OpenAI shipped Workspace Agents on Tuesday afternoon and the SEO landscape still hasn’t caught up. The keyword chatgpt agent has been running at 22,200 searches a month for six months, but the long-tail around Workspace Agents is brand new — and for the next 48 hours, every single one of those searchers will find two kinds of content: OpenAI’s own launch post, and tech-press recap articles.
No tutorials. No “here’s the one I built for my sales team.” No honest reviews.
This is the tutorial.
We’re going to build a single Workspace Agent that runs on a schedule, reads your Salesforce pipeline and your Gmail on Monday morning, generates a weekly sales report with named deals at risk and call-outs by rep, and posts it into your team Slack channel at 9 AM. Setup takes 10 minutes. It replaces the five to six hours a week that most sales teams spend pulling this together by hand — the exact number OpenAI’s own launch-day cookbook calls out for a related workflow.
The kicker: it’s free until May 6, 2026. After that, credit-based pricing kicks in. Build it now, run it for two weeks, and your company’s procurement team will have a hard time saying no when the bill comes.
What Workspace Agents actually is
Think of it this way. A Custom GPT is a ChatGPT prompt with a saved personality — you talk to it, it talks back. A Workspace Agent is a Custom GPT that goes and does the thing. It reads calendars. It queries Salesforce. It posts to Slack. It drafts emails and schedules meetings. It runs on a timer. It remembers what it did last week.
Under the hood, Workspace Agents run on Codex — OpenAI’s autonomous-work model family, the same tech that powers ChatGPT’s coding agent. The reasoning loop is a plan-act-approve cycle: you describe the outcome, the agent plans the steps, it executes them one at a time, and it pauses for your approval before anything leaves your company (sending an email, posting to Slack, updating a CRM record). You see every step in the conversation view.
OpenAI is calling Workspace Agents the successor to Custom GPTs. That’s not marketing; it’s a planned migration. Here’s the roadmap they published on launch day:
- Individuals keep Custom GPTs for the foreseeable future — no change today.
- Organizations (Business, Enterprise, Edu, Teachers) will be required to convert Custom GPTs to Workspace Agents at a yet-to-be-determined date.
- A conversion tool is coming — Custom GPT → Workspace Agent, probably one-click.
- Until then, both formats coexist.
If your team already has a library of Custom GPTs, nothing breaks today. But the writing is on the wall: the format you should be authoring in, as of this week, is Workspace Agents.
What’s inside the box
Right now, Workspace Agents plug into the tools most sales teams actually use:
- CRMs: Salesforce, HubSpot
- Messaging: Slack, Microsoft Teams
- Docs: Google Drive, Gmail, Google Calendar, Microsoft 365 apps, Notion, SharePoint
- Sales-specific: Gong (call transcripts), Atlassian Rovo
- Utilities: web search, code execution, file handling
Memory is persistent. The agent keeps a per-user folder where it stores the running context — last week’s report, the named accounts, the exec’s preferences. You don’t re-explain what “our ICP” means every Monday.
Three templates shipped on day one. Remember these names; they’ll show up when you open the builder:
- Weekly Metrics Reporter — pulls data from your stack, generates charts, shares reports on a schedule. This is the one we’re building today.
- Software Reviewer — reviews employee software requests and files IT tickets.
- Product Feedback Router — monitors Slack channels and support tickets, tags and routes feedback.
The OpenAI cookbook also published a Sales Meeting Prep agent as a same-day bonus template. Its description says the workflow previously took sales reps “five to six hours per week.”
The 10-minute setup
We’re building the Weekly Sales Report agent. Before you start, you’ll need:
- A ChatGPT Business, Enterprise, Edu, or Teachers account with admin access
- Salesforce credentials (read-only to your pipeline is fine)
- Access to your team’s Slack channel and the ability to install apps
- A SharePoint or Google Drive folder to store the generated briefs
Total time: about 10 minutes. Promise.
