OpenAI's new Dots are always-on AI agents designed to keep working toward your goals across apps, tools, and cloud computers. This could change AI from something you ask into something that works beside you.
What if you gave your AI a task at 9 AM...
...closed your laptop...
...went to sleep...
…and the AI was still working on it at 3 AM?
Not waiting.
Not asking you what to do next.
Not sitting behind a chat box.
Actually working.
That is the idea behind Dots, OpenAI's new always-on AI agents announced at DevDay 2026.
OpenAI describes Dots as agents that can work continuously toward your goals, use their own cloud computer, learn from feedback, and connect to more than 4,000 apps through its plugin ecosystem.
And that changes something fundamental.
AI is moving from something you talk to...
to something that works for you.
For years, the interaction model looked like this:
┌─────────────────────────┐
│ CHATBOT │
│ │
│ ASK → ANSWER │
│ │
│ Waits for your prompt │
└─────────────────────────┘Then came AI assistants.
┌─────────────────────────┐
│ AI ASSISTANT │
│ │
│ ASK → THINK → USE TOOL │
│ │
│ Helps complete a task │
└─────────────────────────┘Now OpenAI is pushing toward something different:
┌─────────────────────────┐
│ ALWAYS-ON AGENT │
│ │
│ GOAL │
│ ↓ │
│ PLAN │
│ ↓ │
│ ACT │
│ ↓ │
│ OBSERVE │
│ ↓ │
│ ADAPT │
│ ↓ │
│ CONTINUE │
└─────────────────────────┘The AI doesn't necessarily stop when the conversation ends.
The work continues.
OpenAI describes Dots as highly capable, always-on agents powered by GPT-6 Astra.
Each Dot has its own cloud computer and can work toward user goals around the clock.
It can also connect to external applications and communicate through surfaces such as ChatGPT, Slack and Teams.
Think of it less like a chatbot.
And more like a digital employee with a computer.
YOUR GOAL
│
▼
┌───────────┐
│ DOT │
└─────┬─────┘
│
┌───────────┼───────────┐
▼ ▼ ▼
CLOUD PC APPS TOOLS
│ │ │
└───────────┼───────────┘
▼
EXECUTE
│
▼
RESULTAnd unlike traditional automation, the agent can reason about the task while it works.
This may sound like a small product feature.
It isn't.
Traditional ChatGPT:
You ask → AI responds → conversation pauses.
Dots:
You define a goal → AI continues working.
OpenAI's own examples describe Dots investigating bugs in Slack, turning designs into working applications, keeping planning cycles moving, and preparing an invoice after noticing that one had been forgotten.
That's a completely different relationship with software.
OLD WORLD
USER
↓
PROMPT
↓
AI
↓
ANSWER
↓
DONE
DOT WORLD
USER
↓
GOAL
↓
AI
↓
WORK
↓
CHECK
↓
WORK
↓
CHECK
↓
WORK
↓
RESULTThe word that matters here is:
Instead of telling an AI:
"Write me a market report."
You could potentially give an agent a larger objective:
"Track this market, research new developments, update the report and tell me when something important changes."
That's not a prompt anymore.
That's a responsibility.
PROMPT
"Write this report."
↓
ONE-TIME TASK
GOAL
"Keep this report updated."
↓
ONGOING RESPONSIBILITYAnd that distinction could become one of the biggest changes in software.
Here's where Dots become particularly interesting.
OpenAI says each Dot has its own cloud computer.
That means the agent isn't simply generating text.
It can operate within a computing environment.
DOT
│
▼
┌──────────────┐
│ CLOUD │
│ COMPUTER │
├──────────────┤
│ Browser │
│ Files │
│ Tools │
│ Apps │
│ Code │
└──────┬───────┘
│
▼
REAL WORKThat is a major architectural shift.
The AI isn't merely describing how to do something.
It can potentially perform the work itself.
OpenAI says Dots can connect to more than 4,000 apps through its plugin ecosystem.
Think about what that means.
DOT
│
┌─────────────┼─────────────┐
▼ ▼ ▼
SLACK EMAIL CRM
│ │ │
▼ ▼ ▼
DATABASE CALENDAR DOCS
│ │ │
└─────────────┼─────────────┘
▼
YOUR WORKFLOWThe power isn't just inside the model.
It's in the connections.
An intelligent model with no tools can mostly talk.
An intelligent model with tools can act.
An intelligent model with thousands of tools can potentially become an operational layer across an organization.
Imagine a company with:
10 HUMAN EMPLOYEES
+
50 AI AGENTSThe humans define strategy.
The agents handle repetitive operational work.
One agent monitors support tickets.
Another prepares reports.
Another investigates production issues.
Another tracks competitors.
Another maintains documentation.
Another follows up on invoices.
Another writes code.
Another tests the code.
Suddenly, the organization isn't simply using AI tools.
It has an AI workforce.
OpenAI says it envisions teams of Dots working together on a user's behalf.
That creates another interesting architecture:
HUMAN
│
▼
┌───────────┐
│ MANAGER │
│ DOT │
└─────┬─────┘
│
┌───────────┼───────────┐
▼ ▼ ▼
RESEARCH CODING OPERATIONS
DOT DOT DOT
│ │ │
└───────────┼─────────────┘
▼
RESULTSOne agent coordinates.
