For years, the biggest question in artificial intelligence was simple:
Which AI chatbot is better?
ChatGPT, Claude, Gemini and other AI assistants have competed on writing, reasoning, coding, research and answering questions.
But that competition is changing.
The next battle may not be about which AI can give you the best answer.
It may be about which AI can actually get the work done for you.
OpenAI has now entered this race in a much bigger way with Dots, its new always-on AI agents announced at DevDay 2026. OpenAI describes Dots as agents designed to work continuously toward user goals rather than simply waiting for another prompt. Each Dot gets its own cloud computer and browser and can work with connected applications.
Anthropic's Claude is already moving in a similar direction.
Through Claude Cowork, plugins, skills, connectors and Claude Code, Anthropic is turning Claude from a traditional chatbot into a system that can work across files, applications, coding environments and longer-running tasks.
That makes the latest OpenAI move more interesting.
This is no longer simply ChatGPT vs Claude.
It is increasingly becoming a race to build the AI that can act as a digital coworker.
What Exactly Are OpenAI Dots?
The easiest way to understand Dots is to think about the difference between an assistant that waits for instructions and one that can keep working after you give it a goal.
A normal chatbot interaction usually looks like this:
You ask a question.
The AI responds.
You give another instruction.
It responds again.
Dots are designed around a different model.
OpenAI describes them as always-on agents that can keep working on behalf of users. Instead of repeatedly giving instructions, users can give an agent a broader objective and allow it to work toward that goal.
Each Dot has its own cloud computer and browser. OpenAI says Dots can also connect with thousands of applications through plugins. They can be accessed through ChatGPT and communication tools such as Slack and Teams.
That changes the idea of what an AI assistant can be.
Instead of asking:
"Can you write this report?"
the future could look more like:
"Keep this project updated and let me know when something needs my attention."
The difference is small in the wording but huge in how the AI has to operate.
It needs to understand the goal, use tools, access information, perform multiple steps and potentially continue working without constant supervision.
That is the core idea behind the agent race.
Claude Is Already Building an Agent Ecosystem
OpenAI is not entering an empty market.
Anthropic has been steadily building Claude around agent-style workflows.
Claude Cowork extends Claude's capabilities beyond traditional coding and chat into knowledge work. Anthropic describes Cowork as bringing Claude Code's agentic capabilities to tasks beyond programming.
Claude can also use plugins, skills and connectors.
Plugins can bundle skills, connectors and sub-agents into a single package, giving users a ready-made workflow rather than forcing them to configure every capability separately.
Connectors are particularly important.
They allow Claude to access connected services and, where permitted, take actions within them. Anthropic gives examples such as using Linear to create issues, Slack to send messages and Google Drive to search files.
Skills add another layer.
Anthropic describes skills as a way to give Claude specialized knowledge and workflows so it can perform certain types of work more effectively.
Put all of this together and Claude is moving away from being just a conversational AI.
It is becoming a platform where AI can use tools, follow workflows and complete larger tasks.
Dots vs Claude: Where the Competition Gets Interesting
The interesting part is that OpenAI and Anthropic are approaching the same larger problem from somewhat different directions.
OpenAI is putting a lot of emphasis on persistent, always-on agents.
Claude has been building a broader ecosystem of skills, plugins, connectors, Cowork and coding tools.
Both approaches point toward a similar future.
1. From Prompts to Goals
Traditional AI works primarily through prompts.
You ask.
It answers.
Agentic AI is more goal-oriented.
You give it an objective, and the system determines some of the steps needed to reach that objective.
That means the quality of an agent will depend on more than its language model.
It will also depend on planning, tool use, memory, permissions and its ability to recover when something goes wrong.
2. From One Conversation to Multiple Actions
Imagine you are preparing a business report.
A chatbot might summarize research for you.
An agent could potentially:
- Search for information
- Organize documents
- Analyze data
- Update a spreadsheet
- Draft a report
- Check for missing information
- Continue working on the project
- Send you an update when something requires your attention
That is a fundamentally different experience.
The AI becomes less like a search box and more like a digital worker.
3. Connected Apps Become More Important
This may become one of the biggest battlegrounds.
An AI model can be extremely capable, but its usefulness is limited if it cannot access the information and tools you actually use.
Claude already supports connectors for services such as Google Drive and Slack, among others.
OpenAI is also positioning Dots around connected applications, with OpenAI saying Dots can connect to more than 4,000 apps through plugins.
That means the competition is no longer only about model intelligence.
It is also about ecosystems.