Step 1: Open the Workspace Agent builder
Sign in to ChatGPT. Click Agents in the left sidebar, then Build Agent. You’ll land in a conversational builder — you describe what you want, it drafts the agent configuration.
Step 2: Describe the workflow in one paragraph
Type this into the builder (adjust team names to yours):
“Every Monday at 9 AM, pull all open opportunities in Salesforce where the close date is within the next 30 days. Flag any where the last activity was more than 10 days ago. Pull call summaries from Gong for those accounts from the past week. Write a 400-word report grouped by deal stage: new, advancing, at-risk. For each at-risk deal, name the account, the rep, the last-activity date, and one concrete next step. Save the report to /SharePoint/Sales/Weekly-Reports/ with filename ‘weekly-sales-YYYY-MM-DD.docx’. Post a 3-bullet summary to the #sales-leads Slack channel with a link to the full doc.”
The builder generates a draft agent. It names it something reasonable like “Weekly Sales Report.” You can rename it.
Step 3: Connect your apps
The builder surfaces a list of apps the agent will need. You’ll click to authenticate each one:
- Salesforce: click Connect, sign in, approve the read permissions. Pick “My org” (not “My account”) so the agent sees team-wide pipeline.
- Gong: same flow. Grant read access to call summaries.
- SharePoint: connect with a service account (shared team identity) if you can — not your personal account. This lets the agent write to the team folder regardless of who triggers the run.
- Slack: install the ChatGPT agent app in your workspace, then pick the channel it’s allowed to post to. The agent will only post to channels you grant.
OpenAI’s design choice here is that personal data (calendar, email) authenticates per-user, while shared data (CRM, Slack, team folders) uses a service account. That separation is important: if Sally triggers the agent, it reads Sally’s email but writes to the shared team folder. Different permission model for each category.
Step 4: Add a “skill” for consistent output
Skills are reusable instructions the agent loads when it needs to do a specific type of task. You can upload them as Markdown files or build them inline in the builder.
For the sales report, create a skill called “Weekly Sales Report Format” with this content:
# Weekly Sales Report Format
When generating the weekly sales report, follow this structure exactly:
## Headline (1 sentence)
"Pipeline moved $X this week across Y deals. Z at-risk."
## New Deals (bullet list)
- Account Name — Rep — $Amount — Stage entered YYYY-MM-DD
## Advancing Deals (bullet list)
- Account Name — Rep — $Amount — Previous stage → Current stage
## At-Risk Deals (bullet list)
- Account Name — Rep — Last activity YYYY-MM-DD — One concrete next step
## Call-Outs
- Name 1 rep who had an unusually strong week, based on Gong call sentiment.
- Name 1 rep who may need help, based on at-risk deal count.
- Note any deals over $100K that haven't had activity in 14+ days.
Total word count: 300–500 words. No filler.
Never fabricate numbers — if Salesforce data is incomplete, flag that explicitly.
The “never fabricate numbers” line is important — especially on GPT-5.5, which has a documented high hallucination rate on factual recall. Telling the model explicitly to flag gaps instead of filling them in is the simplest way to keep the report trustworthy.
Step 5: Schedule it
In the agent settings, set the schedule to Monday at 9 AM local time, recurring weekly. Pick your time zone carefully if your team spans multiple.
Step 6: Test it once before it goes live
In the conversation view, type “run now.” The agent starts executing. Watch the action trace — every step shows in the side panel: connecting to Salesforce → querying open opportunities → filtering by close date → querying Gong → composing report → saving to SharePoint → posting to Slack.
Before the Slack post and the SharePoint write, the agent pauses. The button says Post for Slack, Save for SharePoint. Each has a risk indicator. Click through both to approve the first run, then inspect the output.
