How we use AI at The Collecting Group

AI
1

I recently went on a podcast/videocast to discuss how we use AI at TCG: collectingcars.com and watchcollecting.com

I’d love to hear your views on it.

LinkedIn (more interaction here): How We Think About AI at The Collecting Group
Blogpost (for those that dislike LinkedIn): How We Think About AI at The Collecting Group – Olly Brand

More worryingly, I’ve been asked to do an AMA at a conference this week. If I make it through that without vomiting on myself, I’ll share that here too :smiley:

2 Likes
2

I made it through my AMA at the conference without any incidents :slight_smile:

I’d almost go so far as to say I enjoyed it after surviving the opening minutes.

I also forgot how much I enjoy tech conferences. The session replays won’t be available for about three months, but I’m working on a write-up of my key takeaways. I’ll share the write-up and the replay link here once they’re ready.

2 Likes
3

Interesting discussion! Looking forward to the AMA writeup.

I have a bit of experience with this field. I was working on an auto auction web app last year (in the US.) We were looking at conservative uses of Bedrock, e.g. natural language to set our filters and detecting obvious discrepancies in a window sticker.

It was also going to be helpful on the coding side for upgrading high tech debt parts of the app, e.g. generating a set of replayable migrations from some manually written SQL.

I wouldn’t be surprised if they’ve set up more tooling since I’ve stopped working with them, probably Claude on Bedrock. There wasn’t anyone with power thinking through it like you have, though.

1 Like
4

The replay will be live in October, I’ll add more to that and share here.

Until them I did a post from the event…

5

The replay is LIVE, CRINGE!!!

I’m consoling myself by coming up with a blog post for next week.

Also: Moving On: From TCG to Cognito

All feedback and Qs appreciate :smiley:

1 Like
6

Thanks for sharing this! Re: the question about letting business teams build their own software, we are also trying to fend this off from a couple of enthusiastic individuals. I think we are going to end up maintaining a set of APIs for these kinds of apps and we’ll categorize endpoints or data attributes as safe for any of our coworkers to consume, or requiring per-app vetting (or some such.)

I think you’ve got the right idea to let people try things, bring a buddy along, share what was learned.

1 Like
7

Thanks, would MCP help vs. an API?

8

We could provide an MCP server. We use swagger to generate documentation so we’d probably just make the API documentation available through the server and let the no code tool work from that information. We could try to build safety into the MCP server, but we’d want to backstop it with rules at the level of the API’s server anyway.

1 Like
9

Write-up, Cursor 2.0 is currently dominating my thoughts :smiley:

1 Like