A blueprint for how AI agents run a real business — one Chief agent, a named lead for every function, and the task agents beneath them. Every lead here is a live agent — chat with one, hear it reply, then see how ready your business is.
Give it a plain-language task → it routes to the right function lead → merges their answers back to you.
▸ Brief AdaOpus · orchestrator
Function Leads
Tap a lead to collapse its team · tap ＋ Warehousing under Operations to open a vertical
Operations ▸ Vertical
Warehousing
The example vertical built under Operations — same shape, its own specialist task agents.
Stock & Inventory
Live counts, low-stock & discrepancy alerts
Pick-Path / Slotting
Optimises pick routes & where SKUs live
Goods-In
Books deliveries, checks POs, flags shortages
Dispatch & Labels
Picks courier, prints labels, tracks parcels
Returns
Triages RMAs, restocks, spots return patterns
Swap Warehousing for Manufacturing, Field Service, Hospitality, Clinic or any operation — the Operations lead just gets a different specialist team beneath it.
Legend
The stack behind each agent
Nothing exotic — off-the-shelf models + automation, wired per function. This is the key for the whole blueprint.
◇
Human
Owner or employee — briefs a lead, reviews & approves outputs. Always in the loop.
◆
Chief agent · Opus
Orchestrator. Routes tasks to leads, merges answers. Your one front door.
■
Function lead · Sonnet
Owns a domain, holds its knowledge base, delegates to its task agents.
▫
Task agent
One job, done well — draft this, update that, chase this. Cheap + fast.
⚙︎
Automation
Make / n8n — the triggers, schedules and hand-offs that run without you.
◎
Voice
Call a lead and talk to it like a colleague.
⧉
Tools
CRM, inbox, docs, spreadsheets, data — connected via MCP / APIs.
▤
Knowledge base
Each function's private, gated "brain" — clean data = useful agents.
How people use it: staff brief a function lead by chat or voice for their day-to-day. Execs query any lead's brain over Slack / WhatsApp ("how many new leads this week?"). Leads pass context sideways so Sales knows what Marketing just shipped.
Where to start
Your first three agents
You don't switch everything on at once. Start with the high-volume, low-risk, rules-based work — prove it with a human approving everything, then expand.
01
N
Content & posts
Nova · Marketing
Highest volume, lowest risk. Drafts blogs, posts and emails in your voice — nothing goes out until you approve it.
≈ 5 hrs/week back
02
P
First-line support
Pepper · Customer Service
Answers the same questions 24/7 from your FAQs and past tickets; flags the ones that genuinely need a human.
≈ 8 hrs/week back
03
S
Invoicing & admin
Sterling · Finance
Raises invoices from completed jobs and reconciles receipts — repetitive and rules-based. Money movement stays human-approved.
≈ 4 hrs/week back
Phase 1 · Weeks 1–2
Switch on one
Pick the biggest time-sink above. Run it on real work with you approving everything (Amber).
Phase 2 · Month 1
Add the low-risk two
Wire the tools, set Green/Amber/Red guardrails, let the safe work run unattended (Green).
Phase 3 · Quarter 1
Bring in the rest
The judgement-heavy functions (HR, deals, ops) — under tight guardrails, always human-approved.
Time-back figures are typical illustrations, not a promise — the audit sizes the real numbers for your business.
N
Nova
Marketing lead
Impressed? This is one generic agent — imagine it on your data.