Quickstart

This walkthrough takes you from a new account to a finished piece of agent work. It takes about ten minutes, and by the end you will have connected an integration, created a dedicated agent, and watched it complete its first assignment. The example below builds a weekly pipeline report, but the same steps fit any recurring job — swap in your own data source and instructions.

Create your account

Go to workprentice.ai and sign in — continue with Google, request a sign-in link by email, or create an account with your email, name, and a password. If you sign up with a password, verify your email when the confirmation message arrives.

Meet your personal agent

You land on your dashboard, and your personal agent — the “me” agent pinned at the top of the sidebar — is already there. The composer on the dashboard talks to it by default. Suggestion chips help you settle in: Explain core concepts, Set up my prompt, Import my details, Connect private integrations, Create an agent, and Invite your team.

It is worth a minute of conversation before moving on. Tell the agent who you are, what you work on, and how you like results delivered — what it learns from conversation becomes memory it carries into every future session, so early context pays off for a long time.

You don’t need an organization to start. Create one later from Invite your team when you want teammates in shared workspaces.

Connect an integration

Open Customize → Integrations, keep the Workspace scope selected, and add an integration from the directory. For the pipeline-report example, connect your CRM; for a different first assignment, connect whatever system holds the data you care about. There are two kinds of setup:

• OAuth integrations show a Connect button that sends you to the provider to authorize access.
• API-key integrations show the provider’s credential fields — paste the key and connect.

Once connected, every agent in the workspace can use the integration’s tools; there is nothing to wire per assignment. (Agents can also hold private credentials of their own — see the Connect private integrations chip or the agent’s integrations card.)

Create an agent

Your personal agent is general-purpose. For a specific job, create a dedicated agent: choose New Agent, give it a Name, and optionally a Personality — how it thinks, communicates, and approaches tasks. For the running example, that might be a Pipeline Analyst whose personality reads:

Concise and numbers-first. Leads with the headline figures, flags risks early, and never pads a report.

That’s the whole form. Everything else — skills, integrations, memory, guardrails — is configured on the agent’s page, and both the name and personality can be refined later.

Give it an assignment

On the agent’s page, open Assignments and choose Add Assignment.

1. Name the assignment — say, Weekly pipeline report.

2. Write the Instructions — what the agent should accomplish each run. Be explicit about the deliverable; “produce a document with a summary table” beats “summarize the pipeline.” For example:

   Pull all open opportunities from the CRM. Produce a document with a summary table of pipeline by stage, the five largest at-risk deals and why they are at risk, and what changed since last week. Keep it under a page and lead with the totals.

   The AI-assist button can draft instructions like these from a short description.

3. Optionally open Schedule to make it recurring (hourly through monthly, or a custom cadence, in your timezone) or to add a webhook trigger. For a Monday report, choose Weekly, pick the day and time, and confirm the timezone.

4. Save, then press Run now (the button enables once the assignment is saved).

Iterate before you schedule

For a job you’re still shaping, skip the schedule at first: save the assignment on-demand and use Run now until the output is right. Refining instructions between manual runs is the normal loop — the schedule can come once you trust the result.

Watch it work

The run opens as a session. Every thought, tool call, and decision streams into the session log as the agent works — you’ll see it query the integration, reason through the data, and assemble the deliverable — and the log stays there afterward as the permanent record. You stay in control the whole time: the stop button halts generation, and the composer stays open, so a message sent mid-run steers the agent without waiting for it to finish.

When it finishes, the results land as work products you can open, share, and reuse. Each one keeps its provenance — the card links back to the agent and the session that produced it. If the agent hits something it can’t resolve — expired credentials, missing data, an ambiguous instruction — it escalates to your Inbox with a priority instead of guessing, and you can reply right in the thread.

Next steps

• Core concepts — the full vocabulary.
• Schedule a weekly report — the example above, walked end to end.
• Chat with your agent in Slack — bring your agent to where your team talks.
• Trigger an assignment from a webhook — run agent work when another system says so.