What auto and AI leaders posted, October 17. A dealer drops the tools AI can rebuild at 90%

For a second day the feed is about the Tesla Roadster, but the most useful thing for a dealer was said more quietly. A dealer group CMO explained the rule he uses to drop vendors in favor of AI. Toyota showed how it wants to catch up with China in software. And the founder of SaaStr reported that his AI agent took a $100,000 deal all the way to payment.

News from the trade press is in the morning issue. AI tips are in a separate post. I added my take to each post.

What dealers say#

The 90/90 rule. A dealer drops a vendor when AI can do its job#

Ben Sporn, CMO at VIP Automotive Group of Long Island, told the Car Dealership Guy show how his team decides what to pay for. If AI can recreate 90% of a tool's functionality for 90% less, the work comes in-house. "So we no longer have to hire a third-party company, hire a developer, hire an SEO agency, and do all these things". Car Dealership Guy thinks this puts a large chunk of dealership software spend under pressure.

My take. The rule is good because it has numbers. Not "AI will replace everything", but two thresholds of 90%. What passes both goes in-house. What does not stays with the vendor, and nobody argues.

What you can do today. List every tool the store pays for each month and answer two questions for each one. How much of it do you use day to day. How much of that could AI plus one employee rebuild in a week. The first candidate to cut will be obvious.

Car Dealership Guy post on X

Inconsistent op codes break service reports in every store#

Maureen Martin of Dynatron Software says service departments spoil their own data. Technicians book the same job under different op codes. Across the 4,000-plus dealerships her company works with, this is the number one driver of what her team calls fixed ops "data fog". "Because that is an issue that is in every single dealership in the US and Canada."

My take. Sales has the same disease. One salesperson writes the lead source as "website", another writes "internet", a third leaves it empty. Then the manager reads a report by channel and makes a decision on data that does not exist. A list of allowed values in the CRM is boring work, but without it your analytics lie.

What you can do today. Open the "lead source" field in your CRM and count how many different values live there. More than ten means the report by channel is not worth reading until you clean it up.

Car Dealership Guy post on X

Porsche sells supercars to people who have somewhere to service them#

CJ Wilson, owner and GM at Porsche Fresno, explains the brand's advantage in the exotic-car market. You can service a $500,000 Porsche without living in Miami or LA. The dealer network is much larger than the rivals have. About his own McLaren P1 he says this. "I have 1,000% anxiety about servicing my P1."

My take. A buyer of an expensive car buys peace of mind for three years, not a 0-60 time. The nearest service center 250 miles away kills the deal faster than the price does.

A salesperson of any brand can use this. A map of service centers, time to get an appointment, a loaner car. A rival without a network cannot copy any of it.

Car Dealership Guy post on X

Toyota keeps its customers better than anyone#

In the J.D. Power 2026 U.S. Automotive Brand Loyalty Study, Toyota led mass market cars at 64.8% and mass market SUVs at 63.8%. In trucks it came second at 62.2%, behind Ford at 66.2%.

Tyson Jominy of J.D. Power credits the result to vehicle quality and service history. Brands with high loyalty, he says, give the customer the same good experience in sales and in service.

My take. Loyalty is not born in the showroom. It is born in the service lane a year after the purchase. If a customer comes back for a new car after three service visits, the service advisor earned it as much as the salesperson did. The service department needs a repeat-sales target too.

Car Dealership Guy post on X

A dealer expects that not every automaker survives#

Car Dealership Guy published a text from a dealer with several domestic-brand rooftops about the future of auto retail. His own answer is this.

It depends on the brand. There will be consolidation among automakers. Part of them will not survive. The ones that do will be stronger and take more share. The industry is changing, but the future is not as bleak as the dealer writes.

My take. For a dealer the question is not whether his brand survives. It is what he will do if it does not. A second brand under the same roof, a strong service department and a trade-in process that work with any make. A dealer who depends 100% on one automaker risks everything.

Car Dealership Guy post on X

What carmakers and the market say#

Toyota will put 50 models on its own software platform by 2030#

Automotive News reports on Toyota's plan for its Arene software platform. Today it runs on two models. By 2030 it will run on 50 nameplates and 10 million vehicles a year. The platform launched a year ago in the RAV4 after more than five years of development. The goal is the same one China and Silicon Valley chase, to make the car more like a smartphone.

CTO Hiroki Nakajima told investors about the plan at a technical workshop in Europe on October 9. By 2035 Toyota wants to offer Arene to alliance partners.

My take. Toyota is slow to decide and massive when it does. 10 million cars a year on one software base means one thing. Updates and paid features will reach a Corolla buyer, not only the owner of a premium EV. A Toyota salesperson will soon explain subscriptions and updates the way he explains trim levels today.

