OpenAI Workspace Agents Now Plug Into Salesforce and Slack: The 4-Question Audit Sales Ops Should Run Before July 6

ChatGPT just stopped being a solo chat window for your reps and became a shared team artifact that plugs directly into Salesforce and Slack. The meter starts July 6, 2026.

That's the short version of what OpenAI shipped this week with Workspace Agents in ChatGPT. And for Sales Ops teams that built shared workflows on custom GPTs, the implications aren't abstract. You've got a migration decision, a cost decision, and a governance decision to make before a hard deadline.

What OpenAI Just Shipped

According to VentureBeat's coverage of the launch, OpenAI's Workspace Agents are positioned as the successor to custom GPTs, built from the ground up for shared team use rather than individual chat sessions. They're available now in research preview.

The plan eligibility is narrow: ChatGPT Business, Enterprise, Education, and Teachers. If your organization is on one of those tiers, you can start building and testing today. Free and Plus users don't have access in this initial wave.

Under the hood, Workspace Agents run on Codex, OpenAI's code-capable model. That's the same model family behind GitHub Copilot's most capable features, which tells you something about what OpenAI envisions these agents doing: not just answering questions, but executing structured workflows. Sales Development Representatives (SDRs) asking an agent to pull account history from Salesforce, Account Executives (AEs) asking it to log a call summary automatically -- those are the use cases Codex-backed execution is designed to handle.

OpenAI's own introduction post confirms that first-class connectors for Slack and Salesforce ship with the launch. No custom API work, no middleware. Admins wire the connector, and the agent has access. That's a meaningful difference from the workarounds teams were building with custom GPTs.

The admin controls are worth noting too. Admins can now define exactly who can build a Workspace Agent, who can share one across the org, and who can invoke one. That governance layer didn't exist with custom GPTs in any meaningful way.

On pricing: the free window was originally set to close May 6. OpenAI extended it. The new cutover is July 6, 2026, and that's when credit-based pricing kicks in. Per-credit rates haven't been fully published yet, but they're expected before the deadline.

Reworked.co's coverage of the announcement also highlights the roadmap items OpenAI has signaled: event-triggered agents (so a new Salesforce opportunity could auto-kick off an agent workflow), usage dashboards for admins, broader tool access, and tighter integration with Codex apps.

Key Facts

• Free through July 6, 2026, then credit-based pricing -- OpenAI Help Center
• Powered by Codex -- OpenAI
• Plan eligibility: ChatGPT Business, Enterprise, Education, Teachers -- OpenAI

Why This Is Different From the Enterprise Agent Platform

Sales Ops teams with long memories will note that OpenAI shipped an Enterprise Agent Platform in February 2026. That announcement was about a developer building surface -- a way for your data science or IT team to construct custom agents using OpenAI's API and tooling. If you want context on that layer, our earlier coverage of the OpenAI Enterprise Agent Platform covers what RevOps needs to understand before your engineering team starts building.

Workspace Agents is a different thing. It's not a developer surface. It's a shared-team surface inside ChatGPT itself, the direct successor to the custom GPTs feature that millions of enterprise users have already been using. The key distinction is where the agent lives and who manages it.

Custom GPTs lived in an individual user's account. They could be shared, but governance was loose. Workspace Agents live at the organizational level, with admin controls on creation, sharing, and usage -- and they come with the Slack and Salesforce connectors built in. The ChatGPT workspace agents enterprise story is about making ChatGPT an organizational asset, not just a personal one.

That shift has direct implications for Sales Ops: teams that built shared GPTs to support reps now need to evaluate what migrates, what gets rebuilt, and what gets retired. And they need to do it before July 6.

The 4-Question Audit Sales Ops Should Run

This is the core of what we're calling the Workspace Agent Audit (WAA): four questions that give Sales Ops a structured way to evaluate the Workspace Agents launch before the pricing meter starts. The WAA isn't about feature excitement. It's about turning a vendor announcement into a concrete action list with an owner and a deadline.

Q1: Which workflows already live in shared GPTs that need to migrate?

Start with an inventory. Pull up your organization's current shared custom GPTs and map each one to its primary use case. Common examples: an "Account Brief" GPT that reps use before discovery calls, a "Proposal Draft" GPT that AEs trigger for mid-funnel deals, a "Call Summary" GPT that managers use to review rep activity.

For each one, ask two questions. First: does this workflow benefit from Salesforce or Slack connectivity? If yes, it's a migration candidate. An "Account Brief" agent that can actually pull the live Salesforce account record rather than relying on what the rep pastes in is a meaningfully better tool.

Second: is this workflow actively used, or is it a zombie? Custom GPT sprawl is real. Many organizations built a dozen GPTs in 2024 and 2025 that nobody opens anymore. The July 6 deadline is a good forcing function to deprecate the low-value ones before carrying them into a usage-billed system.

Q2: What is your Slack and Salesforce admin doing -- separately -- about the new connector?

This question surfaces a coordination risk that Sales Ops teams often miss. Workspace Agents have connectors for Slack and Salesforce, but those connectors require approval on the platform side, not just on the OpenAI side. Your Salesforce admin controls what objects and fields an OAuth-connected app can read and write. Your Slack admin controls which apps can post to which channels and which users can install workspace-level integrations.

If Sales Ops moves ahead with Workspace Agents without looping in both admins, one of two things happens: the agent is blocked at the connector level and doesn't work, or the agent is approved with overly broad permissions because nobody defined the scope in advance. Neither outcome is acceptable.

