Agents in ChatGPT for Work: How Businesses Can Automate Recurring Workflows
Written by Mike Pearlstein, CISSP, CEO of Fusion Computing Limited. Helping Canadian businesses build and manage secure IT infrastructure since 2012 across Toronto, Hamilton, and Metro Vancouver.
Most people still use ChatGPT the way they used Google in 2005, one question at a time, then back to the inbox. The repetitive work that actually eats the week, the morning email triage, the Monday briefing, the weekly status roll-up, never makes it into the prompt.
ChatGPT Agents are built for that second category. This newer mode of ChatGPT can read email, browse websites, work with uploaded files, fill forms, and edit spreadsheets on its own virtual computer. The user stays in control of every action that has consequences (OpenAI, Introducing ChatGPT agent).
This article explains what the new mode is, where it helps a Canadian SMB the most, what a good workflow actually looks like, and where the real risks are. It also explains why an MSP belongs in the room when a business decides to roll the technology out.
Key Takeaways
- ChatGPT Agents are workflow assistants, not chatbots. The system can navigate websites, read uploaded files, connect to email and document repositories, fill forms, and edit spreadsheets while pausing for confirmation when needed (OpenAI, 2026).
- Workspace agents are the enterprise wrapper. They let teams build agents for repeatable tasks, share them inside a workspace, run them in Slack, and put them on a schedule (OpenAI Help Center).
- Scheduled tasks can run one-off or recurring, with a hard cap of 10 active tasks at a time, on every plan except Pro models (OpenAI, Tasks in ChatGPT).
- Agent mode is off by default for ChatGPT Enterprise. A workspace owner has to enable it, control which apps it can reach, and assign it through role-based access control.
- The risks worth taking seriously are permissions, confidential data exposure, hallucinations, over-automation, auditability, and Canadian privacy obligations under PIPEDA, Quebec’s Law 25, and the proposed Bill C-8.
What Are ChatGPT Agents?
Tenant hygiene: ChatGPT Agents and Copilot agents both read whatever the user account already accesses. The Pre-Copilot SharePoint Audit covers the SharePoint side of that exposure for Microsoft tenants.
OpenAI uses three closely-related terms. ChatGPT agent is the mode you turn on inside a regular ChatGPT conversation, by selecting it from the tools menu or typing /agent in the composer. Once active, it carries out multi-step tasks on its own virtual computer. The system uses screenshots of a virtual browser to click buttons, fill forms, and work across websites the way a person would (OpenAI, ChatGPT agent).
Workspace agents are the enterprise and business wrapper around that capability. A workspace owner builds an agent for a repeatable task, tests it before publishing, connects it to apps such as Google Calendar, Google Drive, Slack, and SharePoint, and shares it with the team. The same agent can run inside a Slack channel and execute on a schedule (OpenAI, Workspace agents).
Scheduled tasks are the recurring layer. ChatGPT can create a task for itself to run later, either at a set time, on a recurring cadence, or via API. Tasks execute whether or not the user is online, and there is a hard limit of 10 active tasks per account (OpenAI, Tasks in ChatGPT).
According to OpenAI’s 2026 Tasks in ChatGPT documentation, scheduled tasks support up to 10 active tasks per account at any time and execute whether the user is online or not. The 10-task ceiling matters because it forces an explicit prioritisation conversation. For most Canadian SMBs, that constraint is a feature, not a limit.
The practical takeaway is that an “agent” in business use is usually a stack of all three. A workflow is defined once, connected to a few trusted data sources, and scheduled to run on a predictable rhythm.
Why ChatGPT Agents Matter for Businesses
The everyday business problem isn’t a shortage of intelligence. It is a flood of small information chores. A typical SMB owner reads four inboxes before lunch, then two CRMs, a project tool, three news feeds, and a folder of PDFs. None of it is high-value work. Skipping it has consequences anyway: missed follow-ups, stale numbers in the Monday meeting, the deal that quietly went cold.
