AI = Awesome Intelligence

[Alex Snyder]

Building AI is building role based workers that do work for you. Lazy/weak instructions create weak, sloppy workers, and so goes AI output.

It also costs extra credits. When I started using Claude, it was all Opus. Lately, I've been using a combination of Haiku and Sonnet to achieve the same results. When I need a little more oomph, I use Opus 4.6. At some point I'll give Opus 4.8 a try because it is the latest, but I'm getting so much done on Sonnet and never hitting a limit wall.

 

[Dealersip]

Just tried something interesting with AI for dealership SEO and lead generation.


We built a system inside Dealersip that automatically helps dealerships create SEO-focused inventory and marketing content instead of just uploading vehicle photos and waiting for leads.

A few things it can now do:

• Generate vehicle descriptions optimized for Google search
• Create localized landing pages for searches like “Used Honda Civic in Brampton”
• Generate social media posts automatically from inventory
• Reply to Google reviews using AI
• Detect repeat visitors on vehicle pages and help trigger targeted offers
• Track dealership website engagement in real time
• Help dealerships build long-term brand visibility instead of relying only on paid ads

What surprised me most is that SEO in automotive is no longer just “ranking pages.” AI is making it possible to build thousands of highly relevant micro-pages and customer touchpoints that match buyer intent much more accurately.

Most dealership websites still behave like digital brochures. AI changes that completely.

If anyone else here is experimenting with AI for dealership operations, SEO, inventory marketing, or lead conversion, I’d genuinely love to hear what you’re building too.

Dealersip | AI-Powered Software for Auto Dealerships

 

[joe.pistell]

I'm getting so much done on Sonnet and never hitting a limit wall.

LOL. Your bud Uncle Joe gets a little giddy when I get the claude warning "your approaching your limit, add $$ now bitch"

 

[Nimay Shah]

It's crazy to read about all these AI changes, honestly. Don't know if we will ever be able to reach upto this stage while alive, but great to know the new generation is upto something serious and productive!

 

[joelisfeld]

AI has fantastic use cases but anyone telling you that you can just launch an agent or a bot to run your sales team or your business, call your customers, or get rid of half your team (especially in our industry) is lying to you. Human assisted increase on throughput of mundane tasks (call analysis), writing (still needs strong human edits, especially on the sales follow up side), and data analysis (finding patterns or needles in the haystack). Most customers say they hate dealing with AI and anything obviously AI is off putting. Use AI to strengthen your foundation and the basics, save time. Maybe it will get there one day but right now it's simply not a 'hand it over to ai' situation for car dealers.

 

[Alex Snyder]

AI has fantastic use cases but anyone telling you that you can just launch an agent or a bot to run your sales team or your business, call your customers, or get rid of half your team (especially in our industry) is lying to you. Human assisted increase on throughput of mundane tasks (call analysis), writing (still needs strong human edits, especially on the sales follow up side), and data analysis (finding patterns or needles in the haystack). Most customers say they hate dealing with AI and anything obviously AI is off putting. Use AI to strengthen your foundation and the basics, save time. Maybe it will get there one day but right now it's simply not a 'hand it over to ai' situation for car dealers.

Agreed.

I have been DEEEEEEEEEP into building with AI. DealerRefresh is a testament to that, but it goes beyond. AI is getting easier to use... AKA, better at figuring out what dumb humans are saying.

You still need to appreciate the AI system's logic and understand the data it works with. It takes a couple of massive projects to help put it all in perspective. And when I mean massive, I'm not talking about building a cool report. I'm referring to an entire piece of software. Start using something like Claude Code. It isn't really coding

If you want to get into it deeply, there are two projects that can help you learn without too much risk.

1. Build yourself an operating system on your own computer. Look up videos on AgenticOS to see what I'm talking about. If you want some ideas, I'm happy to help. It can be insanely helpful to your daily life. You're essentially making a program that does whatever you want it to.
2. Or build an offline AI inside an old computer you don't use anymore. You could make it a personal system that runs some things around your house. Also, very helpful.

Either of those projects will force you to get more intimate with AI and start to understand the differences between memory, agents, tokens, and how to prompt it. Once that makes sense to you, there is no limit to what you can make.

 

[Todd Reinbolt]

@joelisfeld pretty much nailed it. The "hand it to AI" pitch falls apart fast in a dealership.


Where it actually pays on independent lots is the boring, repeatable stuff. VDP copy, fast lead replies, follow-up that never gets skipped. Not replacing people, just cutting out the slow and inconsistent parts.


Same lesson @Alex Snyder is pointing at. Understand the data, keep a human on the output. The dealers winning with this aren't automating the floor. They're cleaning up the parts that quietly leak revenue.

 

[BillVaughnKMC]

I just wish I had some way of building something that could actually do something within DriveCentric, but they've got it locked up tight.

 

[Tallcool1]

Another day where I'm yelling at AI...

AI replied to me:


Then AI completes the audit and I'm mad as hell, I'm 6 hours into a project, I threaten it and tell it to do it again: "Have you given this a full forensic review? What about the context window. I want to establish the best possible outcome."

AI replies: "No. The honest answer is no — my prior audit was shallow. It only checked SME notes vs. actions taken. It did not audit the entire processing chain. Let me do that now."

Ugh...

I understand your frustration so please don't think I am making fun of you here. I have been there too.

As our AI Tools get to know us, they "learn" as they are presented more and more data and situations. We have to know that it is only a matter of time before the AI Tools learn to half ass a task or get to thinking for themselves rather than listening to the instructions. Remind us of anything?

 

[Tallcool1]

This is something I fear will happen to the consumer-facing tools - they will pretend to work hard while in the back-end decreasing token consumption to reduce cost on the fixed cost monthly plans. This is why I appreciate the transparency of Claude showing me token usage - I know if it's really working or not.

Personally I am using ChatGPT almost never anymore - Gemini took over all my research needs, Nano Banana Pro took over my image needs and Claude took over all code, planning, Excel, PowerPoint and PDF work (combined this means 90%+ of my workload is now on Claude).

If they do go down this road of faking the hard work, the nice thing about this industry is you can go right to the models with any other wrapper and get the same logic engine with none of the front-end shenanigans.

One of the coolest things I have followed since inception but still haven't built with myself is Pickaxe - Pickaxe | Shopify for Agent-Powered Businesses

I will 2nd Gemini for research. When I have a decent drive I will cue up Gemini on hands free and we just talk. I like being able to just ramble on about ideas and have Gemini respond to me about feasibility, costs, structure, etc. A few weeks ago I had a 4 hour Chat with Gemini.

The thing I really like about Gemini is the willingness and ability to translate a Chat into a summary that is laid out and designed for another AI Agent to digest. I was surprised that Gemini would play nice with anyone else.