonomous drones Andreessen was discussing in context (to be identified from the source; likely a drone product such as Skydio's). True: technical documentation, company statements or credible reporting show that the drones' core autonomy (perception, navigation, obstacle avoidance) relies on deep learning / deep neural networks. Mostly true: deep learning is a central component, alongside substantial classical methods (for example SLAM or geometric computer vision), so 'all' is a slight overstatement. Mixed: deep learning is one of several comparably important components. Mostly false: deep learning plays only a minor role and the autonomy is mainly classical or non-learning methods. False: the drones do not use deep learning.

Set before the research began.

1. E1 Skydio - Wikipedia Wikipedia · 31 August 2021

   “In 2018, the company's first consumer product was the Skydio R1. The Skydio R1 had 12 cameras around the body of the drone and a gimbal stabilized 4K main camera. It had subject follow mode and obstacle avoidance.”

2. E2 Skydio - Wikipedia Wikipedia · 31 August 2021

   “In 2012, Bry and Bachrach helped develop autonomous-control algorithms that could calculate a plane's trajectory and determine its location, physical orientation, velocity, and acceleration.”

3. E3 Skydio - Wikipedia Wikipedia · 31 August 2021

   “Early investors included venture capitalist Andreessen Horowitz.”

4. E4 Skydio - Wikipedia Wikipedia · 31 August 2021

   “The X2 uses the same Skydio Autonomy engine with 6 navigation cameras located on the outside of the drone but now with a thermal camera,”

5. E5 Skydio - Wikipedia Wikipedia · 31 August 2021

   “It had subject follow mode and obstacle avoidance.”

6. E6 Skydio - Wikipedia Wikipedia · 31 August 2021

   “The X2 uses the same Skydio Autonomy engine with 6 navigation cameras located on the outside of the drone but now with a thermal camera,”

7. E7 Skydio Demonstrates Incredible Obstacle-Dodging Full Autonomy With New R1 Consumer Drone IEEE Spectrum · 13 February 2018

   “we have a deep learning-based system based on all the people we’ve recorded during our testing, and we use that to make predictions about how somebody might move.”

8. E8 The autonomous “selfie drone” - Robohub Robohub · 14 March 2018

   “Using a deep neural network, it compiles information on each object and identifies each individual by, say, clothing and size.”

9. E9 HOT CHIPS, 2021 Hot Chips conference (Skydio keynote slides by Abraham Bachrach) · 12 March 2026 Own record

   “8 models used in flight”

10. E10 HOT CHIPS, 2021 Hot Chips conference (Skydio keynote slides by Abraham Bachrach) · 12 March 2026 Own record

    “A large optimization problem that jointly solves over 50 objectives including:”

The 2018 reports show the Skydio R1 used deep learning to predict people's movement and identify individuals (E7, E8). They do not show how far its navigation and obstacle avoidance relied on deep learning compared with other methods. The only signs of non-learning methods come from 2012 work on planes (E2) and a 2021 deck (E10), so the evidence cannot settle how accurate "all based on deep learning" was as of 2018.