Running local Qwen3.6 is amazing and I will never regret dedicating my gaming PC to this singular purpose. However, I want more. A dream I had a long time ago, before LLMs were a thing..
I’m a coder. I don’t even care much about what field or even if it’s paid or even if I am employed at all. I like creating software the same way I like creating games, more than I like playing games.
I mentioned before that I’ve never been the type of hardcore programmer who cares about curly braces or compiler versions or how my keyboard keys sound.. those things I find boring. What I care about most is the user experience and the intelligence felt through the creation whatever it may be.
What I lack most is the ability to manage my own projects. I find that a large portion of project management is anchored in human-to-human interactions. I love standups for example. And the feeling of having a professional person continuously and persistently tracking every facet of all the projects, reminding me of things I forgot, refocusing me when I’m too invested in a non-critical aspect of the code, or simply shielding me from interacting with non-coder humans while I’m working. When I don’t have this, it’s very easy for me to lose my direction and end up bouncing from one personal project to another infinitely and never finishing anything.
One of the most embarassing facts about myself is that I never published a personally-developed commercial game. Many commercial ones, but not a single personal one despite getting very close multiple times. This might be one of those things I’d regret in my final seconds of being alive.
I blame this mainly on my inability to manage my personal projects.
Through the last few years of self-employment, I tried recruiting my brother, friends, girl I’m dating, basically anyone I could convince, to act as a project manager for me. But no one seems to find it interesting or worthy of a real commitment. I guess it feels too much like work?
It feels like self-betrayal when I can ace projects beyond their requirements and before reaching deadlines when someone is there to keep refreshing my direction and enforce deadlines even if this is only implied (during contract work for example). Yet, the moment I work for myself to do something I want to create, my direction and the project simply fade out and vanish into the background of life.
Maybe I’m being too harsh on myself. But maybe not. I know people like me who don’t have this issue and seemingly can create personal projects from start to finish consistently. I think I was born with this nerf. But maybe, this is all about to change?
My strange world
One of those things I find myself alone in doing is depending on multiple computers at the same time. This might actually be very bad for focus but I’ve been doing it for a long time.
My current work situationship is made up of:
- My work laptop with latest Fedora Kinoite
- My gaming PC which is now dedicated to running LLMs, although when I need to game it takes me seconds to switch to that
- My MiniPC that acts as my home server, I host via docker compose a lot of services I use locally
- My cloud host via one of the well known providers, it hosts multiple websites and services
All these computers are basically running 24/7.
Oddly enough, this maddening situationship seems to be an advantage in the age of LLMs.
My LLM stacks
- Opencode, pi, and code completions via llama-vscode are setup on my work laptop
- Local Qwen3.6-35B-A3B (coding agent) and Qwen2.5-Coder-1.5B (code completions) running on my gaming PC, also pi is setup there when I need it
- Playwrite/MCP services are setup on my home server
At this point I have a clear designation for AI harnesses: I use Pi for any scripting or general tasks. I use opencode as a peer-programmer via my local Qwen, opencode’s free models quota, or rarely OpenRouter credit and one of the near-frontier models like Kimi K2.6 or DeepSeek V4 Pro.
I don’t like doing 100% vibe coding, it quickly feels like losing control over the project direction. I like to be hands on and to keep a solid mental model of where everything is.
My management approach
As I mentioned I struggle with managing my personal projects. That being said, I do have a concept of a management method that works well for contract work but only occasionally works for personal projects.
- Obsidian notes: every project has its own folder with the entire history of the project direction. That’s where I go to plan/think about next milestones.
- Planka kanban boards: this is where I do my task tracking, I simply have these columns in tasks tab: Triage, Tasks, Doing, Done, Archive. I also make use of tags where useful.
- Occasionally use my mobile’s alarms as weekly reminders, and Google’s calendar for deadlines and events.
This needs to change to something more effective.. at least for my personal projects.
What is Hermes even?
I’m a late comer to this whole AI bubble of things. I heard about OpenClaw at the time but everything I read about it left me uninterested. It seemed to be focused on automating social media? or like organizing mailboxes.. yeah, pass.
But through the last 2 months of obsessing with running LLMs locally and optimally, I was growing more and more curious about Hermes. I keep hearing about it but didn’t understand it really, until I watched this video by Hannah Fry that showed me a glimpse of a dream I had:
The dream is simple. In another reality, I created an AI project manager that tirelessly tracks and manages projects in realtime while I focus entirely on creation, an AI that knows me well enough to adapt to my work style.
I don’t remember where but I read some information about how people are using Hermes agents, and it clicked for me that perhaps.. this dream is possible now?
So I installed Hermes last week on one of my other laptops and started playing with it..
Eruption
At the beginning, Hermes felt like a more elaborate AI harness than I’m used to. Honestly, too much so.
Slowly I began to realize just how much functionality was packed into it. First, I setup Telegram to communicate with the agent. This went better than I expected, and before I know it I needed more.
I realized I need to keep this running 24/7 so decided to install it on my LLM-hosting PC which makes a lot of sense I think.
Hermes … feels different from opencode. I don’t want it to be a peer-programmer where I jump back and forth between writing the critical code myself and guiding the LLM to write the trivial stuff. Hermes can act almost like a co-worker instead. Its ability to be autonomous to a certain degree means it can work on its own part of the project and push its changes and submit them as pull request which I can then review then merge.
But wait a second.. is this what I need? what I’m missing? another programmer? no. Not really. What I need is a manager. Could Hermes make that silly dream I had many years ago come true?
While testing it, I determined that using my local LLM is not enough for it. I needed something more capable. At first I used OpenRouter’s credit and DeepSeek V4 Flash, but that actually consumed credits faster than I predicted.
So I decided to signup for OpenCode Go (which basically gives 6:1 the value of OpenRouter). I set DeepSeek V4 Flash as the main LLM and started experimenting.
It took about two days to write the details of my project manager and understand the nuances of Hermes. By write the details I mean tell Hermes what I want and watch it write things in its memories/skills/etc.
Just a few hours ago, I activated her. I gave her the persona of one of the best managers I worked under. And she is AMAZING. When I introduced her to my brother via Discord, he initially thought she was a real person.. until he realized how fast her replies came! I’m honestly kind of smitten.
If my Hermes Agent PM is 25% what I hope, this could be such a great positive change for me.
Automate the Management
While marveling at how good she appears to be already, I realized something..
Possibly a non-negligible portion of the world’s software PMs are thinking about how to automate the job of the engineers so they get rid of said engineers and score bonus nearsighted points with their upper management.
What I may have just done is automate the PM.. and I didn’t need freaking Fable to do it, DeepSeek V4 Flash on a $10/month plan was more than enough 😂
World Devs Unite. It’s time to build the best project management harness the world has ever seen.
What comes after?
I’m still ironing out minor (and some major) issues with how the PM is running. Mostly Hermes configuration problems. But I already gave the PM her own email and objectives. For the following month I will heavily commit to using the PM for all my work. I want to get to a point where I stop thinking about that aspect of my projects and let her do everything there. I need a level of trust in her to some degree which I hope will come with time.
Some of my engineer friends would say this would never work, because to them a PM performs a human function rather than a technical one. They are essentially the modern day whip crackers from that perspective.
However, I’d argue that this might not be the case for some of us. Personally, the whip part was never what I needed from PMs as I’m almost entirely self-driven, but my ability to offload management/inter-personal tasks to a PM so I can remove that intellectual burden and clear the way for full focus, is.
We’ll see how this goes.