From Handcrafted Workflows to AI Agents to Agentic Workflows
This evolution — from handcrafted chatbot/RPA flows to AI-driven adaptive workflows — is transforming conversational AI, automation & decision-making.
Graph Approach
For decades we had an approach of handcrafting and mapping out processes from start to finish.
For instance in the case of an enterprise that built a chat bought, a flow had to be designed and developed for each and every possible scenario.
Each of these flows were linked to and intended to address a certain customer intent and once the intent is detected certain a certain predetermined, fixed flow was invoked.
This approach became the backbone of automation, manually designing and encoding decision trees, dependencies and execution paths.
The problem however with this approach was that it was obviously predetermined and not flexible and could not adapt to changes especially related to user behaviour.
There was also an element of maintenance which demanded that these flows were reviewed and updated as vulnerabilities would take it and processes and products changed.
If you like this article & want to show some love ❤️
-Clap 50 times, each one helps more than you think! 👏
- Follow me on Medium and subscribe for free. 🫶
- Let’s connect on LinkedIn, and stay in touch on X!
AI Agents
I have written extensively about AI agents, but suffice to say that AI agents brought about a process where a user query is automatically decomposed into sub-steps and the sub-steps are chained together in the sequence or a flow which is executed by the AI agent to reach a final conclusion.
The agent iterates through a process of observation, thought, action, observation, and this process iterates until the final answer is reached.
The AI agent has a part of its output the chain or sequence of events it created to reach the final conclusion.
Then advent of AI Agents, shifted automation from static rule-following to dynamic decision-making. AI Agents integrated reasoning, tool use, and adaptability, allowing them to execute tasks with greater flexibility.
Using AI Agents To Create Agentic Workflows
Recent research from OpenAI used their reasoning model, taking knowledge articles and converting them into a sequence of events with conditions.
The o1 model, with its advanced reasoning capabilities, is seemingly well suited for creating routines that convert knowledge articles into process flows.
Its ability to handle complex, structured information without extensive prior training allows it to deconstruct intricate knowledge articles — such as those containing multi-step instructions, described decision trees, or diagrams — into actionable routines.
By leveraging its zero-shot capabilities, o1 can efficiently interpret and break down tasks into clear, manageable steps without requiring extensive prompting or fine-tuning.
And this is a new paradigm I really find interesting, where AI Agents are not merely used to reach a final conclusion but are leveraged to create workflows or graphs based on existing data.
This is something that is referred to as the agentic workflows, where the user expresses a desire and a workflow is generated and presented to the user.
The user then performs supervision, approves or changes to workflow and fire it off to be executed.
So there's approach lies somewhere in between handcrafted manual workflows and completely autonomous AI agents where there is a level of human supervision.
Agentic Workflows, where AI is no longer just a worker — it’s becoming the architect of its own workflows.
- AI can synthesise information, break down problems, and create optimal task sequences.
- Workflows can adapt dynamically based on context, environment, and real-time data.
- Human oversight shifts from designing workflows to validating AI-generated processes.
As AI Agents gain more autonomy over workflow creation, the boundary between automation and decision-making is blurring.
Human Supervision / Agentic Workflows
Considering the image below…it is a good example of introducing agency in a measured fashion.
The user asks a question: I need to set an alarm for every weekday morning at 7:30, and then cancel the alarm for Thursday, changing it to 8:00 in the evening.
This is a compound and multi-intent utterance, but see how the Agentic Assistant breaks down the request into a sequence of tasks and sub-tasks. And the user has the option to delete steps, or refine steps by splitting them.
This image shows the plan generated on the left, and the conversation on the right.
And below a last example of a compound user input / request, and the decomposed task with multiple steps (agentic workflow) for human approval.
Chief Evangelist @ Kore.ai | I’m passionate about exploring the intersection of AI and language. From Language Models, AI Agents to Agentic Applications, Development Frameworks & Data-Centric Productivity Tools, I share insights and ideas on how these technologies are shaping the future.
