A reflection after a recent conversation with builders in this space.
Let me begin the way I always do-by going back to the basics.
There are Mainly Two Kinds of Agents
There are mainly two kinds of agents people are building today: horizontal and vertical.
Horizontal Agents
Horizontal agents are like generalists. For example, they can schedule meetings, summarize emails, draft blogs, answer research queries, and much more - all without being tied to a specific domain. These agents follow broad instructions with ease, making them powerful generalists.
Vertical Agents
Vertical agents? They’re the specialists. Laser-focused. You give them one task-like helping someone manage returns in an e-commerce store or onboarding vendors-and they get really good at that one thing. These agents aren't built for everything, just the one thing they’re supposed to do, and they do it well.
It’s the same lens we’ve used for SaaS over the years. General tools vs. niche tools. And now, it’s playing out again-this time with agents.
So... Why Now?
This reminded me of something.
Back in 2018, I was knee-deep in the blockchain world-reading, exploring, and obsessing over smart contracts and token economies. I remember telling a junior of mine, “This is a wave worth riding.” He did. Joined one of the earliest smart contract platforms, and today, he’s one of the leading engineers in that space.
The takeaway? Being early matters. Not just in tech trends-but in how you bet on your own time and energy.
I feel the same now, in 2025, about AI agents.
We’re still living in the age of ANI-Artificial Narrow Intelligence. But AGI (Artificial General Intelligence) isn’t far off. And ASI? Let’s just say the whispers are turning into serious conversations. You don’t need to bet the house. But you can bet your next 10 years here. You won’t regret it.
But Not Every Problem Needs an Agent
Before you start throwing agents at every use case, take a breath.
Let’s ask the obvious: What is an agent, really?
Here’s the textbook version:
“Agents are systems that independently accomplish tasks on your behalf.”
But here’s the street-smart version:
Agents are what you build when rules and checklists just don’t cut it anymore.
Think about it. Some problems are best solved with a simple workflow. Zapier. Notion. A few conditionals and you’re done.
But then there are those messy problems. Ones that…
- Involve a bunch of fuzzy decisions
- Depend on understanding natural language
- Or keep breaking down every time the rules change
That’s when an agent makes sense.
Filters to Decide If You Need an Agent
Use these as your north star:
- Is there complex decision-making involved?
- Are your current rules becoming too hard to maintain?
- Are you relying heavily on unstructured data-documents, text, conversations?
If the answer is yes to one or more, you’re in agent territory.
Building Blocks of an Agent
At its core, every agent is made up of:
- Model – The brain (usually an LLM like GPT-4o)
- Tools – The arms and legs (APIs, functions it can call)
- Instructions – The soul (what you tell it to do, how to behave)
Let’s Make it Real
Example: Cheap Flight Booking Agent
- Model: GPT-4o
- Tool: flightFinder – a tool to search flights
- Instructions: “You’re a helpful travel assistant. Ask for the user’s travel dates and destination. Use flightFinder to get the cheapest options. If nothing works, suggest nearby airports or alternate dates.”
Simple? Yes. But also powerful.
Instead of spending 2 hours opening 7 tabs to compare flights, you just ask. The agent does the rest.
Now stretch that idea. What if the agent helps you buy the right mic for your podcast? Or helps you audit your personal expenses across 3 banks? The potential is crazy.
A Word of Caution
Agents can feel magical. But like any powerful tool, their effectiveness comes down to how well you design them.
You need to:
- Choose the right model
- Give it the right tools
- Set clear boundaries
- And make sure it doesn’t hallucinate its way into serious decisions
TL;DR: This Is Your Moment
AI agents aren’t just hype. They’re the next shift.
We’ve crossed the automation wave. This one’s about autonomy-systems that think, decide, and act.
In the coming issues, I’ll walk you through:
- How to select the right model
- Designing effective toolkits
- Writing clear instructions
- The magic of multi-agent systems
- And setting the guardrails that keep them safe
The wave is here. You just have to decide how you want to ride it.