The right move is a joint planning session -- Sales Ops, Salesforce admin, Slack admin -- before any agent goes into production. Define the minimum permissions the agent needs, document them, and build the approval through your existing change management process.

Q3: What is your cost cap when credit-based pricing kicks in on July 6?

OpenAI hasn't published per-credit rates yet, but the pricing structure -- credits consumed per agent action -- is confirmed. That means Sales Ops needs to work backward from budget, not forward from usage.

Start with a rough model. How many agents do you expect to have active by July 6? How many times per day will each agent be invoked? Multiply by a conservative per-credit assumption (update it when rates are published), then add a 30% buffer for spikes. That number becomes your starting monthly cap request to finance.

The parallel is worth flagging: when HubSpot moved to outcome-based pricing for its AI agents, CROs who hadn't pre-budgeted the variable cost found themselves in mid-year budget reallocations. The HubSpot outcome-based pricing story is a useful reference. The OpenAI Codex sales workflow pricing shift will follow the same pattern -- early movers who define the cap have more control than those who react to a bill.

One additional consideration: who pays by default? If your Sales Ops team is building and deploying the agents, clarify now whether usage draws from a central IT credit pool or your department budget.

Q4: How will you measure agent value versus the SDR and AE time it replaces?

This is the question most teams skip, and it's the one that determines whether Workspace Agents survive their first quarterly business review. Without a measurement framework, you're left defending usage costs with anecdotes. With one, you can show the math.

The baseline measurement is straightforward: before deploying a Workspace Agent, time how long a rep currently spends on the task it's replacing. An SDR building an account brief manually might spend 12-15 minutes per account. If that SDR does 20 per week, the agent is replacing roughly 4-5 hours of work per SDR weekly.

The ROI comparison is then: credit cost per invocation multiplied by weekly invocations versus hours saved multiplied by fully-loaded rep cost per hour. If the agent costs $0.50 per run and the SDR runs it 20 times a week, that's $10 per SDR per week against 4-5 hours of recaptured time. That case is easy to make -- but only if you captured the baseline first.

Decide now what the "before" baseline is, who tracks it, and at what interval you'll run the comparison. That's the value-per-agent baseline that lets you cut low-performing agents before they accumulate credit spend.

A Small Table: Custom GPTs vs Workspace Agents

The Sales Ops Action List for the Next 30 Days

1. Inventory your shared custom GPTs. List every shared GPT your Sales Ops or RevOps team has built or blessed. Tag each as active, dormant, or zombie. Active ones get evaluated for migration. Dormant ones get a 30-day usage check. Zombies get archived.

2. Schedule a joint connector session with your Salesforce and Slack admins. Don't send a Slack message -- book a 30-minute call. Come with the specific objects and channels you want agents to access. Leave with a documented scope and an approval timeline.

3. Write a cost-cap policy. Even without final credit rates, document the process: who approves a new Workspace Agent, what usage threshold triggers a review, and who has authority to suspend an agent that's burning credits unexpectedly. This is easier to write before the meter starts than after.

4. Define a value-per-agent baseline for your first migration candidate. Pick one workflow -- the highest-frequency one -- and time it manually before you migrate it to a Workspace Agent. Document the baseline. That's your control group.

5. Build a deprecation calendar for low-value agents. Set a 60-day review date for every agent you deploy. At 60 days, if the agent isn't showing measurable value against its credit cost, it gets suspended. This prevents the sprawl problem from migrating forward from custom GPTs into the new system.

6. Monitor OpenAI's pricing page before July 6. When credit rates are published, update your cost model and revalidate the cap request with finance. Don't wait for the first bill to learn what the rates are.

FAQ

Do Workspace Agents replace shared GPTs?

Yes, functionally. OpenAI has positioned Workspace Agents as the successor to custom GPTs for team use cases. Existing custom GPTs won't disappear immediately, but OpenAI's investment is clearly moving toward Workspace Agents. Teams should plan to migrate their highest-value shared GPTs to the new system before the credit-based pricing transition.

Who pays for credit-based usage by default?

That depends on how your organization's ChatGPT Enterprise plan is structured. Usage typically draws from an organizational credit pool managed by the account admin. Sales Ops teams deploying agents should confirm with their IT or procurement contact whether agent usage is attributed to a central pool or a department-level budget, and set up usage visibility before July 6.

Can these agents write back to Salesforce?

Based on what's been announced, yes -- but the scope of write access depends on how your Salesforce admin configures the connector. Read and write permissions are set at the connected app level in Salesforce. Before you deploy an agent that logs call activities or updates opportunity stages, confirm the exact permissions with your Salesforce admin and document what the agent is allowed to touch.

Learn More

• OpenAI Enterprise Agent Platform: What RevOps Needs to Know Before Your Team Starts Building -- our February 2026 coverage of the developer-layer story
• Outreach's MCP Server Move Changes How You Build AI Sales Workflows -- the connector architecture context that makes Workspace Agents' Salesforce plug make more sense
• HubSpot Just Rewrote the AI Pricing Rulebook -- the credit-based pricing parallel for how this plays out at QBR
• Salesforce Put an AI Coworker in Every Search Bar -- how the competing Salesforce-native agent story affects your connector decision

If your team is on ChatGPT Business or Enterprise, the next move isn't reading more about Workspace Agents. It's opening your list of shared custom GPTs and starting the inventory. The July 6 deadline is real, the pricing shift is confirmed, and the teams that do this work now will have cleaner governance, a cost model, and a value baseline before the meter starts. The workspace agents credit-based pricing transition doesn't have to be a surprise.