If the output is wrong, tell the agent what’s off (“the headline is too generic — pull a specific dollar amount from the new deals”) and it redrafts. Repeat until the template is right. Once the Monday run produces what you want, toggle the “skip similar approvals” dropdown for low-risk actions (the SharePoint write is low-risk; the Slack post you might want to keep as manual approval for the first month).
Step 7: Share the agent with your team
In the agent’s Sharing settings, generate a workspace share link. Send it to your sales leaders. Each person who accepts it authenticates their own Salesforce / Gmail / Gong credentials — they only see their own data inside the shared agent. The SharePoint service account stays shared.
That’s the whole build. Ten minutes if you move fast.
Three more agents a sales org should have this quarter
Once the Weekly Sales Report is running, these three are straightforward extensions. All use the same integration stack.
Daily Pipeline Standup Digest. Runs every weekday at 8 AM. For each rep, it lists the 3 deals they should touch today (ranked by close-date urgency and last-activity recency), generates a draft follow-up email for each, and posts the digest privately to each rep in Slack. Saves 20 minutes of CRM-squinting per rep per day.
Deal Stage Push Notifier. Triggers on a Salesforce webhook — any deal sitting in the same stage for more than 14 days. The agent pulls the last three call summaries from Gong, drafts a “here’s why this deal is stuck and three moves to unstick it” note, and sends it to the rep + their manager. Catches the slip-through-the-cracks deals before they die.
QBR Prep. Runs on the 28th of every month. Pulls closed-won and closed-lost deals for the quarter, segment revenue by rep and product line, generates a 10-slide deck in PowerPoint via the agent’s file tools, and emails drafts to each rep’s manager for review. Replaces the 2-day “let’s build the QBR deck” scramble that happens at the end of every quarter.
All three are similar enough to the main Weekly Report agent that you can clone it, change the prompt, and ship.
What Workspace Agents can’t do (yet)
It can’t see your Slack history beyond the last 30 days — the connector pulls recent messages, not long-history archives. If your team has internal-knowledge Slack that’s 3 years deep, the agent won’t learn from it.
It doesn’t run fully unattended. Every outbound action (send email, post to Slack, update a record) pauses for approval on the first run. You can toggle “skip approval for similar actions” per type, but the default is cautious. This is good for safety, slightly annoying for pure automation. If you’re picturing a fully autonomous agent sending cold emails overnight, that’s not what ships today. A human still clicks Send on anything external.
The free tier has a ceiling. Workspace Agents are free for Business/Enterprise/Edu/Teachers accounts until May 6, 2026. After that, credit-based pricing begins — OpenAI hasn’t published the exact numbers yet. Factor that into your procurement conversation. Most orgs will end up somewhere in the $30–$150 per user per month range once it stabilizes.
Integrations are North America / English-heavy. Salesforce, HubSpot, Slack, Microsoft, Google — all first-class. Pipedrive, Monday, ActiveCampaign, Freshsales — not yet. If your CRM is outside the supported list, you’ll either be waiting or building a webhook bridge yourself.
Memory isn’t infinite. Each user’s agent memory is a scoped folder. Long-running agents that try to remember every meeting for two years will run into context limits. Good agents archive their own memory — you can tell yours “every Friday, summarize this week’s context into a single note and drop older per-day notes.” The builder supports this pattern natively.
Workspace Agent vs Custom GPT — the migration conversation
Your company probably has 20 or 30 Custom GPTs floating around. Here’s what to tell your team about those.
| Custom GPT | Workspace Agent | |
|---|---|---|
| What it is | A saved prompt with a name | An agent that acts across your tool stack |
| Can it send emails / update CRM? | No — only chat | Yes, with approval |
| Schedules and recurring runs? | No | Yes |
| Memory across sessions? | Limited | Persistent, per-user |
| Integration connectors? | Actions via OpenAPI only | Pre-built for Slack, Salesforce, Gmail, Gong, etc. |
| Future of format | Deprecating for orgs (date TBD) | Standard going forward |
| Conversion path | One-click tool coming | — |
For individual users on Plus or Pro: nothing changes today. Keep your Custom GPTs.
