Video Summary
Agentic workflows let you describe outcomes in natural language and the system implements the automation.
Drag-and-drop platforms (n8n, Make, Zapier) are becoming less productive compared to agentic tools.
Enterprise adoption of agentic AI is accelerating; the market and built-in agentic features are growing fast.
Demo: build an agent triggered from ClickUp using Claude Code and trigger.dev to research and publish content.
Key risks include context drift, hallucinations, and scope creep — verify generated code and test with real data.
Because agentic workflows that accept natural language are speeding up automation development — the same automations that took days in drag-and-drop tools can be built in minutes, and enterprises are rapidly adopting agentic AI.
A workflow where you describe the desired outcome in plain language and an AI agent interprets the request, connects tools, writes and debugs code, and executes the automation autonomously.
Claude Code is used to create the agent logic and Trigger.dev (with integrations like ClickUp) is shown for execution and observability; other APIs like Nano Banana Pro and key.ai are referenced for image generation.
Context drift (agents forgetting earlier instructions), hallucinations (fabricating functions or data), and scope mismanagement — mitigate by testing with real data and keeping concise project summaries.
Use error notifications, observability/monitoring, and version control; verify and test generated code, and maintain clear scope and project summaries for long-running agents.
"A new phase for AI automations has begun."
The landscape of AI automation is shifting dramatically from traditional drag-and-drop platforms, such as Make.com and Zapier, to more intuitive systems that utilize natural language processing.
These new tools simplify the automation process, allowing users to build complex automations in hours rather than weeks, merely by describing their needs in plain language.
A significant change is predicted, with estimates suggesting that 50% of enterprises will adopt these new systems by 2027.
"Instead of telling the system how to do something step-by-step, you're just telling it what you want."
Previously, creating AI automations involved a detailed, step-by-step configuration, where users manually connected nodes, passed variables, set up API calls, and debugged errors.
While tools like end-to-end platforms democratized automation for non-developers, they still required extensive input and management from the user.
The rise of agentic workflows means users now simply convey their desired outcomes to the system, which interprets the requirements and manages the implementation process autonomously.
"The agentic AI market was valued at just over $5 billion in 2024, and by 2034, it's projected to hit nearly $200 billion."
There's a massive financial shift toward agentic AI, with 96% of enterprises planning to enhance their agentic AI usage within the year.
By 2028, a third of all enterprise software is expected to include built-in agentic AI, indicating strong industry acceptance and integration.
This trend reflects the long-term potential and preference for automation methods that prioritize ease of use and rapid implementation.
"We've seen this pattern play out before."
The evolution of AI automation can be characterized by distinct waves, starting with the emergence of AI chatbots that allowed users to interact conversationally.
The second wave integrated AI with existing automation platforms, enabling users to create workflows that can perform complex tasks automatically.
The current wave emphasizes agentic tools, which prioritize natural language inputs and reduce the responsibility on users for meticulous programming, thus enhancing productivity.
"Agentic tools don't change what you build, but they significantly change how you build it."
Comparing the old method of building automations with newer agentic systems illustrates the complexity of traditional methods, which involved managing numerous detailed configurations manually.
Current agentic tools like Claude Code allow users to describe the automation they want in natural language, which the system then translates into functional workflows without the user having to manage the underlying logic actively.
This represents a shift from a focus on the end result to a focus on simplifying the process of achieving that result.
"You can literally come into Claude Code and just tell it in natural language what you want."
A practical scenario illustrates the efficiency of using agentic workflows to create an automation that checks for updates on a specific YouTube channel every eight hours.
Traditionally, this process would have involved gathering data, structuring logic, and checking a database, tasks that can be streamlined using agentic tools where the user simply states their requirements.
The rapid development of the automation highlights the effectiveness and ease of using natural language to achieve complex tasks quickly and efficiently.
"Every time your YouTube Check TypeScript runs, we get this sort of idempotency key, which is the video ID."
The automation process incorporates a deduplication mechanism that prevents processing the same video multiple times by using a unique video ID as an identifier. This feature simplifies the logic since there's no need to manually manage duplicates within the workflow.
The main advantage of using agentic workflows is that they streamline automation without requiring extensive logic or database setups from the user. Tools like Claude Code and Trigger.dev empower users to build these workflows more easily.
"I built this simply by telling Claude Code, 'I want an AI agent that can see tasks in ClickUp, do research, and then send me a brief.'"
The speaker demonstrates the creation of a competitive research agent triggered by a task in ClickUp, showcasing how the agent autonomously decides on a research strategy. After gathering sufficient information, it generates a structured brief which is then delivered back to the ClickUp platform.
The integration of visual components in Trigger.dev facilitates real-time tracking of automation processes, providing clarity on how tasks are being executed.
"The longer you work within an agent, the more it starts to forget what you told it."
Context drift can occur during prolonged sessions with an AI agent, leading to potential miscommunication or "hallucinations" where the agent produces inaccurate or fabricated results. It is crucial to implement methods to maintain context, such as breaking tasks into shorter sessions or keeping a project summary updated.
Hallucinations, or the generation of non-existent functions, can cause issues if the generated code is applied without testing. Developers should verify the code with actual data to catch subtle errors often overlooked.
Scope management is important; agents can either over-engineer simple requests or under-engineer complex ones. By providing clear and specific instructions, users can help agents stay within the intended boundaries.
"With agentic workflows, you're writing code that needs to be managed somehow."
After building an automation, it’s essential to have proper management practices in place, including error notifications for runtime issues, observability for tracking the performance of automations, and version control to oversee changes and collaborations within projects.
Awareness of these management factors is important, as users may mistakenly view failures as tool flaws when they simply need to learn how to interact with the system correctly.
Prior experiences with end-to-end automation can prove beneficial in transitioning to agentic systems, as foundational knowledge about automation principles, triggers, and error handling remains relevant and assists users in effectively directing AI workflows.
"I want to build another agent triggered by a ClickUp task, similar to our company researcher, but for a different list."
The process begins by determining the desired functionalities before diving into building the agent. The prompt suggests creating an agent that is triggered by tasks from ClickUp, aimed at generating a LinkedIn post along with a relevant infographic.
The agent needs to conduct research on a provided topic and output both the LinkedIn content and the infographic utilizing tools like Nano Banana Pro and key.ai for API access. It’s crucial to specify where the outputs should be directed, which in this case is a specific ClickUp task.
"It will do a research phase, a writing phase, and an infographic phase, and it will poll to see when the image is done."
The newly created agent is designed to follow a structured workflow, which includes phases for research, writing, and infographic creation. Importantly, the agent autonomously manages the polling mechanism to check the status of the tasks it initiates.
The agent asks questions for clarification, fine-tuning its approach to ensure proper configuration, such as the LinkedIn post style and the naming of the ClickUp list. It's notable that it operates without needing extensive manually inputted parameters.
"It ran into two issues, but worked on fixing them automatically."
During the execution of the agent's tasks, some errors surfaced related to ClickUp integration and infographic generation. However, the agent proves its efficacy by automatically diagnosing and rectifying the issues without manual intervention.
The system indicates that the agent is capable of conducting multiple web searches and reading URLs, which allows it to gather context and information necessary for fulfilling its objectives.
"Considering I didn't give any prompts for writing styles, that's a really solid first iteration."
Upon completion of the task, the agent successfully generates a LinkedIn post along with an infographic detailing statistics about AI agents in enterprise automation. The execution time for this entire process was approximately 2 minutes and 23 seconds.
The output includes valuable content that indicates a significant gap in AI agent adoption and implementation, showcasing the effectiveness of the automation process without extensive direction on formatting or required data.
