What auto and AI leaders posted, October 9. F&I profit per car rose 5.7%

Friday is about earning more from what you already have. Dealers make more on each deal, even with slightly fewer deals. Atlassian says that after two years with AI, the hard part is not speed. The hard part is who steers.

Here is what car dealers, automakers, and AI-in-sales people posted on X and YouTube in the last day. News from outlets is in this morning's issue. AI tips are in a separate post. I added my take to each post.

What dealers say#

F&I profit per car rose 5.7% with fewer deals#

Automotive News looked at StoneEagle numbers for the first half of 2026. Finance and insurance gross profit per car sold rose 5.7% year over year, to $1,989. Average monthly F&I income per store also hit a record, up 2.6% to $217,705.

The average store did 109 deals a month, down from 113 a year ago. The number of products per deal stayed about the same. The second quarter drove most of the gain. Deals per month rose to 112 from 106 in the first quarter, and profit per car rose to $1,996 from $1,982.

My take. Products per deal did not change, but profit did. So the price of what dealers sell went up, or they sell more of the costly products. The post does not say which. I would not copy this result before I see what it is made of.

What to do today. Take your last 30 deals.

Calculate F&I profit per car for each manager. If the gap is big, the market is not the reason. The talk in the F&I office is. You can review that talk and improve it.

Post by Automotive News on X

A group grew from 6 stores to 17 by buying troubled ones#

Car Dealership Guy posted a podcast episode with Michael Brown, president and owner of Empire Automotive Group. Matthew Haiken asks the questions. The group grew from 6 stores to 17 by buying distressed dealerships.

The logic is simple. When a group has scale, it has room to fix other people's problems.

Brown puts it this way. Buy problems that size can fix. The previous episode was about 2008. The group cut headcount then, and had a record year in 2009. It grew from 10 stores to 26.

My take. This trick works only for people who know how to fix. If you buy a troubled store without a clear process for sales and service, you only add losses. Scale is a condition here, not the cause.

If you run one store, you may not have the scale to buy. The method is still the same. Find one department in your store that is always "in trouble" and fix it with one process. Do not change people before you change the process.

Episode on Car Dealership Guy's YouTube

Lithia sold two Los Angeles stores#

Buyer Victor Oh bought Audi Downtown LA and Volkswagen of Downtown LA in California from Lithia Motors. The deal closed on September 28, 2026. The Dave Cantin Group advised the seller. The post adds that Lithia also owned a Porsche store, but the end of that line is cut off in the feed.

My take. A large chain sells two stores in one area. The post gives no reason, and I will not guess.

I see one thing here. Store sales now go through specialist advisors, and every deal leaves a public record. If you think about buying or selling, read these records as a reference book.

Post by Car Dealership Guy on X

What automakers say#

Porsche rules out a U.S. factory, GM and Lear talk about partnership#

Automotive News published two videos after its congress in Detroit. In the October 8 news show, the North American heads of GM, Mercedes-Benz, and Hyundai talk about U.S. investment and the China threat. The CEO of Porsche showed a turnaround plan and ruled out a U.S. factory.

In the second video, GM president Mark Reuss and Lear CEO Ray Scott talk about trust. It builds over decades and turns a supplier into a strategic partner. They also touch on AI.

My take. One line matters to a dealer.

If Porsche does not build a U.S. factory, prices and delivery times for U.S. buyers will depend on tariffs and logistics. I would put this into the delivery-time talk with a client. I would not promise a date with a small margin.

Videos on YouTube. News, October 8 and Behind the Wheel with Reuss and Scott

AI and sales#

Atlassian put AI in 20 products. Three lessons against common advice#

Jason Lemkin, founder of SaaStr, summarized a talk by Sherif Mansour, who runs AI at Atlassian. More than 5 million people use these AI tools every month, across more than 20 products.

Lesson one. Build chat into the product, even if it feels redundant.

Two years ago, people at Atlassian thought chat was temporary. They thought customers would use ChatGPT or Claude instead. That was wrong. People use the AI that is closest to their work.

