The new agents don't require engineering
GPT-5.5 Workspace Agents are designed for operators, not engineers. There is no API to call, no webhook to configure, no Zapier account to set up, and no code to write. The agent runs inside ChatGPT Enterprise, connected to the tools your team already uses, configured in plain English. The hardest part of your first agent is not the technology. It is deciding which task to delegate first.
The 8-minute estimate in the title is not marketing. It is the actual measured time from "new agent" button to "agent activated and scheduled," assuming your CRM and Slack are already connected to your ChatGPT Enterprise account. If they are not yet connected, add 3 to 4 minutes for the OAuth steps. Either way, you will have a running automation before your next meeting starts.
The 8-minute playbook
Minute 0 to 1: Open ChatGPT Enterprise
Navigate to chatgpt.com and log in to your Enterprise account. In the left sidebar, look for "Workspace Agents" below your recent conversations. Click "New Workspace Agent." A setup panel opens on the right side of the screen. Name the agent something specific and functional, not something clever. "Monday Sales Recap" is better than "Sales Bot" because the name is the first signal to the agent about its purpose, and it is easier for teammates to identify in the agent list when you have more than one running.
Minute 1 to 3: Write the one-line goal
Every Monday at 8am, pull last week's closed-won deals from CRM, summarize the top 3 objections, and post a 5-bullet recap in #sales-wins Slack channel.This one line contains everything the agent needs: a schedule (every Monday at 8am), a data source (CRM, closed-won deals), a transformation (summarize top 3 objections), and an output destination (#sales-wins Slack). That structure - when, from where, do what, deliver where - is the template for every effective agent goal. If any of those four elements is missing, the agent will ask clarifying questions before running, which means it will not run on schedule.
Minute 3 to 5: Connect your tools
Click "Connect" next to Slack and your CRM in the tool panel on the right. One OAuth authorization screen for each. The authorization takes about 30 seconds per tool. During the OAuth step, you will see exactly what permissions you are granting: read access to deal data, write access to the specified Slack channel. You do not need to grant write access to CRM for a recap agent. Read-only is correct for this use case and is the safer starting point for any first agent.
Minute 5 to 7: Review the proposed plan
After you save the goal, the agent generates a step-by-step plan for its first run and shows it to you before activating. Read every step. This is not a formality. The plan reveals whether the agent understood your goal correctly. Common issues at this stage: the agent proposes to pull all deals instead of just closed-won, or it formats the Slack message as a paragraph instead of bullets. If the plan is wrong, edit the goal statement to be more specific. If the plan is right, approve it and set recurrence to weekly.
Minute 7 to 8: Activate
Click "Activate." The agent is now live and scheduled. You will receive a notification in your ChatGPT Enterprise inbox when the first run completes, along with a full audit log showing every tool call the agent made, every piece of data it retrieved, and the exact Slack message it sent. Review that audit log after the first run before making any changes. The audit log is your primary feedback mechanism for the first two weeks.
What good looks like after 7 days
Zero manual effort on Monday morning recap. The Slack message is already there when the sales team opens their phones. No one had to remember to do it. No one had to pull data the night before. The meeting starts with everyone already looking at the same summary.
Objections that used to live in scattered notes now appear in Slack with a clear pattern across deals. When a human writes a weekly recap, they tend to remember the most recent deals most vividly. The agent pulls every closed deal from the past 7 days with equal weight, which means it catches the objection that appeared in three deals but none of them were the biggest deal so they never made it into the manual recap.
You only get notified when something actually needs human judgment. The agent handles the routine. It escalates the exception. A $200,000 deal with an unusual objection you haven't seen before gets flagged. The standard Monday recap runs silently. That asymmetry is what makes the agent genuinely useful rather than just a differently-shaped inbox.
The one-line goal template (fill in the blanks)
Every [DAY] at [TIME], pull [DATA SOURCE], [ACTION], and post [OUTPUT FORMAT] in [CHANNEL/TOOL].Use this template for your first five agents. It forces clarity on all four structural elements before the agent runs. Once you have built five agents with this structure and understand how the agent interprets each element, you can start writing more complex multi-step goal statements. But for your first week, one line, four elements, one agent.
Common first-week mistakes (and how to avoid them)
Goal too vague. "Help with sales" gives the agent nothing to act on. The agent needs a schedule, a data source, a transformation, and a destination. If your goal statement does not include all four, it is too vague. "Pull closed-won deals and summarize objections" succeeds because it names the data and the transformation, even if the schedule and destination are still implied.
No approval gate on the first run. Start with "Always ask before posting to Slack" for the first two weeks of any new agent. This gives you a review step before anything goes to a shared channel. Once you have confirmed the output is clean for three consecutive runs, switch to auto-approve for internal-only channels.
No success metric in the goal. Add a line like "Success = 100% of closed deals from the past 7 days captured, zero false positives, posted by 8:15am" to your goal statement. The agent uses this metric in its self-audit. Without it, the agent has no way to evaluate whether it is doing its job correctly, which means you will need to evaluate manually.
Challenge: Create your first agent before you finish reading this. The 8-minute timer starts now. Pick the most boring, repetitive data task your team does on a fixed schedule. Write one goal line using the template above. Activate. Your future self and your team will thank you by Friday.