Tesla isn’t giving up on automation just yet. On Monday, CEO Elon Musk revealed the company is currently hosting a hackathon to fix two of the worst chokepoints in the automated production line. The company has struggled to automate the production of its $35,000 Model 3, leading to delays.

Musk’s announcement was posted in response to an Ars Technica story about the company’s issues. The company’s plan to produce 2,000 cars per week by December 2017 was based on automating the Gigafactory into a “machine that builds machines.” After only meeting this figure last month, Musk admitted that “excessive automation…was a mistake” and that “humans are underrated.” The Ars story notes that General Motors also spent billions on automation in the eighties without ever seeing a return. The lesson in both instances was that automation is best implemented in increments into an already-successful production line.

General Motors faced serious issues in its endeavor. The story quotes Paul Ingrassia and Joseph White’s book “Comeback,” which describes the Hamtramck, Michigan plant: “The spray-painting robots began spraying each other instead of the cars, causing GM to truck the cars across town to a fifty-seven-year-old Cadillac plant for repainting. When a massive computer-controlled ‘robogate’ welding machine smashed a car body, or a welding machine stopped dead, the entire Hamtramck line would stop.”

Musk anticipated teething problems from the start, describing the first few months at the July 2017 start of manufacturing as “production hell.” The company has gradually moved past this stage, but Musk reevaled at the start of last month that he has started sleeping at the factory to fix delays. YouTuber Ben Sullins arranged a crowdfund campaign to buy Musk a decent couch to improve his sleep, which he delivered to the factory this month.

Tesla is set to move past the 2,000-per-week rate with two upgrades, with the goal outlined in a leaked e-mail to reach production levels of 6,000 per week by the end of June. The company will have to work hard to fulfill a backlog around 400,000 reservations.

It may take some more time to fine-tune the machine that builds machines.