For Business/Enterprise orgs: start authoring new work as Workspace Agents, not Custom GPTs. Keep your old Custom GPTs running until the conversion tool ships, then batch-convert. Don’t invest time building net-new Custom GPTs in 2026.
What this means for you
If you’re an SDR or AE: Build the Daily Pipeline Standup Digest agent this week. Not the Weekly Report one — that’s a leader-tier view. The Daily Standup gives you your three most-important-to-touch deals every morning, with drafted follow-up copy, before your coffee kicks in. Saves 20 minutes and catches the deal you were about to forget.
If you’re a rev-ops lead or sales manager: Build the Weekly Sales Report this week and deploy it across your org. Two things will happen. First, your Monday morning staff meetings get shorter — everyone walks in already having read the same brief. Second, your org gets a real-time view of at-risk deals that used to live in your head or in someone’s spreadsheet. Use the two weeks of free access to prove ROI before the May 6 pricing change.
If you have a library of Custom GPTs: Don’t panic, don’t rebuild everything. Keep what works. But your next net-new agent should be a Workspace Agent. When the conversion tool ships (likely Q3 2026), batch-migrate. The format you’re authoring in as of this week matters for the next 3–5 years of your company’s AI workflows.
If you’re just now considering ChatGPT for your team: The Business tier at $20/user/month gets you Workspace Agents as of now. That’s meaningfully better ROI than it was a week ago — the same subscription now includes a platform for team automation, not just a better chatbot. Your IT director should know.
The bottom line: Workspace Agents aren’t a better Custom GPT. They’re the infrastructure layer a sales org builds workflows on. One agent per repeating task. Approval gates for the things that leave your company. Credit pricing after May 6. Start with one — the Weekly Sales Report — and see what moves.
Who should use it
- Sales leaders who keep re-running the same weekly pipeline-review process by hand
- SDRs and AEs drowning in account research before every customer call
- Customer Success teams routing support tickets, tracking QBR prep, chasing renewals
- Rev-ops and sales-engineering who already build internal tools and know where 5 hours of their week is wasted on data-munging
Skip Workspace Agents if: your company isn’t on ChatGPT Business yet (wait until you are), your CRM is niche (unsupported connector), or you’re one person with no team — Custom GPTs still do what you need.
The bottom line
OpenAI’s Workspace Agents launch is the week’s most interesting enterprise AI story, and the SERP hasn’t caught up yet. The keyword landscape for chatgpt agent, chatgpt agent mode, and chatgpt agent builder was built on the older Custom-GPT-era meaning, and the new meaning — “an agent that actually does the work” — hasn’t fully displaced it yet. The first real tutorial to rank for these keywords is going to own the cluster for a long time.
More importantly, the window to actually use them for free is 12 days. Your procurement team will not be able to block this at $0. They will have a harder time unblocking it at $150. Build your first agent today. Ship the Weekly Sales Report by Friday. Decide if it’s worth renewing by May 6.
Sources:
- Introducing workspace agents in ChatGPT — OpenAI
- Building workspace agents — sales meeting prep — OpenAI Cookbook
- OpenAI unveils Workspace Agents, a successor to custom GPTs — VentureBeat
- OpenAI updates ChatGPT with Codex-powered ‘workspace agents’ for teams — 9to5Mac
- OpenAI now lets teams make custom bots that can do work on their own — The Verge
- OpenAI Launches Autonomous Workspace Agents for ChatGPT — AIToolly
- OpenAI Workspace Agents Launch: Enterprise Buyer Guide — UC Today
- Beyond Chat: OpenAI Launches Workspace Agents — Android Headlines
- OpenAI Academy: Workspace Agents
- Workspace Agents in ChatGPT: Bold Leap From Chatbot to Team Automation — Kingy AI
