How to create an agent in ChatGPT (and what to do when you need more)
Davis Christenhuis
ChatGPT has evolved far beyond answering questions. In 2026, you can create AI agents that automate research, draft content, analyze data, and complete multi-step tasks without constant supervision. These agents work with your instructions, remember context, and use tools like web browsing and code execution to get work done.
This guide shows you how to create agents in ChatGPT using both Custom GPTs and the newer Agent Mode, covers when these tools work well, and explains when teams need more capable enterprise platforms.
📌 TL;DR
Don't have time to read everything? Here are the key takeaways:
- ChatGPT offers two agent types: Custom GPTs are reusable chatbots you configure once. Agent Mode autonomously executes complex tasks like research and data analysis.
- Building one takes ~10 minutes: Write clear instructions, upload reference files, enable tools (web browsing, code execution), test, and publish — all inside ChatGPT's built-in builder. It requires a paid account.
- Works best for individuals: Great for personal productivity using public data. Share via link — updates publish to all users when you click “Update”.
- Hits limits for teams: Live data connectors limited to ~30 apps on Enterprise/Business plans, no per-agent data source controls, limited multi-agent orchestration.
- Enterprise platforms close the gap: Platforms like Dust connect to several business systems, support multi-agent workflows, and scale AI adoption across entire organizations.
What is a ChatGPT agent?
A ChatGPT agent is an autonomous AI system that can use tools, remember context, and execute multi-step tasks based on your instructions. Unlike standard ChatGPT conversations that require you to manually copy outputs and take action, agents can browse the web, run code, analyze files, and generate deliverables with minimal human intervention.
ChatGPT offers two types of agents. Custom GPTs are specialized chatbots you configure once and reuse for specific tasks like editing marketing copy or formatting code. Agent Mode represents OpenAI's newest capability, where ChatGPT can autonomously navigate websites, create spreadsheets, and complete complex research workflows using its own virtual computer.
💡 Want to see how AI agents work across a full team? Explore how Dust deploys agents at scale →
How to create an agent in ChatGPT (step-by-step)
Creating a ChatGPT agent requires a paid subscription and takes about 10 minutes to set up your first version.
Step 1: Access the GPT builder
Navigate to the GPT creation interface in ChatGPT by clicking "Create" from the GPTs section. The GPT builder interface shows two panels. The left side displays tabs where you configure your agent. The right side shows a live preview where you can test your agent as you build it.
Step 2: Define your agent's purpose
Start by telling ChatGPT what you want to build using conversational prompts. Be specific about the task, the audience, and the output format. Instead of "help with writing," say "You are a content editor for B2B SaaS blog posts. Review drafts for clarity, remove jargon, and suggest headlines under 60 characters."
The GPT builder will suggest a name, profile image, and conversation starters based on your description. You can accept these or refine them through follow-up instructions. More specific prompts create more useful agents.
Step 3: Configure instructions and knowledge
Switch to the Configure tab to add detailed instructions. This section defines how your agent behaves, what it should avoid, and how it formats responses. You can also upload knowledge files here—PDFs, documents, or data that your agent should reference when generating answers.
Knowledge files work best for static reference material like style guides, product documentation, or FAQs.
Step 4: Enable capabilities
ChatGPT agents can use several core tools. Web Search lets them find current information online. Canvas provides an interactive workspace for writing and coding projects. Image Generation creates visuals based on text descriptions.
Code Interpreter & Data Analysis allows them to run Python code, analyze datasets, and create charts. Actions let your GPT connect to external APIs and services. Define custom API endpoints using an OpenAPI schema, enabling your agent to query databases, pull live CRM data, or interact with third-party tools directly from the conversation.
Enable only the capabilities your agent needs. Each capability adds processing time, so start minimal and expand based on testing.
Step 5: Test and refine
Use the preview panel to test your agent with real queries. Check whether responses match your expectations for tone, format, and accuracy. If the agent generates responses that are too long, too short, or off-topic, return to the Configure tab and adjust your instructions.
Testing reveals edge cases your initial instructions didn't cover. Add specific guidance about how to handle ambiguous requests, format constraints, or situations where the agent should ask clarifying questions instead of making assumptions.
Step 6: Publish and share
When you're ready, publish your agent and choose who can access it. Options include "Only me" for personal use, "Anyone with a link" for selective sharing, or "Public" to list your agent in the GPT Store. Enterprise workspaces can restrict sharing to organization members only.
Once published, your Custom GPT appears in your ChatGPT sidebar. You can edit instructions, update knowledge files, or delete the agent at any time.
