Why I Built Agency
I didn’t build Agency because prompts were annoying.
I built it because I was tired of rebuilding the same damn agent every time.
New project? Cool.
Now define tools.
Wire logic.
Handle variables.
Add some API calls.
Throw in a loop.
Glue everything together.
Congrats, you just rebuilt the same system again.
Even with Claude Code doing most of the coding, it still felt repetitive. Faster, sure. But still the same mental overhead.
At some point I realized: I’m not building products.
I’m rebuilding agent scaffolding over and over.
“But tools like n8n exist”
Yes. n8n exists.
And it’s good.
But it’s built around connectors. APIs. Integrations.
LLMs are just… another node.
That’s not how I was working.
For me, the LLM is the system. Everything else is just support.
So I didn’t want a workflow tool with AI added on top.
I wanted something where the whole thing assumes:
“of course there’s an LLM in the middle of this.”
The real problem
It wasn’t prompts.
It was rebuilding this every time:
- tool calling logic
- variable passing
- branching
- loops
- formatting outputs
Over. And over.
Even when AI writes the code, you’re still designing the same pipeline each time.
That’s the part I wanted to delete.
So I stopped writing agents
And started drawing them.
That’s basically Agency.
You drag nodes, connect them, name outputs, reuse them anywhere with {{varName}}, and that’s it.
No hidden magic. No “where did this value come from” moments.
Everything is explicit. Everything is visible.
What’s actually different
A few things that sound small but aren’t:
- Variables are first-class. You name them once, use them anywhere.
- No implicit data flow. If something is used, you can see it.
- LLM is the default, not an add-on.
- It runs locally. No accounts, no billing layer on top of your API.
I also wanted this to run on the same Groq budget I was already using, not stack another “workflow SaaS tax” on top—frontier models are already expensive, and I’m not running a data center from my laptop.
Also: it’s fast. No waiting for some cloud workflow to wake up.
The feel of it
This is the part I didn’t expect.
It stops feeling like automation.
And starts feeling like sketching ideas.
You try something, tweak a node, rerun. No friction. No “ok let me rewrite this file real quick.”
You just… move things around.
What this is (and isn’t)
It’s not trying to be enterprise anything.
No 500 integrations.
No “platform”.
No dashboard with 12 tabs.
It’s just:
a place to build agent flows without rebuilding them every time
That’s it.
Why it exists
Because I got tired of this loop:
idea → build agent → repeat from scratch
Now it’s:
idea → draw flow → done
Same power.
Way less friction.
Still private. I’ll post a link when there’s something real to play with.