Other agents specialize.
This is where multi-agent systems become much more powerful.
Now imagine the same architecture with a compromised agent.
USER
↓
AI AGENT
↓
4,000+ APPS
↓
EMAIL
FILES
DATABASES
PAYMENTS
CODE
INTERNETSuddenly, an AI mistake isn't just a bad answer.
It could become:
A bad action.
And if the agent is always running?
The mistake could continue after you have stopped watching.
That's why OpenAI emphasizes controls around Dots, including boundaries, progress monitoring and human participation in decisions that require approval.
Traditional AI safety often focused on:
"What is the model going to say?"
Agent safety asks:
"What is the model allowed to do?"
That's a much bigger question.
OLD AI SECURITY
MODEL
↓
OUTPUT FILTER
↓
USER
AGENT SECURITY
MODEL
↓
AGENT
↓
PERMISSIONS
↓
TOOLS
↓
SANDBOX
↓
MONITORING
↓
APPROVAL
↓
REAL WORLDThe security boundary moves outside the model.
And that's exactly where the AI industry is heading.
The old software model was:
Open app → do task → close app.
The agent model is different:
Set goal → agent keeps working → agent observes changes → agent adapts → agent reports.
That means AI starts looking less like software you launch...
and more like a service that is continuously running.
HUMAN
│
▼
GOAL
│
▼
AI AGENT
│
┌────┴────┐
▼ ▼
ACT OBSERVE
│ │
└────┬────┘
▼
ADAPT
│
└─────────┐
▼
REPEATThat's a fundamentally different computing paradigm.
Today companies buy:
AI subscriptions.
Tomorrow they may buy:
AI workers.
Imagine pricing based on:
Instead of:
"We have 500 employees using an AI assistant."
Companies may eventually say:
"We have 80 autonomous agents handling operations."
That changes the economics of software.
They obviously are.
The bigger question is:
An agent that can:
read → useful
write → more useful
execute → powerful
make decisions → extremely powerful
continue without supervision → fundamentally different
READ
↓
WRITE
↓
EXECUTE
↓
DECIDE
↓
ACT AUTONOMOUSLYEvery step increases productivity.
Every step also increases risk.
For decades, software required us to interact with interfaces.
We opened:
Calendar
CRM
Spreadsheet
Browser
IDE
Database
Then we performed the work.
Agents potentially reverse that relationship.
Instead of:
Human → 10 applications
we move toward:
Human → AI agent → 10 applications
TODAY
YOU
↓
APP
↓
CLICK
↓
TYPE
↓
RESULT
AGENT ERA
YOU
↓
GOAL
↓
AGENT
↓
APPS
↓
ACTIONS
↓
RESULTThe interface becomes the agent.
OpenAI isn't alone in this direction.
The entire industry is moving toward increasingly autonomous systems.
The progression looks something like:
2022
AI GENERATES TEXT
↓
2024
AI USES TOOLS
↓
2025
AI WRITES & RUNS CODE
↓
2026
AI OPERATES COMPUTER
↓
NEXT
AI MANAGES LONG-RUNNING GOALS
↓
FUTURE
AI TEAMS WORK TOGETHERThe interesting part isn't that AI is becoming smarter.
It's that AI is becoming persistent.
A chatbot can make a mistake.
You correct it.
Conversation ends.
An autonomous agent can make a mistake.
Then:
retry
↓
adapt
↓
try another method
↓
continue
↓
execute
That's why autonomous systems require fundamentally different safety controls.
The problem isn't simply intelligence.
It's intelligence + persistence + access.
INTELLIGENCE
+
PERSISTENCE
+
TOOLS
+
ACCESS
=
AUTONOMOUS POWERDots are important.
But the bigger story is the direction they represent.
For years, we asked:
"How can AI answer my question better?"
Now we're beginning to ask:
"How can AI take responsibility for the task?"
That is a much bigger question.
Because once AI can take responsibility for a task...
we no longer need to think of it purely as a chatbot.
We can think of it as a digital worker.
The first AI revolution was:
AI that could talk.
The second:
AI that could reason.
The third:
AI that could use tools.
The next:
AI that keeps working.
CHATBOT
↓
ASSISTANT
↓
AGENT
↓
ALWAYS-ON AGENT
↓
AI WORKFORCEAnd that's the real significance of OpenAI's Dots.
Not another chatbot.
Not another model release.
But a new idea about what AI should actually be.
Something that doesn't wait for you to ask.
Something that keeps working toward the goal.
Something that can operate across the software you already use.
And eventually...
something that may become a persistent digital coworker.
The biggest question of the agent era may therefore not be:
"How smart is your AI?"
It may be:
Next in sequence: #3 — “AI Agents Are Becoming Hackers — The Internet Wasn't Built for This”.
Nextgen Blog — YouTube Channel Owner
IIT Mandi CSE graduate and Full Stack Engineer with 6+ years of experience. Creator of the Nextgen Blog YouTube channel, covering AI, web development, and modern software engineering.
Discussion
Comments powered by Giscus
To enable comments, fill in GISCUS_REPO, GISCUS_REPO_ID, and GISCUS_CATEGORY_ID in components/blog/GiscusComments.tsx