When to use ChatGPT agents vs. enterprise platforms
ChatGPT agents serve individuals well, but teams require different capabilities. Here's how they compare to enterprise platforms built for organizational use:
Feature | ChatGPT Agents | Enterprise Platforms |
Target audience | Individuals | Teams and organizations |
Data access | Manual file uploads, public web (Enterprise/Business) | Live company data (Slack, Notion, Salesforce, GitHub, etc.) |
Agent management | Individual creation, link sharing | Centralized deployment, agent versioning and centralized updates, team collaboration |
Multi-agent workflows | Limited | Full orchestration across specialized agents |
Model flexibility | OpenAI models only | Multiple providers (OpenAI, Anthropic, Google, Mistral) |
Permissions | Basic sharing controls | Role-based access, audit logs, compliance certifications |
Updates | Manual per agent | Centralized updates propagate across organization |
Best for | Personal productivity, experimentation | Enterprise workflows, team collaboration, regulated industries |
The core difference comes down to data and scale. ChatGPT agents work with information you manually provide, while enterprise platforms connect directly to your business systems.
When you need agents that access live company data, work together on complex workflows, and deploy across teams with proper governance, platforms like Dust fill that gap.
How Dust works for enterprise teams
Dust follows the same basic agent creation process as ChatGPT: define instructions, configure capabilities, and test. The difference is that agents connect to your live company data and deploy across your organization.
Live data connections instead of file uploads
Rather than manually uploading documents, Dust connects to your actual business systems:
- Notion
- Google Drive
- GitHub
- Confluence
- Salesforce
- HubSpot
- + Many other sources
When your team updates a document in Notion or closes a deal in Salesforce, Dust agents see those changes immediately. A sales agent can pull the latest product specs from your docs, pricing from your CRM, and past proposal language from Google Drive to draft RFP responses that stay current as your business evolves.
Multi-agent orchestration
Instead of building one agent that tries to do everything, Dust lets you create specialized agents for different functions and route work between them. A customer support flow might send simple questions to a knowledge base agent, technical issues to an engineering agent, and escalations to a human specialist.
Team deployment and governance
When you build an agent in Dust, you can deploy it across your organization with role-based permissions. An agent with access to financial data stays restricted to the finance team, while a general knowledge agent remains available company-wide. Audit logs track every interaction for compliance.
Model flexibility
Choose OpenAI, Anthropic Claude, Google Gemini, or Mistral models for each agent based on what works best for that task. Switch models without rebuilding agents or compare performance across providers.
No-code builder
Non-technical team members create agents using Dust's visual interface—connect data sources, write instructions in plain English, test, and deploy. Technical users can extend with custom code when needed.
💡 Ready to move beyond individual agents? Try Dust free for 14 days →
Companies deploying Dust agents
Persona: Nearly 300 agents across 11 of 13 departments
Persona, an identity verification company, started with a single engineering agent and scaled to an organization-wide AI transformation:
- PersonaEngineer agent answered technical questions by accessing GitHub repos, the data warehouse, help center, and employee directory—reducing the #ask-engineers Slack channel bottleneck
- RFP response agents pull from previous RFPs and current company documentation to draft responses in hours instead of days
- SQL query agents cut fraud analyst query time from hours to under 30 minutes by generating queries from data warehouse schemas
Results: More than 80% employee adoption across the company. Nearly 300 specialized agents (published and unpublished) now handle workflows from customer support to documentation. Read the full customer story →
Mirakl: 91% utilization with 30% of employees building agents
Mirakl, a marketplace software company serving major retailers, increased AI utilization from 80% to 91% within one month of switching to Dust — without the drop in usage that typically follows a platform migration.
- Sales brief agents pull information from Salesforce and Google Drive automatically to prepare client meetings
- Agents help create multilingual product documentation and microcopy across market
- Customer service response times improved without sacrificing accuracy.
Results: 91% AI utilization within one month of deployment. 30% of employees now actively build agents, with a company goal of reaching 50% builders. Read the full customer story →
💡 These are two examples. See how other teams are using Dust across sales, support, engineering, and more. Read all customer stories →
Frequently asked questions (FAQs)
How do I share a ChatGPT agent with my team?
You can share a Custom GPT via link. When you update your agent's instructions or knowledge files, those changes apply to everyone accessing the shared GPT after you click “Update” to publish a new version. For teams that need synchronized agents with shared updates, enterprise platforms provide centralized deployment and version control.
What's the difference between Custom GPTs and Agent Mode in ChatGPT?
Custom GPTs are personalized chatbots you configure with specific instructions, knowledge files, and capabilities. You create them once and reuse them for consistent tasks. Agent Mode is ChatGPT's autonomous execution capability where you describe a complex task and the AI handles the entire workflow using its own virtual computer to browse websites, create files, and deliver completed work. Custom GPTs help maintain consistency across conversations. Agent Mode handles multi-step tasks with minimal supervision.
How is Dust different from ChatGPT agents?
Dust connects to your company's live data sources like Slack, Notion, and Salesforce, so agents stay current as your information changes. It supports multi-agent workflows where specialized agents collaborate on complex tasks. You can choose from multiple AI models (OpenAI, Anthropic, Google, Mistral) for each agent. Dust provides enterprise security with role-based permissions and audit logs. Teams can build agents collaboratively with centralized updates that propagate across the organization. ChatGPT agents work well for individuals, while Dust is built for teams and enterprises.