ChatGPT Workspace Agents Bring Codex to Knowledge Work

In Brief

OpenAI just launched Workspace Agents in ChatGPT, an evolution of custom GPTs that lets teams build shared agents to run long, multi-step workflows across Slack, Google Drive, Microsoft, and Salesforce.

Powered by Codex in the cloud, they keep working when users are offline, run on schedules, and live inside Slack channels.

The BIG news: any non-technical knowledge worker can build them.

What Happened

OpenAI introduced Workspace Agents in ChatGPT, positioning them as shared agents that handle complex tasks and long-running workflows across tools and teams. A team builds an agent once, uses it together inside ChatGPT or Slack, and the agent improves over time. The agents are powered by Codex running in the cloud, so they keep working when the user is offline. They also can run on a schedule,and be deployed directly into Slack channels.

OpenAI offers prebuilt templates for finance, sales, and marketing with out-of-the-box connections to Slack, Google Drive, Microsoft apps, and Salesforce. Workspace Agents are available as a research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans. They are off by default for enterprise workspaces pending admin enablement, free for the next couple of weeks, then move to a credit-based model. Governance includes role-based admin controls, required human approval for sensitive actions, a compliance API, and prompt-injection safeguards.

The launch happened in the same week Microsoft 365 Copilot's agentic capabilities went generally available and Google previewed an Agent Designer at Google Cloud Next. Three platforms, one direction, in a single week.

SmarterX founder and CEO Paul Roetzer broke down what all of this means on Episode 211 of The Artificial Intelligence Show.

The Key Numbers

$20 - Monthly per-person plan cost that currently gives ChatGPT users access to new agents. This could increase in the future.

100 - Human work hours Roetzer estimates a custom agent, such as a newly built customer success assistant, could potentially take over

Codex Inside ChatGPT Changes the Math

The templates show the workflow rewrite. Roetzer pulled up the agent browser in ChatGPT and read the prebuilt templates aloud. A Chief of Staff agent that prepares a daily operating brief from schedule, inbox, and team chat. A Data Analysis agent that sharpens the question, writes SQL, and builds visuals. A Sales Assistant for account intelligence and seller coaching. A Customer Support agent for ticket triage and case investigation.

"Every one of those, if we connect it to HubSpot, completely changes our workflows and potentially our staffing plans," says Roetzer. "If these things actually work, it completely evolves the way I think about how we're going to do our hiring this year."

Codex is the engine within ChatGPT. The detail buried in the announcement is that Workspace Agents are powered by Codex running in the cloud. Codex and Claude Code are the tools developers use to do agentic work with files, tools, and memory, far beyond what a chat prompt can do. OpenAI just took that engine and put it behind a builder interface inside the chat product millions of people already use.

"This is a different level," says Roetzer. "The ability to build agents for each role in the company really starts to change how I think about this because it's so easy to do. You could literally train anybody to do this. Even somebody who has been hesitant to do anything with AI."

Capability is here. Adoption is the bottleneck. Most people have never built a custom GPT, despite having had access for nearly three years. Workspace Agents are entering a market that has not fully exploited the previous generation of tools. Roetzer's response was to put a calendar invite on his team's schedule for an hour-long internal lab to figure out what the agents can actually do.

"It's pretty clear that by fall of this year, if not sooner, depending on which platform you're on, they're all going to enable a non-technical knowledge worker to build agents and run them," says Roetzer.

"We are clearly in the very early stages of not just the agent capabilities for the technical people and for development work, but now bringing that to knowledge work to make it as simple as building a GPT."

Paul Roetzer, founder and CEO of SmarterX, Episode 211 of The Artificial Intelligence Show

SmarterX Take

The right move for leaders now is to run an internal lab the way Roetzer is. Pick a role on your team, open the agent browser, build one agent against a real workflow, and see what breaks. The barrier is no longer technical capability or budget. It is whether your people try these tools while the templates are free and implementations are fresh.

The deeper shift is that agent-building is becoming a baseline knowledge worker skill, not a specialty. If OpenAI, Microsoft, and Google all enable non-technical employees to build agents by fall, every functional leader needs an answer for how their team uses them. That is a hiring question, a training question, and a workflow question all at once.

What to Watch

The credit-based pricing model. Workspace Agents are free for the next couple of weeks, then move to a credit system OpenAI has not yet disclosed. Pricing will determine how aggressively teams can experiment.

The platform convergence. OpenAI, Microsoft, and Google all moved on knowledge-worker agents the same week. Whichever platform builds the smoothest path from "browse template" to "agent running in production" is positioned to win with enterprises.

Further Reading

OpenAI: Introducing Workspace Agents in ChatGPT → openai.com

@OpenAI on X: Workspace Agents Announcement → x.com

Microsoft: Copilot's Agentic Capabilities Are Generally Available → microsoft.com

Heard on The Artificial Intelligence Show, Episode 211
Paul Roetzer and Mike Kaput discuss OpenAI's Workspace Agents and what Codex inside ChatGPT means for knowledge work. Listen →