Learn With Nathan
Sign in
Full sample chapter · Build with Agent Worker
Build Your First AI Assistant
Build a useful first assistant in Agent Max by choosing the right preset, writing a bounded job, granting the minimum capabilities, testing, and approving its first actions.
Chapter 01
Build Your First AI Assistant
Build a useful first assistant in Agent Max by choosing the right preset, writing a bounded job, granting the minimum capabilities, testing, and approving its first actions.
Agent Worker has several AI surfaces: start in Agent Max when you want an assistant you invoke in conversation; use Ava, Voice, or Automations later when work begins from an event or channel.
AI can answer, retrieve, and propose actions through the capabilities you explicitly grant. The accountable person must choose the preset and job, prepare sources, set permissions, review proposals, test failures, and decide release.
This chapter keeps your first Agent Worker assistant practical for a business reader: the job, evidence, controls, finished artifact, and success measure are all visible. A low-to-medium technical learner can later implement the same method in approved software.
The business case for your first Agent Worker assistant
Start with a measurable result, not an AI feature
your first Agent Worker assistant creates value only when the Agent Worker configuration completes a defined business job with accepted quality, visible evidence, controlled permissions, reliable handoffs, and measurable operating results.
For the first build, stay private, use one narrow job, grant read capabilities before writes, leave every write on Ask me first, and use a low-risk workspace case.
Agent Worker provides official guidance relevant to the workflow, obligation, or control used here.
Design the work before asking AI
Good inputs and acceptance rules reduce rework
Choose the correct Agent Worker surface, prepare authoritative workspace data, define the job and limits, grant only necessary knowledge and capabilities, keep writes approval-first, test representative cases, then review activity and outcomes before expanding.
The two concepts to remember are and . Agent surface means the Agent Worker area that matches how work begins: Agent Max for an owner conversation, Ava for business events, Voice for calls, or Automations for fixed triggers. Commit policy means the per-action choice between Ask me first and Just do it.
| Weak setup | Business-ready setup | Why it matters |
|---|---|---|
| Create an AI assistant that can run my business and do whatever is necessary. | In Agent Max, configure a private “Daily Work Assistant” to read today’s appointments, open and overdue tasks, and pending approvals; it may propose a follow-up task but must Ask me first. It cannot send, book, alter contacts, or use private client content outside the workspace. | It maps directly to Agent Max capabilities and creates a safe first learning loop. |
Confirm consent, confidentiality, account settings, permissions, and applicable policy before giving any tool the information used in your first Agent Worker assistant.
FIRST Agent Max build: the practical playbook
A repeatable sequence a busy professional can follow
- 1. Prepare the workspace. Open Contacts, Tasks, Calendar, and Approvals; correct one small set of records the assistant will read.
- 2. Open Agent Max. Choose an existing assistant or create one from Chief of staff; open Settings.
- 3. Configure four tabs. Set Instructions, optional Knowledge, What it may do, and review the Limits.
- 4. Test the conversation. Ask five normal, two ambiguous, one missing-data, and one prohibited question; inspect tool activity.
- 5. Approve + review. Inspect the proposed task in conversation or Approvals, edit or reject as needed, then revise instructions before wider use.
Context to provide: Agent Worker workspace, business owner, target user, trigger, contact or case state, approved knowledge, enabled modules, permissions, success criteria, test cases, and escalation owner.
AI job: Configure and run the tested first-assistant configuration inside Agent Worker; preserve source, proposed actions, approvals, corrections, run evidence, and release decision.
Return: Tested first-assistant configuration with workspace route, source and permission map, instructions, test cases, approval evidence, result, exceptions, and next review.
Quality rules: Do not grant or imply access, consent, facts, knowledge, contact identity, availability, authority, approval, send, booking, system update, or success that Agent Worker has not verified.
Worked example: A private daily-work assistant for one owner
Follow the evidence from messy input to an approved result
FIRST Agent Max build: the working loop
Five Agent Worker build gates move from a real job and governed workspace data to approval-first action, tested release, and observed results
Audiobook description
The figure shows a five-step path from left to right. Step 1, Prepare the workspace, Open Contacts, Tasks, Calendar, and Approvals; correct one small set of records the assistant will read. Step 2, Open Agent Max, Choose an existing assistant or create one from Chief of staff; open Settings. Step 3, Configure four tabs, Set Instructions, optional Knowledge, What it may do, and review the Limits. Step 4, Test the conversation, Ask five normal, two ambiguous, one missing-data, and one prohibited question; inspect tool activity. Step 5, Approve + review, Inspect the proposed task in conversation or Approvals, edit or reject as needed, then revise instructions before wider use. A reminder below the path says to begin with a bounded use case, verify the result, and improve the workflow.
| Starting material | AI-assisted result | Human review |
|---|---|---|
| Ten accurate contacts, one week of appointments, open and overdue tasks, supported questions, prohibited actions, and nine test cases. | Private Agent Max assistant with Read your day and Create tasks, approval-first task creation, test log, corrected instructions, and release decision. | The owner rejects a vague task, requires a named contact and date, and keeps Send a text and Book appointments switched off. |
Apply it safely and measure the gain
A faster draft is useful only when the finished work is better
Build in one controlled workspace with representative but non-sensitive or appropriately protected data; leave every write at Ask me first and every public or outbound channel off until the tests pass.
Measure: Track correct reads, missing-data honesty, prohibited-action refusals, approval edits, duplicate tasks, owner time, and task usefulness. Measure the complete workflow, including preparation, review, correction, exceptions, and recovery—not generation speed alone.
Keep the tested first-assistant configuration, approval record, exception notes, and metric result as the evidence for whether this Agent Worker build should be repeated, changed, or stopped.
Pause and apply
Reflection questions
- Where does your first Agent Worker assistant create the most avoidable delay, inconsistency, or risk today?
- Which input, judgment, or external action must remain under explicit human control?
- What evidence would justify expanding this Agent Worker build after the first test?
Sources and further reading
Chapter 1 endnotes
- Agent Worker. Agent Max, 2026. Read the verified product reference.
- Agent Worker. Approvals, 2026. Read the verified product reference.
- Agent Worker Academy. Build With Agent Worker Production Brief, 2026. Academy blueprint.