How to design an Agentic workflow?

Over the past five years, the tech world has been abuzz with the rise of no-code platforms. Tools that empower users to build custom software without writing a single line of code. Companies like Webflow, Retool, Zapier, AppSmith, and Bubble have ridden this wave to unicorn status, each offering a unique spin on a simple yet transformative idea: designing software around workflows rather than code.

For those who crave a bit more control, platforms like Pipedream even blend workflows with the flexibility of custom coding, supported by a robust library of integrations.

This shift has democratized software creation, allowing non-developers to craft solutions tailored to their needs. But as workflows become the backbone of modern software design, a new question emerges: what happens when we hand these workflows over to artificial intelligence?

This is the motivation behind the concept of an Agentic flow — a term used to describe a workflow where key components are autonomously managed by AI. To unpack this idea, let’s first define workflows using a framework I call T.T.C.A.D and then explore how AI transforms them into something more dynamic.

Understanding Workflows with T.T.C.A.D

Most workflows, whether in no-code platforms or traditional systems, can be broken down into five essential elements, which I’ve anonymized as T.T.C.A.D:

• Trigger Event: The initiating action that starts the workflow (e.g., a customer makes a bank deposit).
• Task: The specific job to be performed (e.g., checking the account’s active status or recording the transaction).
• Condition: The rule or precedent that must be evaluated (e.g., is the deposit over $10,000, or is the account blacklisted?).
• Action: The outcome or process executed based on the condition (e.g., accept the deposit, flag it, or report it as fraudulent).
• Data: The information fueling the workflow (e.g., depositor’s name, amount, recipient account number).

Consider a real-world example: a bank processing a deposit. When a client deposits money (Trigger Event), the system checks the amount and account status against predefined rules (Task and Condition). Depending on the results, it either accepts the deposit or flags it for review (Action), all while handling critical details like the depositor’s identity and account number (Data). This TTCAD framework is reflected in operational workflows across industries, from banking to e-commerce, and it’s the foundation of tools like Zapier and Bubble.

What Makes a flow “Agentic”?

A flow becomes Agentic when an AI system takes full control of three of these components: Tasks, Conditions, and Actions (TCA). In an Agentic flow, the AI doesn’t just assist—it autonomously manages the workflow’s core logic, leaving humans to define the Trigger Event and provide the Data. Here’s how this plays out:

• Tasks: The AI executes the necessary steps, such as verifying account status or analyzing transaction patterns, without human intervention.
• Conditions: The AI evaluates rules or precedents dynamically, potentially adapting them based on real-time learning (e.g., adjusting fraud detection thresholds as new patterns emerge).
• Actions: The AI determines and executes the outcome—whether that’s approving a deposit, flagging it, or escalating it to a human overseer—all based on its analysis.

Imagine our banking example with an Agentic twist: an AI system monitors deposits (Trigger Event) using customer and transaction details (Data). It autonomously checks for fraud indicators (Task), assesses whether the deposit meets suspicious criteria it has learned over time (Condition), and decides to approve or flag the transaction (Action), This evolves automation into decision making intelligence, with a smart system adapting and refining the process without constant human oversight.

Why Agentic flows Matter

The shift to Agentic flows builds on the no-code revolution but takes it a step further. No-code platforms empower users to design workflows, but Agentic flows enable AI to run them with a level of autonomy and adaptability that static rules can’t match. Research from Gartner supports this evolution, predicting that by 2025, over 70% of enterprises will use AI-driven automation to enhance operational efficiency (Gartner, 2023). Similarly, a McKinsey report highlights how AI can reduce human error and optimize decision-making in workflow-driven systems (McKinsey, 2022).

Platforms like Zapier already hint at this future by integrating AI features—think smart suggestions for workflow steps—but Agentic flows go beyond suggestions to full execution.

My Take — and Yours?

I believe Agentic flows represent the next frontier in software design: a marriage of no-code’s accessibility with AI’s intelligence. By handing Tasks, Conditions, and Actions to AI, we create systems that don’t just follow instructions but evolve with the problems they solve. That said, I’d love to hear your thoughts! Do you agree that AI-managed TCA is what defines an Agentic flow? Or do you see other factors like human oversight or ethical constraints playing a bigger role?