Automotive News post on X

Lexus moves 35,000 cars from other markets to U.S. dealers#

Lexus will redirect about 35,000 vehicles from overseas markets to U.S. dealers. This eases the supply squeeze that dragged the brand's sales down all year. The brand's 246 U.S. dealers heard this at the annual meeting in Atlanta.

They also heard that about 90% of their lineup will be all-new or refreshed within 18 months. The GR and Century sub-brands will go on sale over the next two years.

My take. No cars, no sales, however strong the brand. Lexus admitted it and moved stock from other countries.

The lesson for a dealer in any country is simple. When supply is short, fight for allocation, not for discounts. Allocation decides the year's revenue more than marketing does.

Automotive News post on X

Chinese luxury brands come to Europe, and in Paris you cannot miss them#

Automotive News writes that Chinese luxury brands that took share from BMW, Mercedes-Benz and Audi in China are bringing the fight to Europe. Analysts say it will be much harder than it was for Chinese volume brands. European premium buyers are slow to switch to new names.

Nestor Wu of Li Auto says it plainly. It will be tough, he admits. But in China they did a good job, and that means the product is good.

Lei Xing covers the Chinese car market. He spent a day at the UTAC track with Geely cars before the Paris Motor Show. At the show itself he counted about five Chinese driver-assistance systems doing demos, one of them from WeRide with Bosch. Automotive News released a video titled "China's automakers are taking over the Paris Auto Show".

My take. Chinese volume brands take Europe on price. Premium brands cannot do that. They need service, status and trust, and that takes years.

For BMW and Mercedes dealers this is a delay, not a defense. Five years ago nobody believed BYD would sell in Germany.

Automotive News post on X · Lei Xing post on X · Automotive News video

The Roadster, one day later. 10 billion views and questions about dates#

Sawyer Merritt writes that the Tesla Roadster unveil and its demos got more than 10 billion views across all social media in a day. The car will be built in its own building at the Texas plant, called RFTX. The slowest trim does 0-60 mph in under 1.6 seconds, quicker than the fastest version of the $2.5 million Rimac Nevera R. The battery charges from 15 to 80% in 10 minutes.

Automotive News collected the skeptics. Karl Brauer of iSeeCars recalls that the last Roadster was shown in 2017 with production promised for 2020. "Not holding my breath." Gene Munster of Deepwater called the reveal "an engineering flex with no commercial value beyond embarrassing traditional car makers". Production is promised for 2028.

My take. No ad budget can buy 10 billion views in one night. But a buyer in a store a year from now will not remember what he watched. He will ask when the car arrives and what it costs. Tesla already pushed the Roadster back by years once, and its dates are less trusted than its 0-60 times.

Sawyer Merritt post on X · Sawyer Merritt post on the factory · Automotive News post on X

Europe and Britain soften their EV plans#

Michael Dunne posted twice in a row. Britain, per an FT story, will cut the EV sales targets for carmakers by less than they asked. The energy department holds on to its climate goals. Dunne calls it being "stuck in a 2020 moment".

The second post, captioned "Facing reality", quotes Emmanuel Macron. France and Germany reached an agreement on the future of the European auto sector that keeps both climate ambitions and production in Europe.

My take. When heads of state discuss how to save the car industry, a dealer should review his EV plan for the year. Quotas and fines will change, and prices and stock will change with them. Selling what the customer wants today is safer than selling what the regulator wanted yesterday.

Michael Dunne post on Britain · Michael Dunne post on France and Germany

The U.S., Mexico and Canada trade deal waits for the election#

Automotive News writes that automakers and suppliers will not get answers on the future of USMCA until after the November 3 midterms. The U.S. and Canada trade war drags on, and talks with Mexico continue.

Companies hold back major investments. Neal Ganguli of AlixPartners says this. "I don't think the current conditions are conducive for making long-term capital decisions."

My take. For a U.S. dealer this means prices on cars built in Mexico and Canada can change by December. Tell the customer who is thinking until the new year about it honestly. He will either buy now or not blame the salesperson later.

Automotive News post on X

AI and sales#

An AI agent at SaaStr found a $100,000 deal and took it to payment#

SaaStr founder Jason Lemkin reported a $100,000 deal brought in by his AI agent called 10K. At the company the agent holds the title of AI VP Revenue. Humans closed the deal. But the agent found the customer, followed up, sent the contract, got it signed, prepared the invoice and got the payment in the same day. "That kid's good."

