Why Agents and when to use them
“The best way to predict the future is to invent it.” — Alan Kay
Large language models (LLMs) are ushering in a new era of autonomous workflows. With the ability to reason, use tools, and perform multi-step tasks, we now have agents — systems that act independently to accomplish user goals.
In this first article of my Agentic Experiments Series, we’ll explore:
- What an agent is
- When to use one — and when to avoid them
- Helpful resources to start experimenting
- A sneak peek into the upcoming series of agentic experiments
🛠 What is an Agent?
Unlike conventional software that simply automates workflows, agents take action on your behalf.
A workflow is any sequence of steps required to achieve a goal — from resolving support tickets to generating reports or managing bookings.
Core traits of agents:
- Decision-making: They use LLMs to plan, execute, and correct tasks autonomously.
- Tool usage: They interact with external systems dynamically, following clear guardrails.
Note: Simple chatbots, sentiment classifiers, or single-turn LLMs are not agents, because they do not execute workflows end-to-end.
⚡ When to Use Agents
Agents shine in workflows that are complex, ambiguous, or data-heavy.
Examples of agent-friendly scenarios:
- Complex decisions: Refund approvals with exceptions or nuanced customer requests
- Difficult-to-maintain rules: Legacy systems with sprawling conditional logic
- Unstructured data: Processing claims, analyzing text, or interacting conversationally
When agents might not be necessary:
- Simple, repeatable tasks with deterministic rules
- Scenarios without ambiguity or multi-step logic
🧩 Designing Agents
A functional agent usually combines three components:
- Model: The LLM powering reasoning
- Tools: APIs or functions the agent can call
- Instructions: Clear guidelines and guardrails
Tips for effective agent design:
- Start with a single agent before scaling to multiple
- Equip it with relevant tools for data retrieval, actions, or orchestration
- Write precise instructions and anticipate edge cases
- Implement guardrails to ensure safety, reliability, and compliance
🚀 Agentic Experiments Series
This article is just the start. Over the coming posts, we’ll:
- Build autonomous agents step by step
- Monitor and evaluate their performance
- Share practical insights for real-world workflows
Expect hands-on experiments in SaaS automation, productivity tools, and agentic workflows — all with actionable takeaways.
Stay tuned for the first experiment: creating an agent to automate a simple SaaS workflow.
📸 Visual Inspiration
“Automation applied to an efficient operation will magnify the efficiency. Automation applied to an inefficient operation will magnify the inefficiency.” — Bill Gates
🔗 Resources to Explore
- Project vent agentic experiment
- Building agents with Claude Code sdk)
- Autonomous myth
- Model Context Protocol
Next article: Building Your First Autonomous Agent — Step by Step
We’ll dive into a practical agent example, including code, tools, and orchestration patterns you can replicate.