This is where AI workflow automation pays for itself. The gather-summarize-organize-output cycle collapses into something that runs on its own and lands a consistent artefact in the same place at the same time. The value isn’t replacing human judgement. It is reducing the repetitive information work that crowds the judgement out.
If you want a starting point for thinking about where AI fits in your operations, our AI strategy guide for Canadian SMBs walks through the prioritisation framework we use with clients.
Not sure which workflow to automate first? Book a free 30-minute AI readiness assessment →
FIELD NOTE FROM MIKE
The first agent workflow I shipped for a client was a Friday-afternoon weekly operations report. The client was a 28-person Toronto professional services firm, and the engagement ran in March 2026. Their controller had been spending roughly 90 minutes every Friday assembling the same five spreadsheet pulls and writing the same paragraph of commentary.
We scoped a workflow against three named SharePoint files and one Power BI export, defined a fixed Markdown template, and added a human review step before the email went out.
Three things surprised me. The first run produced the right artefact in 4 minutes. The second run drifted because one of the source files had been renamed. We had to add an explicit file-path-pinning instruction to the prompt. By week six, the controller was spending around 12 minutes on review instead of 90 on assembly. The surface area she could not previously cover, variance commentary against a rolling 12-month average, finally got attention.
Anonymized FC engagement, March 2026. Numbers reflect the engagement; client identifiers changed.
Practical Examples of ChatGPT Agent Workflows
According to OpenAI’s 2026 launch announcement, ChatGPT agent is the first time the product can take actions on the web. Useful first prompts include “brief me on client meetings based on recent news” and “analyze three competitors and create a slide deck.” Most workflows for a Canadian SMB fall into one of six patterns.
- Daily email briefing. The agent reads the previous 18 hours of inbox traffic and groups messages by sender and thread. It surfaces anything that mentions a deadline or invoice, then posts a short summary to a Slack channel before 8:30 AM.
- PDF and document review. Drop a vendor contract or policy document into the workflow. The agent extracts the renewal date, payment terms, indemnification clauses, and any non-standard language, and writes a one-page memo for review.
- News and social media monitoring. Once a day, the agent checks named sources for mentions of the company, top three competitors, and three industry topics, and produces a ranked briefing.
- Meeting preparation. Before a sales or client meeting, the agent reads the calendar invite, pulls public information about the company and the people attending, references the last interaction in the CRM, and drafts a one-page brief.
- Weekly operations report. On Friday afternoon, the agent collects data from named spreadsheets and dashboards, fills in a template, and emails it to the leadership team. Numbers stay in the source of truth; only the narrative is generated.
- Sales follow-up tracking. The agent reviews the pipeline and flags deals that have gone quiet against an agreed cadence, with a suggested next action for each one.
According to OpenAI’s 2026 launch announcement for ChatGPT agent, the system can browse websites, work with uploaded files, connect to email and document repositories, fill forms, and edit spreadsheets on its own virtual computer. That breadth is what unlocks the six workflow patterns above. For a Canadian SMB, the gain is being able to point one tool at work that already crosses six.
Book a Free IT & AI Readiness Assessment
What Makes a Good AI Agent Workflow?
According to OpenAI’s 2026 documentation, agent mode “will pause for clarification or confirmation when needed” and “can be guided or interrupted mid-task.” That control surface is what separates workflows that survive contact with real operations from one-shot prompts. A reliable workflow has six ingredients below. Skip any one and it either drifts, leaks data, or quietly stops being useful.
- A clear trigger. Either a time (08:00 every weekday) or an event (a file lands in a folder, a Slack mention, an inbound webhook). Vague triggers produce vague output.
- Trusted data sources. The agent should pull from named, sanctioned places: a specific shared drive, a specific calendar, a specific CRM view. “The whole inbox” is rarely the right answer.
- Specific instructions. What to include, what to ignore, what tone to use, what format to output. Most failed agent rollouts trace back to a one-line prompt that worked once and was never refined.
