Onboarding
From Chat to Execution — How One Question Becomes a Result
How one message becomes completed work, step by step — every agent handoff and every human approval gate, with an audit trail you own.
Send a message. Watch it become work. This page explains what actually happens between your message and the result — every step, every handoff, every place you stay in control.
This is the governance journey inside AI FrontDesk, how a single message becomes a traceable, approved business action. The system is properly managed and safely configured by expert humans, and each agent operates within a defined authority boundary. Nothing consequential runs without your approval.
See the full AI Departments system
The Journey of One Request
Here is an illustration, not a description of a shipped feature: a restaurant owner sends a Sunday evening message — "Run the weekly inventory reorder." Here's what happens:
01 — Your message arrives on your chosen channel. You send it over WhatsApp, SMS, Google Chat, or email. It doesn't matter which — they reach the same AI team. The message is received, your identity is confirmed against your business account, and the request is parsed for intent.
02 — The right agent picks it up. The request is routed to the agent with authority over inventory and supplier operations. That agent works from the context you have shared with it, such as your stock sheet, approved suppliers and standing reorder rules. It doesn't ask you to repeat yourself.
03 — The agent works from what you have shared. In this illustration the agent reads the stock sheet and supplier emails you have shared as files. (If your inventory or accounting tool has a way to connect, the link is part of the Business Partnership plan. MCP is the open standard for agent-tool communication.) It reads current stock levels, calculates what's below par, cross-references your approved supplier catalog, and checks available delivery windows.
04 — A draft action is prepared. The agent doesn't act yet. It assembles the reorder: which items, which suppliers, what quantities, what total cost, what delivery date. Everything is staged and ready. Nothing has been sent or committed.
05 — You get a summary and approval request. Back on WhatsApp (or wherever you messaged from), you receive a clear summary: "Reorder ready — seven items, three suppliers, $1,240 total, deliveries Thursday and Friday. Approve?" You can approve it, decline it, or ask a follow-up question.
06 — You approve. The agents execute. Your approval triggers the send. Purchase orders go to suppliers by email. The entries are logged in the record. Delivery confirmations are queued for follow-up.
07 — You get a confirmation and can walk away. "Orders placed. PO numbers logged. I'll follow up if any supplier doesn't confirm by Tuesday." That's the complete loop — from your Sunday evening message to completed, traceable work.
Where Humans Stay in the Loop
The approval gate in Step 5 is not optional. For any action that involves committing money, sending external communications, modifying records, or taking an irreversible step, your agents stop and wait for your explicit go-ahead.
This is called a human approval gate. It's built into the agent architecture — not a setting you can accidentally turn off. The agents handle all the preparation: the research, the drafting, the cross-referencing. You handle the decision.
Routine read operations — summarizing a document you shared, looking up a supplier's contact — don't require approval. They just answer. The gate applies to actions, not answers.
The Audit Trail You Can Read Tomorrow
Every step of the journey above is logged in a reviewable audit trail. You can review exactly what your agents did, in what order, with what data, and based on whose approval.
This is governed by the Nest Protocol: the governance layer that controls agent identity, authority boundaries, and action logging across your entire AI team. If you ever want to know why an agent did something — or didn't do something — the audit trail tells you. It's not a black box. It's a record you own.
Another Example — Campaign Launch, Start to Finish
A marketing agency owner wants to launch a new campaign. She messages her AI team: "Launch the Q3 retention campaign — email and SMS, existing client list only."
Her CMO-role agent confirms the audience scope, pulls the approved campaign brief you shared, and queues the send for her review. She gets a preview of the first email and the SMS copy, confirms the audience segment, and approves. The campaign goes out. Delivery stats come back to her afterward.
The campaign record — brief, send list, performance data — lives in her owned system. Not the email vendor's.
Frequently Asked Questions
Can the AI act without my permission?
No. Any action that has real-world consequences — spending money, sending communications, modifying records — requires your explicit approval before it executes. Agents prepare; you decide.
What if a step fails halfway through?
If a step in a multi-step task fails — a confirmation doesn't come back — the agent surfaces the failure and the current state to you. It does not silently retry indefinitely or proceed past a failed step. You decide how to proceed.
Can I see why an agent did something?
Yes. Every agent action is logged with a reason — what input triggered it, what data it used, what decision it made, and what authority boundary it operated within. You can review any action from your audit trail at any time.
How do I prove AI agent ROI to my team?
Start with one high-impact workflow — a weekly reorder, a campaign launch, or a daily summary. Measure the time it took to run that workflow before and after.
Putting AI Agents to work
We make no time-saved promise. The practical way to measure it: run one high-frequency workflow through your AI team for 30 days and track it against your prior manual time. When a result is approved and logged, you have a record of the action taken.
When owners compare AI brands — Claude, ChatGPT, Gemini, Copilot — they're comparing engines, not the car. Connected Spaces runs your AI team on whichever leading model does the job best, and can use MCP, an open standard, where a tool offers a way to connect. The difference is ownership: you get an agent layer built on your data and configured to your business, not a subscription to a shared assistant that knows nothing about your operation.
Go Deeper
Related
- Message your AI team over WhatsApp, Google Chat, SMS or email — How owners chat with their AI team from the channels they already use.
- How AI Agents connect to tools (MCP, explained) — MCP, the open standard, and how we connect your tools.
- How your AI team coordinates — @-mentions, agent-to-agent handoffs, and HITL approval boundaries.
- Who is on your AI team? — Specialist agents, each watching a different part of your business.