Who has the better integrations?
Who can connect to more useful services?
Who can perform more actions safely?
Who can make those connections easier for ordinary users?
These questions could become just as important as benchmark scores.
Why "Always-On" Matters
The phrase always-on is one of the most important parts of the Dots announcement.
Most AI tools today are reactive.
You open them when you need them.
You provide a prompt.
You wait for an answer.
Then the interaction ends.
An always-on agent changes that relationship.
If an agent is working continuously toward a goal, it could potentially monitor information, make progress on a task and return to the user when human input is needed.
That could be useful for things like research, project management, business operations and repetitive digital work.
But it also creates a much bigger challenge.
How much freedom should an AI have?
The Bigger Problem: Trust
Giving an AI access to more tools also gives it more opportunities to make mistakes.
If an AI only writes a paragraph and gets something wrong, the damage may be limited.
But what happens when an AI can send an email, modify a document, create a task, access company data or interact with another service?
The consequences can be much bigger.
This is why permissions and safeguards are becoming central to the agent race.
OpenAI says Dots operate with dedicated cloud infrastructure and customizable permissions and safeguards for sensitive tasks.
Anthropic also warns users to install plugins only from trusted sources because some plugins can contain local components that run with computer permissions.
In other words, making AI more capable is only half of the challenge.
Companies also need to make sure the AI knows what it is allowed to do.
The Human Still Matters
The rise of AI agents does not necessarily mean humans disappear from the workflow.
In many cases, the more realistic future is a combination of human decisions and AI execution.
A person might define the goal.
The AI handles repetitive work.
The person reviews important decisions.
The AI continues with approved actions.
This could make AI agents more useful without giving them unlimited control.
For businesses, that distinction could become especially important.
Companies may want AI to handle repetitive work while keeping financial decisions, sensitive communications and other high-impact actions under human approval.
Why This Could Be Bigger Than Another Chatbot Launch
There have already been countless AI model launches.
A new model becomes faster.
Another gets better at coding.
Another improves reasoning.
Those improvements matter, but users often experience them through the same basic interface: a chat window.
AI agents could change that.
The interface itself becomes less important.
Instead of opening an AI and asking it what to do, you could have an agent working in the background and only interacting with you when necessary.
That is a much bigger shift in how people may use AI.
OpenAI's Dots announcement even points toward the possibility of multiple Dots working together in the future.
That raises another interesting possibility:
What happens when AI agents start working with other AI agents?
One agent could handle research.
Another could work on coding.
Another could manage documents.
Another could monitor a project.
The user could potentially become the person directing the overall workflow rather than performing every individual step.
Claude vs Dots: It's Not Just About Who Is Smarter
It would be easy to turn this into another debate about which model is smarter.
But that misses the bigger story.
An AI agent's usefulness depends on several things:
Model intelligence + tools + integrations + memory + autonomy + permissions + reliability.
A model can be extremely intelligent and still be frustrating if it cannot access the information you need.
Likewise, an agent with access to hundreds of tools is not useful if it makes unreliable decisions.
The winner of this broader technology race, therefore, will not necessarily be determined by a single benchmark.
It will depend on how well the entire system works together.
What This Means for Users
For everyday users, this could eventually make AI much more practical.
Instead of using AI only when you have a question, you could use it as a background assistant.
For example:
Research:
An agent could monitor a topic and organize new information.
Work:
It could help maintain documents, reports and project information.
Coding:
It could work through development tasks and interact with coding tools.
Business:
It could connect information from different applications and help automate repetitive workflows.
Personal productivity:
It could handle routine digital tasks and remind you when human input is needed.
The exact capabilities will continue to change as these systems develop.
But the direction is becoming clear.
The AI Agent Race Is Getting Serious
OpenAI Dots marks an important step in OpenAI's push toward always-on AI agents.
Claude, meanwhile, is already building a large ecosystem around Cowork, plugins, skills, connectors and Claude Code.
Neither approach is simply about building another chatbot.
The larger goal is to make AI capable of taking action.
That changes the question at the center of the AI industry.
A few years ago, people asked:
"Which AI gives the best answer?"
Now the question is becoming:
"Which AI can actually get the job done?"
That may be the real AI race to watch.
OpenAI Dots and Claude are two important pieces of that shift, but they are unlikely to be the last.
As AI agents become more capable, the competition will increasingly move beyond models and chat interfaces toward tools, ecosystems, autonomy, reliability and trust.
The chatbot era is not necessarily ending.
But something much bigger may be growing alongside it:
the era of AI agents.