- A defined output. A specific channel, a specific document template, a specific email format. Consistency is what makes the workflow useful three months in.
- Human review. An identified person who reviews the agent’s output before it goes anywhere external, and a fallback path when the agent is uncertain.
- Security boundaries. What apps it can read, what it cannot, what it is allowed to write to, who can run it. These are configured at the workspace level for Enterprise and Business plans.
Where Businesses Need to Be Careful
According to OpenAI’s 2026 ChatGPT agent documentation, agent mode is off by default for ChatGPT Enterprise workspaces, requires admin enablement, and can be assigned through role-based access control. Workspace owners also control which connector apps the agent can reach. The default-off posture is the cleanest signal that this is meant to be governed before turning on, not after.
Agents extend the reach of any account they run under. That is the productivity gain and the risk in the same sentence. The categories that matter to a Canadian SMB are below.
Permissions. An agent inherits the permissions of the user who configured it, plus whatever connectors it has been granted. OpenAI explicitly recommends “extra caution when using apps with ChatGPT agent to prevent unauthorized access to sensitive information” (OpenAI, 2026).
For Enterprise, agent mode is off by default, app access is controlled centrally, and access can be assigned through role-based access control. The accompanying ChatGPT agent system card documents the safety evaluation work behind these defaults.
Confidential data. Personal information, client matter data, financial records, and HR files should be classified before any agent is allowed near them. PIPEDA and Quebec’s Law 25 both require organizations to know what they collect, why they collect it, and where it goes. The proposed Bill C-8 raises the bar on automated decision-making transparency.
Hallucinations and verification. Agents cite sources and screenshots when they browse, but they still produce confident text that turns out to be wrong. Any output that drives a financial, legal, or customer-facing decision needs human review, not just a glance.
Over-automation. Not every recurring task should run unattended. The cost of a wrong invoice, a wrong tax filing, or a wrong client communication is much higher than the cost of someone reading the email before it sends.
Auditability. Conversations involving agent tasks appear in OpenAI’s Compliance API logs for Enterprise plans, but individual agent actions such as virtual computer usage and chain of thought are not logged the same way. Your own audit trail in the destination systems matters.
User training. The biggest risk to a sensible AI policy is a confident user who doesn’t know what consumer ChatGPT does with their prompts. Training is part of the rollout, not an afterthought.
Why IT and Cybersecurity Still Matter
According to IBM’s 2025 Cost of a Data Breach Report, one in five organizations experienced a shadow-AI-linked breach in 2025. Most lacked proper AI access controls. The fastest way to break an AI rollout is to let it happen one user at a time, on personal accounts, with no record of which workflow is connected to what.
An MSP earns its keep at four points in the agent lifecycle. First, identifying which workflows are worth automating and which are not. Second, reviewing security and permissions before the connectors are turned on. Third, integrating ChatGPT Agents safely with Microsoft 365, SharePoint, Teams, and email so the data classification and DLP rules already in place still apply.
The fourth point is helping users adopt the tooling without creating a parallel set of unsanctioned accounts. Fusion Computing does this work as part of a broader AI services and managed IT engagement, with cybersecurity reviewing the perimeter on the way in.
Getting Started With ChatGPT Agents
The lightest-touch way to start is one workflow on one team, scoped tightly, on a paid plan with proper governance. The five-step approach we use with clients is below.
- Pick one recurring workflow. Something that happens at least weekly, takes more than 30 minutes, and produces a written artefact. Daily email briefings and weekly ops reports are usually the safest first picks.
- Define the data sources. Name the inbox, folder, calendar, or spreadsheet. Avoid “everything I have access to” as a starting scope.
- Define the output. Where it goes, in what format, to whom. A consistent destination is what makes the habit stick.
- Add human approval. For the first 30 days, every output gets reviewed by a named person before it goes anywhere external. After 30 days, decide what can run unattended.
