What if AI could recognize when it’s time for the next step?
AI can already perform individual steps in a workflow.
But in many workflows, the user still has to manually decide when one step is actually finished and tell the AI to move on.
I think AI could help with that transition too.
Instead of only:
Time → Action
or:
Event → Rule → Action
we could also have:
Workflow progresses → AI recognizes that a meaningful condition has been reached → Next step
The important part is that the condition itself may require understanding the context.
A simple example: learning with ChatGPT
Suppose I’m using ChatGPT to learn something one section at a time.
At the beginning, I might define a workflow like:
Whenever you determine that we’ve finished a section, summarize what I learned before moving to the next section.
Right now, I often still have to act as the workflow controller myself:
ChatGPT teaches me
→ we discuss the topic
→ I decide that the section is probably finished
→ I tell ChatGPT to summarize it
→ ChatGPT summarizes it
→ I tell it to continue
But ChatGPT is already participating in the entire learning process.
It can see what has been covered, the questions I’ve asked, my answers, what I still don’t understand, and how the discussion has progressed.
So ideally, the workflow could become:
ChatGPT teaches me
→ the learning conversation progresses
→ AI determines that the current section has actually been completed
→ automatically trigger the section-summary step
→ continue to the next section
I define that workflow once.
I shouldn’t necessarily have to manually tell the AI when every transition should happen.
This isn’t just scheduled checking
I’m not talking about:
Every 10 minutes → check whether the section is finished.
There doesn’t need to be a timer.
There may not even be one specific machine event that means:
“The section is now complete.”
Instead, that state may emerge gradually from the interaction.
The AI might recognize it from the conversation as a whole.
That’s the capability I’m interested in:
AI recognizing that a workflow has reached a meaningful state, and using that recognition to trigger the next predefined step.
This could apply beyond learning
For example:
A project develops over time
→ AI determines that it has reached the predefined criteria for testing
→ trigger the testing step
Or:
Research accumulates new information
→ AI determines that there is now enough evidence to make a decision
→ move to the decision step
Or:
An agent produces a result
→ AI evaluates the result
→ AI determines whether the goal was actually achieved
→ only then continue the workflow
Sometimes there may be a clear event that causes the state to change.
Sometimes there may not be.
The important part is that AI evaluates the meaning of the current state instead of requiring the user to manually manage every transition.
AI as a workflow state evaluator
Traditional automation is very good when the transition can be expressed deterministically:
9:00 AM → Do X
New email → Do Y
Value > threshold → Do Z
But some transitions are semantic:
“We’ve finished this section.”
“This prototype is ready for testing.”
“We now have enough information to make a decision.”
“This result actually satisfies the goal.”
These aren’t always easy to represent as a timer, webhook, boolean, or predefined machine event.
But evaluating this kind of context is exactly where AI could be useful.
So rather than AI only performing steps inside a workflow:
AI could also help manage the transitions between those steps.
User control would still matter
AI judgment isn’t deterministic, so this shouldn’t mean unrestricted autonomous execution.
The user should be able to define:
- what condition the AI is evaluating;
- what counts as completion;
- what information should be considered;
- what should happen when the condition is met;
- whether confirmation is required before continuing;
- and ideally, why the AI believes the workflow is ready to move forward.
For higher-impact actions, user confirmation could still be required.
And if the AI isn’t confident that the condition has been met, it could simply stay in the current step or ask the user.
The idea isn’t to remove user control.
It’s to remove the need for the user to constantly act as the workflow controller.
I don’t just want AI to perform the steps. I want AI to understand when the workflow is ready to move from one step to the next.
I think this could make AI workflows feel much more natural — especially for learning, research, projects, and multi-agent workflows.
Disclosure: Written with the help of ChatGPT, based on my own idea and experience using the product.