Let's put it straight: in March, I committed to building a product, not a sane one. I wanted to solve what frustrated me the most with tools like ChatGPT, Gemini, or shell apps like n8n.
My initial goal wasn't to sell it to customers or pitch it to VCs, but just to ensure I had the tool to tame LLMs, Agents, or even AGI (if it ever ships), so that what had invaded my work life out of necessity would actually work for me and not glitch randomly or just try to trap me.
So there I was, spending almost every second I had building a platform that should allow for:
- Multi-step, multi-modal LLM workflows
- Declarative behavior for AI Agents
- Proper RAG (Retrieval-Augmented Generation)
- Support for any API as a Tool
- Handling cloud provider-level traffic
- Magically splitting the codebase into Open Source and EE (Enterprise Edition)
- ... and, sure thing, doing the core in Go
Now, seven months in, after 60+ hours per week coding and trying to somehow make this more than just my own tooling project, losing my personal money while pulling my hair out, observing market developments, and model-interface providers changing APIs.
I kept discovering more players each month, building some or all of the features on my roadmap. Which kind of validated that there’s no universal solution yet.
A core question always echoed in my mind:
"Is what I'm doing a business?"
Researching and identifying pain points and collaborating with others to identify use cases and verticals beyond my own worldviews.
Hell, I even set aside budgets and evaluated contractors and friends to outsource some coding and essential but non-negligible tasks, for example, copywriting, to ensure I meet my own roadmap...
Since when did I become a PM? Wasn't that a space I always looked down on and never wanted to touch?
naro@xaxen:~/src/github.com/contenox/runtime2$ cloc .
650 text files.
631 unique files.
22 files ignored.
github.com/AlDanial/cloc v 1.98 T=0.40 s (1580.5 files/s, 340245.6 lines/s)
-------------------------------------------------------------------------------
Language files blank comment code
-------------------------------------------------------------------------------
Go 368 9637 9139 53197
JSON 15 2 0 24941
TypeScript 176 1022 159 12702
Markdown 6 2804 66 8572
YAML 6 14 10 5187
Python 25 687 443 3573
CSS 4 45 37 2103
Bourne Shell 3 64 73 378
SQL 2 73 6 285
make 2 32 1 123
HCL 3 19 21 113
Dockerfile 3 23 20 84
SVG 9 8 5 73
JavaScript 6 7 0 69
HTML 1 0 0 13
Text 2 0 0 10
-------------------------------------------------------------------------------
SUM: 631 14437 9980 111423
-------------------------------------------------------------------------------
Here I am, sitting on three major rewrites and a codebase so big I can't navigate using just the file tree.
I achieved my goal. Kinda. Yet never truly released. And I didn't stop there.
While working on this project, nuances about LLMs—their use cases, strengths, weaknesses, and mitigation strategies appeared.
I had to find out what the nature of LLMs is. And soon I was on another mission, validating a new vision:
"I want to define how AI behaves, interacts, and learns."
It seems that, yeah, with the proper implementation, the platform can handle a lot of autonomy-
- it could be used to the point of self-optimizing at runtime
- or let it judge its own performance
- or even to then trigger the generation of content for model fine-tuning.
And since it doesn't require an external model (or can selectively use them), there is no constraint on what it can be used for.
Let's wrap up, me rambling here.
I don't encourage anyone to repeat the same path.
No matter how much, only geeks glamorize the builder path.
Here is what I mean:
I noticed changes and reality catching up; yes, not just minor, fixable things like my bank account's balance, but also longer-lasting changes—from the simple "everyone hates me now because I was so busy" to mindset shifts that can’t truly be undone.
I built the tool I needed, but now it's clear:
This is just the start. And it gets expensive on all fronts from here.
Don't get me wrong. I still don't care that much about monetizing all this effort. But it's clear that without doing so, it will just die.
It’s been clear for a while that my ‘2–3 month’ estimate was a severe miscalculation.
And now what?
What’s the lesson here?
More crucially — what would be different if I had known the game I was playing?
Top comments (0)