L

Agent Activity — owner-first

Show the owner what their agents did — not how they did it.

The business owner is non-technical. They don't want turns, tokens, or prompt tiers — they want outcomes in plain language, conversations they can read, and a nudge when something needs them. The full telemetry from #1493 still exists, but it lives one click deeper as Technical logs for operators. Activity becomes a tab inside the agents area — both org-wide and per-agent.

Owner value
Trust → autonomy
Read what it said → comfortable letting it act on its own.
Owner value
Is it worth it?
Conversations · replies · bookings — the renewal story.
Owner value
Pull me in only when needed
Escalations & opt-outs surfaced; everything else stays quiet.
Screen 1 /ai-agents — the list view gains tabs

AI Agents

Chat with your agents, or see what they've been doing.

Live

The Activity tab shows for owners / operators only (gated by AiAgent:List). Members see just Agents.

Conversations
47
this week
Replies
18
38% reply rate
Booked
6
appointments
Needs review
2
waiting on you
Needs your attention
Lead Follow-up — Acme Co asked a billing question it couldn't answer.
Review →
Text Reply — A customer asked to stop being contacted. Confirm opt-out.
Review →
Recent activity
Lead Follow-up Working now

Texting 12 leads about Tuesday's no-shows — 3 replied, 1 booked so far.

now
Text Reply Booked

Maria asked about her missed cleaning — rebooked her for Thursday 10:00 AM.

4m
Follow-up Drafter Needs review

Drafted 8 follow-up messages and is waiting for your approval to send.

1h
Lead Follow-up No reply yet

Sent the spring promo to 48 contacts — 9 opened, no replies yet.

Jun 9

Plain-language sentences, not telemetry. Each row opens that agent's Activity. Your personal Assistant & Orchestrator are excluded from this org roll-up (privacy) — you still see them on their own detail page.

Screen 2 /ai-agents/:id — the detail view gains tabs

Lead Follow-up

Riley · re-engages missed appointments and books open slots

This week, Riley handled 12 conversations, got 3 replies, and booked 1 appointment. 1 conversation needs your review.

1 booked 3 replied 8 no reply
Conversations
Re-engage Tuesday's no-shows — Maria

Maria rebooked her cleaning for Thursday 10:00 AM.

Booked 4m
Billing question — Acme Co

Couldn't answer a question about an invoice — escalated to you.

Needs review 22m
Spring promo — Daniel

Sent the offer; opened twice, no reply yet.

No reply 2d

Owner-scoped: you own this agent, so you see every conversation — including your personal Assistant's when you open its page. Outcome chips (Booked / Replied / No reply / Needs review) come from the agent's tool calls.

Screen 3 /ai-agents/:id/activity/:threadId — what the owner clicks into

Re-engage Tuesday's no-shows — Maria

Booked

Outcome: Maria rebooked her cleaning for Thursday 10:00 AM. No action needed from you.

Riley (your agent)
Hi Maria! We missed you for your cleaning on Tuesday — no worries at all. Want me to find you a new time this week?
Maria
Yes please! Thursday morning would be great.
Riley (your agent)
You're all set for Thursday at 10:00 AM. We'll send a reminder the day before. See you then! 😊
Booked appointment · Thu 10:00 AM
Reading as the owner — plain and complete. View technical logs →

Billing question — Acme Co

Needs review

Why you're here: Riley wasn't confident answering a question about an invoice and stopped rather than guess.

Acme Co
Why was I charged twice on my last invoice?

The reader reuses our existing chat bubbles — already built. New: the one-line Outcome / Why you're here header, the inline outcome chip, and the escalation banner with owner actions.

Screen 4 — one click deeper · operators only
Back to conversation

Technical logs

Lead Follow-up ·3f9c2a1 ·claude-sonnet-4.6
Turns
3
Tool calls
5
Tokens
48.2k
Wall time
11.3s
1 Turn 12:14:02 PM 14.1k↓ 320↑ · 3.1s
Effective prompt 8 / 12 layers
Tier 0Platform rules1,920 ch
Tier 0.5Org rulesempty
Tier 2Persona286 ch
toolsend_smsok · 240ms
{ "to": "+1•••4821", "body": "Hi Maria! We missed you…" }

This is exactly what #1493 already built — prompt tiers, tokens, tool I/O. Nothing new to build here; it just moves behind the "View technical logs" link so owners never have to see it.

What's new vs. what exists
Already built — reuse
  • Conversation bubbles (the readable transcript)
  • The full debug timeline — prompt tiers, tokens, tool I/O (#1493)
  • Org & per-agent thread queries; the Tabs primitive
New — the owner layer
  • Outcome chips (Booked / Replied / No reply / Needs review) — derived from tool calls (send_sms → "texted", create_appointment → "booked")
  • Plain-language summaries — tool-call-derived first; short generated summaries later
  • "Needs your attention" — escalation / opt-out / low-confidence detection
  • Tabs wiring on both views + the conversation reader