Just started Jake’s courses and wanted to share my first real use case, because it worked way better than I expected.
At my work (we’re a 3PL) I built a Claude routine that reads every client’s Slack channels each weekday morning and posts a sentiment digest for our account managers. Five clients a day, so every client gets a read about once a week. All the rules lived in one Notion page called “Process, Rules, Whatever.” lol.
That page got long. Every time the routine screwed up, I added a rule. It skipped a billing channel and missed a credit we’d already issued, new rule. It marked three channels “not checked” because it ran out of room, new rule. It made up two Slack links (yes, made them up), new rule. Every run had to swallow the whole page: client exclusions, Slack IDs, formatting rules, the list goes on and on.
So I installed the icm-architect skill and pointed it at the page. It barely had to ask me anything because the page already had the answers. What it built:
- a 5 step pipeline: 00 feedback, 01 pick clients, 02 read every channel, 03 verify, 04 post
- 8 rule files, and each rule lives in exactly one place
- a folder for every client with their channels, account manager and digest history
- each step loads only what it needs, about 3,000 tokens. The old way loaded the whole page every single run.
My favorite part is step 02. It writes a checklist with one row per channel, and the run can’t post a client until every row says read. That was the exact mistake that kept happening. Now a file blocks it.
Then we put it on GitHub, pointed the cloud routine at the repo and cut the Notion connection. Before going live I ran a test that only posts to my DMs. Five digests in about six and a half minutes, it pulled 27 client profiles from HubSpot along the way, and nobody on the team saw a thing.
Last piece. Before every run it now reads the Slack channel for feedback from my boss and our two account managers. If someone says a digest got something wrong, it goes back to the source, fixes the rule or the record, and DMs me what changed and why (all automatically). So every correction makes the next run better.
Feels amazing actually taking action instead of just learning.
Big thanks to . First time putting this into practice and it’s pretty wild.