# Likely Person B says nobody cares about skills; likely Person D asks if we'd ship them today

In Wednesday's Agent product review, likely Person B offered a hot take he said usage data backs: “I don't think anyone cares about skills. Outside of maybe us.” Likely Person D partly agreed. He still treats skills as the way to deliver automations. His test for fooling themselves: more skill usage without any change in satisfaction, revenue or memberships is a failure. His other check: with no skills today, would they ship them? His answer was yes.

---

# A UC Davis IT head tried the Agent in a 30,000-person Slack, private channels first

In a beta onboarding call, Person M, executive director of IT for UC Davis's College of Engineering, installed the Agent as workspace admin. Setup offered only public channels. Person M wanted “private channels first,” citing a Slack of about 30,000 people and a risk-averse public university, but called trying it “an appropriate risk for now.” A host said it was probably the largest workspace to install so far. Person M connected only GitHub.

---

# Compounding felt like homework on the launch page; the room tried “Never repeat yourself”

Reviewing the Agent landing page in Wednesday's launch standup, likely Person I said the compounding section “makes me feel like I have to do all of these things.” That did not match the real experience of teaching the agent something once. Likely Person D agreed that compounding “happens automatically” and is not literal Compound Engineering. The room liked “Never repeat yourself” as a header, and someone added your line: don't make the same mistake twice. No final copy was set.

---

# Telling the Agent not to relay internal counts made it drop numbers everywhere

Engineer A's open Plus One prompt PR tried to stop replies from pasting a provider's internal tallies (“112 candidates and kept 5”). Every wording that named counts as something to withhold also made the Agent blur planted timings in unrelated replies: 1/3 twice, against 3/3 on main. Adding “never overrides the numbers rule” scored worst, 0/3. The current head deletes that vocabulary and bans only opaque debugging strings; its score is pending.