In the world of AI, many use the term “agents” inaccurately, referring to basic calls to language models that just respond to prompts. True agents should think independently and employ intelligent reasoning, resembling human decision-making. Current systems, often termed workflows, are helpful but predictable and lack true autonomy. Real agents would reason and adapt to their environment rather than follow hard-coded instructions. Examples like Devin and Anthropic’s initiatives show promise but face challenges such as inconsistency and limited effectiveness. As of now, AI models aren’t quite ready to function as true agents, but advancements may be on the horizon. For specific tasks, simpler solutions are still the way to go.
Watch the Video!
In a fascinating discussion about the evolution of AI, a recent article by Anthropic challenges our understanding of what constitutes an “agent.” Most tools marketed as agents today operate merely as API calls to large language models (LLMs), functioning through a few lines of code and a simple prompt. They lack the ability to act independently or make decisions, merely responding to user queries without any real autonomy.
So, what do we need to consider when defining a “real agent”? To break it down, a genuine agent should possess the capability to make independent decisions, much like our reflective System 2 thinking, as opposed to the quick, automatic responses of System 1 thinking. True agents should not only know how to use tools but also understand when and why to deploy them based on thoughtful reasoning.
While many tools available today can automate processes by coding specific workflows, they still fall short of demonstrating genuine agency. Workflows are critical in creating applications but are predictable and pre-defined, whereas a real agent would operate autonomously. Currently, the development of such intelligent agents remains a work in progress.
Examples like Devin and Anthropic’s computer use attempt to showcase the promise of agent-like systems. Devin acts as a fully autonomous software engineer, managing tasks with some degree of independence. However, these systems often struggle with complex tasks due to their limitations.
In essence, while workflows are valuable for specific tasks and predictable outcomes, they are not true agents. An actual agent would independently analyze needs, iterate solutions, and adapt to various situations—all of which current technologies have yet to perfect.
For now, focusing on simpler solutions using LLM calls is often the most practical approach. As we look toward the future, the dream of creating real agents continues to inspire developers and AI enthusiasts alike.
Links:
Anthropic’s blog on agents: https://www.anthropic.com/research/building-effective-agents
Anthropic’s computer use: https://www.anthropic.com/news/3-5-models-and-computer-use
Hamul Husain’s blog on Devin: https://www.answer.ai/posts/2025-01-08-devin.html
Tags: Artificial Intelligence, Agents, Workflows, LLMs, Anthropic, Devin
What are agents in workflows?
Agents in workflows are like helpers that carry out tasks automatically. They follow specific rules to help manage processes more smoothly, saving time and effort.
How do workflows benefit a team?
Workflows help a team by organizing tasks clearly. They make sure everyone knows their responsibilities, which leads to better teamwork and fewer mistakes.
Can I customize my workflows?
Yes, you can customize your workflows to fit your needs. You can change the steps, add new tasks, and set rules based on what works best for your team.
What tools can I use to create workflows?
There are many tools available for creating workflows, like Trello, Asana, or Monday.com. These tools make it easy to design and manage workflows online.
How do agents improve workflow efficiency?
Agents improve workflow efficiency by automating repetitive tasks. This means that agents can handle routine jobs, allowing team members to focus on more important work.
