OpenAI shipped something meaningful on April 22. Here's what it actually is, why the shift from chatbot to agent matters, and how to get started before pricing kicks in.
OpenAI has had a rough few months in the public narrative. Anthropic and Claude have been shipping reasoning improvements, tools, and model updates at a pace that's hard to match — and increasingly hard to ignore. Google Gemini's advantage is different: not just the model, but the distribution. Gmail, Docs, Search, Android — Gemini is already where hundreds of millions of people work, which is an infrastructure moat that no amount of product velocity closes overnight. The general sense in certain corners of the AI world is that ChatGPT had become the thing your parents use while the serious players moved on.
Then OpenAI shipped Workspace Agents on April 22nd. Whether or not it closes the gap, the product is worth paying attention to — particularly if you run a small business and you're still treating AI as a one-prompt-at-a-time tool.
Most people using ChatGPT today are using it as a very good answer engine. You type a question, you get a response. You paste a draft, it improves it. You describe a situation, it helps you think through it. That's genuinely useful. It's also the beginning, not the destination.
An agent is different in one fundamental way: it acts. It doesn't just respond to your prompt — it takes a sequence of steps, uses tools, makes decisions along the way, and completes a task from start to finish without you holding its hand through each step. The difference between asking someone a question and delegating something to them entirely.
That shift — from answer engine to autonomous actor — is what makes workspace agents worth a separate conversation. This isn't a new feature in the usual sense. It's a different category of use.
The short version: shared, persistent agents that your team builds once and uses together. They run in the cloud, connect to the tools you already use — Slack, Google Drive, Salesforce, Microsoft apps — and they can execute multi-step tasks on a schedule without someone manually kicking them off every time.
These aren't chatbots. They're not glorified autocomplete. A workspace agent can research, produce a document, check its own work, and follow a task from start to finish. The kind of task that, today, requires three interruptions, two handoffs, and a calendar reminder.
One practical note before you get too excited: Workspace Agents are available on ChatGPT Business, Enterprise, and Edu plans — not the free or Plus tiers. Business runs $20 per user per month. The feature is free to use until May 6, after which OpenAI moves to credit-based pricing. If you've been considering a team plan, the next week is a reasonable window to test before the meter starts running.
Think about the workflows that repeat in any business. New client onboarding. Weekly reporting. Follow-up sequences after a sale or event. Inventory summaries. Scheduling confirmations.
Every one of those is a candidate for a role-specific agent. You build it once, connect it to the tools you're already using, and it runs. The compounding effect is real — teams that have iterated on agents over multiple cycles report work that used to take days collapsing to hours. That's not a marketing claim. That's what happens when you stop rebuilding the same workflow from scratch every time.
If you're on a ChatGPT Business, Enterprise, or Edu plan, Workspace Agents are live now. Open ChatGPT and look for the Agents option in the left sidebar. Click it, describe a workflow your team runs regularly, and ChatGPT walks you through the build — connecting tools, setting a schedule or trigger, testing before you publish to the team.
The Slack integration is the most immediately useful at launch. Once deployed, your agent lives in a channel, responds to @mentions, and runs on a timer without anyone opening ChatGPT. A common pattern: create a dedicated channel, deploy the agent there, and let teammates interact with it directly without switching tools.
Start with one workflow, low stakes. Admin tasks, scheduling, basic data pulls work well first. Not your most complex process, and not anything customer-facing until you've tested it thoroughly. Connect only what you need — agents inherit the permissions you give them, so scope that carefully from the start.
If you're not on a Business plan yet, the free trial window closes May 6. That's your low-friction moment to test before pricing kicks in.
This is where a lot of small businesses will make mistakes, and the mistakes are worth naming clearly.
Agents get access. When you connect an agent to your Google Drive or your CRM, you are extending permissions into an automated system. Decide in advance who can build agents, what data they can touch, and what actions require a human to approve before they execute. That governance layer is not optional — it's what separates useful automation from an expensive liability.
Agents also require maintenance. The knowledge sources that power them need to be updated as your business changes. A one-and-done build will degrade. Budget time for that, not just money.
And on pricing: the credit-based model starting May 6 can be unpredictable at scale. High-frequency agents running multiple times a day accumulate costs quickly. Start with centralized development — one person owns the build — so you can track what you're spending before it compounds.
Most small businesses are still at the "I use ChatGPT sometimes" stage. That was the right first step. But workspace agents represent a meaningful jump — from answer engine to operational infrastructure. From tool you use to system that runs.
The question is not whether this technology is relevant to how your business operates. It almost certainly is. The question is whether you treat the next 90 days as a real implementation window or keep it on the "we're watching this" list while competitors quietly rebuild their back office around agents.
OpenAI isn't out of the race. And the businesses that wrote them off — along with this product — may have done so a few months too soon.