⚠️ This blog post was created with the help of AI tools. Yes, I used a bit of magic from language models to organize my thoughts and automate the boring parts, but the geeky fun and the 🤖 in C# are 100% mine.

Hi!

In 2025, I built more AI agents in .NET than I expected.

Some were demos.
Some were prototypes.
Some (few) actually made it to production.

And while models kept getting better every month, the hardest problems I ran into had very little to do with AI itself.

They were engineering problems.

Here are a few lessons I wish I had internalized earlier.


1. State management beats prompting

Good prompts help.
But once an agent has memory, tools, retries, and multiple steps, state becomes the real challenge.

Where does context live?
What gets persisted?
What gets reset?

Most bugs I hit were not “bad model answers” — they were broken state.


2. Orchestration matters more than intelligence

A single smart call is easy.

Real agents require:

  • sequencing
  • retries
  • branching logic
  • fallbacks

The difference between a demo and a system is orchestration.

This is where clean C# code and clear flows matter more than model choice.


3. Local testing saves weeks, not hours

Running agents locally, with fast feedback loops, changed everything.

It made failures obvious.
It made logs readable.
It made iteration cheap.

Cloud is essential — but local-first development is still your best friend.


4. Observability beats clever prompts

If you can’t see:

  • what the agent did
  • why it did it
  • where it failed

You’re guessing.

Logs, traces, and metrics turned vague “AI issues” into debuggable problems.


5. Production agents fail in very human ways

They misunderstand.
They loop.
They get stuck.
They make assumptions.

None of this is surprising — unless you expect magic.

AI agents are not magic.

They are distributed systems… with opinions.


Closing

The biggest lesson from 2025?

Building AI agents is less about AI
and much more about engineering discipline.

Models will keep improving.
These fundamentals will still matter in 2026.

Happy coding!

Greetings

El Bruno

More posts in my blog ElBruno.com.

More info in https://beacons.ai/elbruno


One response to “What Building AI Agents in .NET Taught Me in 2025”

  1. […] What Building AI Agents in .NET Taught Me in 2025 (Bruno Capuano) […]

    Like

Leave a comment

Discover more from El Bruno

Subscribe now to keep reading and get access to the full archive.

Continue reading