AI Agents in Operations: Practical Automation

4 MINS

# AI Agents in Operations: Practical Automation

I've been building AI agents as personal projects — using crewAI for Python-based custom agents and Lindy for agent orchestration. The hype around AI agents is enormous. The practical applications are more limited, but genuinely valuable when done right.

Starting with Real Problems

The temptation with AI agents is to build impressive demos. Agents that can browse the web, write code, and compose emails — all in one flow. It looks magical in a video.

In practice, I've found the most useful agents are boringly specific:

An agent that summarizes marketplace listings into standardized formats

An agent that triages support tickets by urgency

An agent that extracts structured data from unstructured documents These aren't exciting. They're useful. That's the difference between demo and deployment.

CrewAI for Customization

When I need precise control over agent behavior, crewAI is my go-to. Python-based means I can integrate with existing systems, add custom logic, and debug when things go wrong.

The key insight from building with crewAI: agents are only as good as their tools. Giving an agent access to the right APIs, the right data sources, the right output channels — that's where the real work is.

The agent itself is often the easy part.

Lindy for Orchestration

Lindy approaches the problem differently — it's about orchestrating agents without deep technical setup. For operations workflows where speed matters more than customization, it's powerful.

I've used Lindy for:

Multi-step approval workflows

Data gathering across multiple sources

Automated reporting pipelines The lesson: different tools for different problems. Don't use a custom Python agent when a no-code workflow will do.

Operations as the Ideal Use Case

Why operations? Because operations work is:

Repetitive — The same patterns recur

Structured — Rules can be defined

Tolerant of errors — Mistakes can be caught and corrected This combination makes operations ideal for AI augmentation. You're not replacing human judgment on critical decisions. You're removing tedious work so humans can focus on exceptions. At Cardino, improving internal efficiency through product meant constantly asking: what's the next thing we can automate? AI agents are another tool in that toolkit.

The Human Loop

Every agent workflow I build has human checkpoints. Not because the AI can't be trusted, but because operations reality is messier than any model can capture.

The goal isn't full automation. It's intelligent assistance. Agents handle the 80%, humans handle the 20% that requires judgment. That's where the leverage is.

From Founder to PM: The Mindset Shift

Building Products Across Cultures: A Multilingual Perspective

Amon skipped presentations and built real AI products.

Amon Fayz was part of the September 2025 cohort at Curious PM, alongside 13 other talented participants.