From: aidotengineer
Agentic workflows are becoming increasingly important in the development of reliable AI solutions [00:13:36]. These workflows are characterized by how much control, reasoning, and autonomy an AI system possesses [00:14:43]. As models evolve, so too does the complexity and capability of these workflows, moving beyond simple API calls to proactive, decision-making systems [00:15:06]. Success in production isn’t solely about the models; it’s about how the system is built around them [00:14:17].
Levels of Agentic Behavior
A framework has been developed to define different levels of agentic behavior, recognizing that every AI workflow has some level of this behavior [00:14:40]. This framework is not static and can expand as models evolve [00:15:06].
L0: Basic LLM Call (No External Planning)
At Level 0, the workflow involves an LLM call retrieving data, potentially with inline evaluations, to generate a response [00:15:20].
- Characteristics:
- No external reasoning, planning, or decision-making beyond what’s embedded in the prompt and the model’s inherent behavior [00:15:32].
- The model performs all reasoning within the prompt itself [00:15:38].
- No external “agent” organizes decisions or plans actions [00:15:43].
- Some reasoning and agentic behavior still exist at the model level [00:15:49].
L1: Tool Use
Moving to Level 1, workflows gain the ability to use various tools [00:15:56].
- Characteristics:
- The AI system decides when to call APIs and take actions [00:16:02].
- The model decides whether to call a specific tool or a vector database to retrieve more data before generating an output [00:16:13].
- Memory plays a crucial role for multi-threaded conversations, capturing context throughout the workflow [00:16:25].
- Evaluation is necessary at every step to ensure models make correct decisions and use the right tools [00:16:37].
- Workflows can range from simple to complex, with multiple branching paths and numerous tools [00:16:50].
- The focus is on orchestration: how models interact with the system and data, and how to ensure correct context from databases [00:21:01].
L2: Structured Reasoning
At Level 2, workflows demonstrate structured reasoning, moving beyond simple tool use [00:17:15].
- Characteristics:
- Workflows notice triggers, plan actions, and execute tasks in a structured sequence [00:17:28].
- Tasks are broken down into multiple steps, involving information retrieval, calling tools, evaluating usefulness, and refining outputs in a continuous loop [00:17:37].
- Agentic behavior becomes more intentional, as the system actively decides what needs to be done and spends more time “thinking” [00:17:55].
- The process is still finite; the workflow terminates once its planned steps are completed [00:18:16].
- Significant innovation is expected here, with AI agents developed for planning and reasoning using advanced models like 01 or O3 [00:21:41].
L3: Proactive Autonomy
Level 3 introduces greater autonomy and decision-making not explicitly defined by human creators [00:18:33].
- Characteristics:
- Systems proactively take actions without waiting for direct input [00:18:45].
- Instead of terminating after a single request, the system remains active, continuously monitoring its environment and reacting as needed [00:18:52].
- Can interact with external services (e.g., email, Slack, Google Drive) to plan and execute next moves in real-time, or request human input [00:19:02].
- AI workflows become less of a tool and more of an independent system [00:19:22].
L4: Fully Creative/Inventive
Level 4 represents the most advanced stage, where AI moves beyond automation and reasoning to become an inventor [00:19:39].
- Characteristics:
- Creates its own new workflows, utilities, agents, prompts, function calls, and tools [00:19:56].
- Solves problems in novel ways [00:20:06].
- Currently “Out Of Reach” due to constraints with models like overfitting (models favoring training data) and inductive bias (models making assumptions based on training data) [00:20:12].
- The goal is AI that invents, improves, and solves problems in unforeseen ways [00:20:30].
Current State and Future Outlook
Most production-grade AI solutions currently fall within the L1 segment, focusing on robust orchestrations that enable models to interact effectively with systems and data [00:20:41]. Significant innovation is anticipated in L2 workflows this year, with the development of AI agents capable of planning and reasoning for complex tasks [00:21:41]. L3 and L4 are still limited by current models and surrounding logic [00:22:22].
Example: SEO Agent Workflow
An example of an agentic workflow that lies between L1 and L2 is an SEO agent designed to automate the entire SEO process from keyword research to content creation [00:23:04].
- Components:
- SEO Analyst and Researcher: Takes a keyword and analyzes top-performing articles using Google Search [00:23:43]. It identifies good components to amplify and missing segments for improvement [00:23:54]. The researcher conducts further searches based on identified gaps [00:25:38].
- Writer: Uses the research and planning output to create a first draft [00:26:04]. This content is context-rich, drawing from analyzed articles and potentially a RAG database of existing articles and learnings [00:26:21].
- Editor (LLM-based Judge): Evaluates the first draft against predefined rules in its prompt [00:24:21].
- Feedback Loop: The editor’s feedback is passed back to the writer via a memory component (chat history), continuously looping until a specific criterion is met (e.g., the editor deems it an “excellent post” or a set number of iterations are completed) [00:24:31].
- Memory Component: Captures all previous conversations between the writer and the editor [00:24:40].
- Outcome: Produces a useful and impressive first draft that leverages context smartly [00:24:51]. This system saves significant time by automating content foundations [00:27:47].
This example highlights the importance of a test-driven approach for building reliable and continuously improving agentic systems [00:21:25].