In the same thread Lemkin set a rule for vendors who ask him to try their AI agents. "We'll try it if 10K, our AI VP Revenue, wants to. It's up to 10K now."

My take. The important part of this story is not the $100,000. It is the list of actions. Found, followed up, sent the contract, got the signature, issued the invoice, checked the payment.

In a dealership this is exactly the routine a salesperson puts off while he talks to a customer on the floor. The human closes, the agent takes it to the money.

What you can do today. Take one deal from this week and write down every step from the customer's "yes" to the money in the account. Mark the steps the salesperson did by hand. That is your list for an agent.

Jason Lemkin post on the deal · Jason Lemkin post on the rule

Three AI tools worked on one Salesforce org, and the cheap one won#

SaaStr released a comparison. Three tools, Claudeforce, Replit and Lovable, were connected to the same Salesforce org with the same prompts. Per the episode description, Lovable won on design and cost, $2 against $12 for Replit and about $15 for Claudeforce. It was the only one that built a login on its own.

Claudeforce is still a read-only beta with 35 pre-built skills. Lemkin writes that running Salesforce through an agent is "10x-100x better than logging into the native Salesforce".

My take. The CRM screen stops being the place where a salesperson works. He tells the agent what he needs, and the agent goes into the database. Dealership CRMs are not ready for this. The first stores that put an agent on top of their system will get salespeople who do not spend an hour a day on buttons.

Jason Lemkin post on X · Episode on YouTube

For people who work with AI#

Mollick recommends a paper on how a team of agents split the roles on its own#

Ethan Mollick recommends a paper on how a team of Gemini 4 agents beat Minecraft. He calls it an accessible example of spontaneous coordination among agents to reach a complex goal, and of how their approach differs from humans. The paper is partly AI-written, he says, which makes it easy to read.

My take. The key word is "coordination". When a task is big, one agent is worse than two that split the work and check each other. In sales this is one agent that finds customers and another that checks its findings before they reach the salesperson.

What you can do today. Give one task to two AI sessions with different roles. The first drafts a reply to a customer, the second looks for mistakes in price and dates. Compare the result with what one session gives you.

Ethan Mollick post on X

OpenAI engineers gave an agent the job of finding what to remove#

OpenAI showed how its engineers use Codex to maintain code. Rodrigo Roim writes that a few months ago he had to tell the agent exactly where to look for dead code. Now the agent finds candidates in Slack, checks them against production telemetry and drafts cleanup PRs.

My take. The pattern moves to any department. The agent reads the chat, compares it with the data and brings a list of what is stale. In a store that is old price lists in chats, leads with no reply, promotions that ended but still sit on the website.

What you can do today. Give AI an export of your team chat for the last month. Ask it to find everything that was promised to customers and never closed. The list will be longer than you think.

OpenAI Developers post on X

The agent works the incident at night, the human steers from a phone#

OpenAI told the story of Ali, an engineer who was paged at 3 a.m. His agent investigated the issue on its own and started a Codex thread, and Ali steered the work from his phone in the ChatGPT app.

My take. A sales manager gets urgent questions on a day off too. By the time the phone rings, an agent has already pulled the customer's history, the messages and the car's status. That saves an hour and a lot of nerves. Steering from a phone matters more than it seems, because the laptop is never nearby at that moment.

OpenAI Developers post on X

In brief#

• Elon Musk says Tesla supervised self-driving is approved in Slovakia.
• CnEVPost reports that Nio will launch a 60-kWh battery for the Firefly on October 23, with a CLTC range of 560 km. Today the car has only a 42.1-kWh pack.
• BYD showed the Ti7 in Paraguay.
• Car Dealership Guy writes that Hyundai recalls nearly 95,000 Genesis vehicles over overheating power seat motors, with two fires reported.
• Lei Xing writes that Volvo Cars and Mercedes extended their partnerships with NVIDIA on compute, software and safety systems.
• Anthropic says Claude designed quantum circuits from plain-language instructions that beat the best published ones, sometimes by an order of magnitude.

Also on video#

• Automotive News, October 16 show. The Roadster and a talk with Jessica Gonzalez of InformedIQ on generative AI.
• Automotive News, the battery industry grows despite slumping EV sales. Where the battery business is going.
• SaaStr, why AI won't make junior employees obsolete. Atlassian flipped its hiring pyramid, more juniors and fewer in the middle.
• All-In, CoreWeave crash and Anthropic's investor day. A talk about whether the AI bubble has popped.
• Autocar India, Tesla Model Y L review. Three rows, two motors, 681 km for Rs 62 lakh.
• Lex Fridman, Jensen Huang in Taiwan. The NVIDIA CEO on why compute is worth a trillion dollars.