Chat also showed what to build. Users asked to group whiteboard notes into themes, and that became a feature. They asked to push notes into a backlog, and that became a workflow.

Lesson two. Add AI to what you already have. Atlassian put one new box in the Jira workflow designer that says "use an agent here."

Now one rule applies to every product. Anything you can do with a human, you can do with an agent. Mention it, give it a task, add it to a workflow.

Lesson three. "Everyone becomes an AI builder" did not hold at scale. About 10 teams got hand-picked builders, including product managers and designers who code with AI. They were fast for a few weeks, then they stalled.

Everyone was rowing and no one was steering. The ratio of product and design people to engineers is 1 to 10. With AI it feels like 1 to 30 or 1 to 40. If the product manager is coding with AI, nobody decides what to build.

The company also changed hiring. More juniors, more seniors, fewer people in the middle. Juniors are almost twice as likely to try AI, and seniors are better at catching bad output. Atlassian puts them in one room for four days every couple of months.

My take. A dealer has the same pattern, without the words "agent" and "backlog."

A sales team gets an AI assistant. Everyone tries it their own way. Then each person counts leads differently.

One person must decide what the agent does in the store. One person, not a committee.

What to do today. Name an owner for the AI assistant in your sales team. Pick the person who answers for lead results, not the person who knows the tool best. Once a week, this person looks at which requests repeat and turns them into standing workflows. This is what Atlassian did.

Post by Jason Lemkin on X. The full talk is on YouTube

After a round closes, write to everyone who bet on you#

Lemkin advises founders. When the round closes, send each person who bet on you a personal thank-you email. Maybe they were lucky to get an allocation. But they bet on you for a decade, and a short email means more than it seems because almost nobody writes it.

My take. This is not only about investors. It is about any long trust.

In the car business, the same thing works with a client who bought a third car from one salesperson. A short personal message after a deal is rare, and people remember it. I would pick five clients who bought from you this month and write to each one. Not a mass message. One line with the name and the model.

Post by Jason Lemkin on X

For people who work with AI#

The full set of tips is in today's AI tips post. Here is what I found separately while preparing this issue.

Simon Willison built a new feature for his blog fully by voice in Codex Desktop, while he cooked dinner. The post has no more detail than that fact. Post on X

What to do today. Pick one small change in a tool you use at work and dictate it to the AI instead of typing. Check how exactly it understood what you asked.

Ethan Mollick shared results of randomized trials with the old GPT-4o. AI access raises student test scores, and a smaller gain stays a week later. The winners use AI as a tutor, which means augmentation.

For students who let AI write for them, which means automation, the gain fades. Post on X

What to do today. When you give work to AI, ask yourself if you learn from it or hand work to it. For habits you want to keep, choose the first.

In brief#

Mollick on what is next for Google. After Gemini 4, the question for Google is what to do with a good model. He says Anthropic and OpenAI move toward one interface for several tasks with orchestrator agents. The Gemini 3 era was a set of separate products. X

Mollick on Gemini 2.5 in urgent care. Physicians rated the advice of the old Gemini 2.5 Pro and Flash as similar to doctors' advice. The models had no patient records, and no safety issues were found. Mollick adds that models have since become much better. X

Lemkin on Claudeforce. The next episode of his show The Agents promises a live demo of Claudeforce. It also compares headless Salesforce with Replit and Lovable. X

Stripe on the first year. Maia Josebachvili of Stripe says $100 million in year one is no longer the exception but the rule. Most of the revenue comes from outside the U.S. YouTube

Reflection AI showed the open model Beam. Co-founder Misha Laskin talks on the No Priors podcast about a 500-billion-parameter open-weight model. He believes open models will take most of the world's token demand. YouTube

What else was on screen#

These are titles without a review.

• Car Dealership Guy. Fixed Ops Friday with Dennis Gingrich, Cody Short, and Travis Graddon
• Carwow. The new electric VW ID. Tiguan, in German
• All-In. David Sacks says the economy is stronger than you think