30-Day Onboarding Runbook

The day-by-day playbook for rolling out your Digital Worker. What you do, what we do, what you walk away with at every milestone.

This runbook is yours to keep. If you ever add a second workflow, a second hire, or hand the system off to someone on your team, you can re-run it from the top. Nothing in here is proprietary to Obsidian AI Labs. The whole point is that you can run it yourself.

Customers redacted throughout. Replace bracketed items with your specifics.

Pre-Day 1: What to have ready

Before kickoff, gather the following so the build moves fast on Day 1.

• One person on your side who can answer "is this email correct?" and "is this what you meant?" within 4 business hours. Their attention for 30 minutes a day during week one.
• Your most-wanted workflow. One. Not three. (Examples: client follow-up emails, monthly reporting, lead enrichment.) We narrow before we scale.
• Existing accounts for whatever the workflow touches: email, CRM, calendar, document storage, billing. We integrate against what you already use.
• 30 days of inbox/CRM/notes context as background reading material for the agent. You don't have to organize anything. We extract.

Week 1: Discovery and first deployment (Days 1 to 7)

Day 2 to 4: Build

We build the agent against your real data. You don't see anything yet. We do not ship a half-finished agent for you to look at; we ship something already running.

• Agent identity, voice, and access scopes set up. We extract a sample of your sent folder so the agent learns your real voice instead of a generic AI tone.
• Integrations wired (email, CRM, calendar, docs, etc.) with read-first scopes. Write scopes only after you approve them.
• First workflow tested on a sandbox dataset.

Day 5 to 6: Shadow run

Agent runs the workflow in shadow mode. It produces what it would have sent or done, but does not act yet. You review.

• You see every output before it goes live (drafts in your email, CRM notes you can review, calendar holds you can confirm).
• You give us 5 to 10 examples of "this is right" and "this is wrong, here's why".
• We tune voice, decisions, and edge cases against your feedback.

Week 2: Live operation and tuning (Days 8 to 14)

Day 8 to 11: Live with guardrails

Agent runs the workflow daily. Every action is logged and visible to you in a single channel (we suggest a dedicated channel in whatever messaging tool you already use).

• You can pause the agent on any specific contact, task, or workflow with one message.
• Anything outside the agreed scope routes back to you for a decision.
• We monitor for errors, weird outputs, missed handoffs.

Day 12 to 14: First adjustment cycle

By now you have real signal. Some things the agent does are exactly right; some need a small turn of the dial.

• You list every change you want (we give you a simple template).
• We ship those changes in batches, not one at a time.
• Agent stays live the whole time. No downtime for tuning.

Week 3: Expansion and tooling (Days 15 to 21)

Day 15 to 18: Add the second workflow (optional)

If your first workflow is humming, we can add a second now. Or we can keep optimizing the first. Your call.

• If expanding: same kickoff -> build -> shadow -> live cycle, compressed to 4 days because we already know your stack.
• If staying: we deepen the first workflow (more edge cases, better integrations, smarter follow-ups).

Day 19 to 21: Tooling and observability

We hand over the dashboards and dashboards-of-dashboards.

• Daily summary report: what the agent did, what it skipped, what needs your input.
• Weekly hours-saved estimate so you can see the math.
• Direct access to logs if you ever want to audit any specific decision.

Week 4: Stabilize and prepare for review (Days 22 to 28)

Day 22 to 26: Smooth out

By now we have three weeks of agent operation data. We use it to:

• Tighten the prompts and decision rules where the agent has been over-cautious.
• Loosen the prompts where the agent has been over-confident.
• Add small workflows that have emerged organically (the "while you're in there, can you also..." asks).

Day 27: Pre-review prep

We compile the 30-day numbers so the Day 30 review is concrete, not a vibe check.

• Hours of admin offloaded per week.
• Outputs produced (emails sent, contacts updated, reports filed, etc).
• Any moments where the agent caught something you would have missed.

Day 30 review

The Day 30 review is an email exchange, not a meeting. By Day 30 we both already know what's happening day-to-day. The point of this review is to put the numbers on paper and make a decision about month two. You'll get an email from us with the metrics already filled in. Your job is to send back three answers. Easiest path: paste our email into your AI and ask it to draft your reply. Edit, send. Five minutes total.

Day 60 review (mid-pilot check)

Also an email. Lighter touch than Day 30, but more meaningful, because by Day 60 you're starting to forget what life was like before the agent ran. We pull the numbers so you can see the gap. Same five-minute reply pattern as Day 30.

Day 90 review and exit interview

Day 90 is two emails. First, a transition email with the final pilot snapshot. Second, a short exit interview (five questions) that helps us improve and helps you reflect on whether the system is doing what you wanted it to. Both come pre-filled with metrics. Use your AI to draft replies.

The Founder Desk Line transitions to standard support after Day 90. Here's what changes and what doesn't.

• Stays the same: The agent keeps running. The integrations stay live. The dashboards keep updating.
• Changes: Response time on questions moves from "Brandon within 4 business hours" to "team within 1 business day".
• Optional: A short case study if the numbers worked out. You see the draft before anything ships public, and you can redact anything.

If you ever want to add a second workflow

Use this runbook from the top. The compressed version: kickoff (1 day) -> build (2 days) -> shadow (1 day) -> live (1 day). Five business days from "I want to add X" to "X is live", because the foundation is already built.

If you ever want to take it in-house

You can. The agent runs on your accounts, with your credentials, against your data. The prompts are yours. The runbook is yours. The whole system is structured so it is not held hostage to us.

Two paths to going fully in-house:

• Figure it out yourself. This runbook plus access to the prompts and integration configs is enough for any technically-comfortable person on your team to take over. We hand over the source on request.
• We show you. Bring whoever on your team is going to own it onto a working session. We walk through the architecture, the prompt files, the integration patterns, and how to extend or modify any of it. You leave that session able to run and grow the system without us.

Our preference is that you keep us around because we are the cheapest line item that does the most work. But we built this so you don't have to.

Questions, escalations, or you spotted something the agent shouldn't have done? Direct line: brandon@obsidianailabs.ca. The first 90 days, that is the actual line, not a ticket queue.