- Review security and improve. Walk through the connectors and permissions with IT or your MSP. Tighten where you can. Then iterate the prompt and the output template based on what you learn.
Already running an AI tool stack and want a second opinion on governance? Our AI readiness assessment walks through tools, data classes, and policy in a single working session.
According to OpenAI’s 2026 workspace agents documentation, a workspace agent can be built once and tested before publishing. It connects to apps including Microsoft 365, SharePoint, Slack, and Google Drive, runs inside a Slack channel, and operates on a fixed schedule. That architecture pattern turns a one-off prompt into a repeatable business workflow with audit trails attached.
Final Thoughts
ChatGPT Agents are workflow assistants, not magic automation. They handle the gathering and summarizing work that fills the calendar without filling the strategy. Pick the workflow first, scope the data, set up governance, and let an experienced partner help you tune what comes out the other side.
Fusion Computing helps Canadian businesses adopt ChatGPT Agents and other AI tools safely.
Frequently Asked Questions
What are ChatGPT Agents?
ChatGPT Agents are an operating mode in ChatGPT that can carry out multi-step tasks on a virtual computer. They navigate websites, read uploaded files, connect to data sources such as email and document repositories, fill forms, and edit spreadsheets, while pausing for confirmation when an action has consequences. Agent mode is available on Pro, Plus, Business, Enterprise, and Edu plans (OpenAI, 2026).
Are ChatGPT Agents safe for business use?
They can be, with the right configuration. For Enterprise, agent mode is off by default, app access is controlled centrally, role-based access control applies, and conversations appear in Compliance API logs. They are not automatically safe for personal accounts running unsanctioned. The risk profile depends entirely on which connectors and data sources the agent is allowed to reach.
Can ChatGPT Agents summarize emails and documents?
Yes. With the appropriate connectors enabled, an agent can read recent email, summarize threads by topic, extract action items, and produce a daily briefing. The same capability applies to PDFs, Word documents, and SharePoint files, provided the user or workspace owner has granted access to those sources.
Do ChatGPT Agents replace traditional automation tools?
No. They sit alongside them. For deterministic, high-volume, rule-based work such as accounts payable matching or ETL, a purpose-built automation platform is still the right answer. ChatGPT Agents shine where the work involves reading unstructured information, making judgement calls, and producing a written output that varies each time.
What is the best first workflow to automate?
For most Canadian SMBs, a daily email briefing or a weekly operations report. Both happen on a predictable cadence, both produce a written artefact, both are easy to review for accuracy, and both have low downside if the agent gets something wrong on the first run. Sales follow-up tracking is a strong second choice once the first workflow is stable.
Do businesses need an AI policy before using agents?
Yes. A short, written acceptable use policy that lists approved tools, classifies data that may be entered into them, and requires annual training is becoming standard in Canadian cyber insurance renewals. PIPEDA and Quebec’s Law 25 already apply, and the proposed Bill C-8 raises the bar further. See our guide on what should be in an AI acceptable use policy for a working template.
Which ChatGPT plans support Agents?
Agent mode is available on ChatGPT Pro, Plus, Business, Enterprise, and Edu plans. Monthly message limits vary by tier: Plus 40, Pro 400, Business and Enterprise 40 per user. Flexible-pricing tenants meter at roughly 30 credits per message. Agent mode is not available on the free plan (OpenAI, 2026).
How many recurring agent tasks can run at once?
OpenAI caps active scheduled tasks at 10 per account at any given time. If you reach that ceiling, the system will not create a new task until you pause or delete an existing active task, or it completes per its scheduled time. All recurring tasks can be reviewed at chatgpt.com/schedules (OpenAI, Tasks in ChatGPT).
How can Fusion Computing help?
We help Canadian SMBs identify the right workflows to automate, review security and permissions, integrate ChatGPT Agents with Microsoft 365 and SharePoint safely, and train teams on responsible use. The work is part of a broader managed IT, cybersecurity, and AI services engagement. Book a Consultation to start.
