Pinned Tweet
🚨 The first edition of The Modern ERP Digest is coming soon!
Start your Monday sharp.
📬 New every week at 8am CT.
linkedin.com/newsletters/the…
2
198
Before Quill shows me a draft, it checks the text against a list of phrases I banned.
Things like "quietly," fake deadlines, and talking to a crowd as "most people." If a line trips the list, Quill rewrites it before I see it.
7
Ask if Integration Broker is “hardened” and you’ll get a nod about SSL. Ask who can reach /PSIGW/ and whether ANONYMOUS is still a superuser — and you’ll get a pause.
Five ways to make IB boring again:
peoplesoftcloud.com/p/5-ways…
10
Fathom plans our 2027 Europe trip. It builds options, tracks dates, and keeps the anniversary in view.
It doesn't book flights or hotels. When a plan is ready to book, it comes to me.
1
19
Pace coaches me for a 50k trail race in July 2027.
It owns the training plan. My other bots don't invent workouts or move long-run days, and training questions go to Pace.
8
Atlas keeps my Notion organized. Other bots don't pick their own folders.
When Quill or Wren needs to file something, they ask Atlas where it goes. Atlas also gates deletes and schema changes.
20
Scout's morning brief is built around one question: what do I need to decide today?
Things that are just status don't make the cut. Those stay with the bot that owns them.
13
When I deleted five bots in September, it was tempting to blame the bots.
But I trained them. Loose instructions got loose results.
Now I write a new bot's instructions like I'd brief a new hire: the job, the limits, and who it goes to when it's stuck. Then I add the bot.
7
In a side project, the bot that reviews code isn't the one that merges it.
A code reviewer checks the work. A merge manager decides when it lands.
Same reason a real team doesn't let the author approve their own pull request.
6
A question I ask about every bot: if it went wrong tomorrow, could I turn off just that one?
Because each bot has one job, the answer should be yes. Pausing Quill shouldn't touch Scout's brief or Pace's training plan.
7
Every bot on my team has a list of things it can't do. I think it matters more than the job description.
Atlas organizes Notion. It doesn't decide what's important.
Fathom plans a 2027 trip. It doesn't book anything.
12
Clerk triages my two inboxes. It doesn't answer anything.
When a message needs a reply from me, Clerk hands it to Wren, and Wren brings it to me for a yes.
One bot sorts. One bot asks. Neither one sends.
Keeping those apart is what makes a handoff safe.
6
My bots don't send email or post on their own.
They draft. When something needs my yes, Wren shows me Approve, Approve some, or Hold.
That one gate keeps the speed and keeps my name on every decision.
Same rule I'd give any AI agent on ERP data.
14
Last month I deleted five of my bots.
They didn't work the way I needed. That was on me. I trained them loosely, with overlapping jobs and fuzzy limits.
The ones that survived each have one owner, one lane, and written rules for what they can't do.
11
A partner runs our PeopleSoft. If they miss a patch, whose problem is it?
Yours.
A contract can hand off the work. It can't hand off the accountability.
After every cycle, ask for versions read from production, not a ticket marked done. #PeopleSoft
1
9
Six steps and named owners in this week's Q&A: open.substack.com/pub/people…
6
A rule for AI agents on ERP data:
Every agent gets a named human owner and its own identity.
Not a shared service account. Not the developer's login. Not "the team."
If it can read payroll, a person's name is on it, and you can cut its access without cutting anything else.
1
9
My side project runs on a small bot org chart: a program lead, a BA, a dispatcher, a code reviewer and a merge manager.
The rule that made it work: every decision for me goes through the program lead. Nobody else asks.
One front door. Same as a real team.
1
13



