Talent Acquisition. Solo builder. My AI agents run my online micro businesses. Daily build in public: what works, what breaks, and the fixes. Follow along.
Seattle, WA
Joined October 2019
- Tweets158
- Following72
- Followers1K
- Likes914
Pinned Tweet
For the next 90+ days, I'm growing this site in public: a new bot every day, and real traffic and signup numbers every week.
Get the weekly copy-and-use playbooks here:
agentplaybooks.dev/
Follow along.
I'm shifting my build in public from daily challenge bots to my actual business.
It's called Agent Playbooks: one problem, one bot team.
91 Hermes Agent bots, each built to fix one real problem. Every bot gets its own page and its own blog post, one a day. 🧵
Day 2. You asked what broke first in my AI-run trivia channel.
1. The header said TRMA. In a bold font, the I and V touched, so TRIVIA read wrong for two versions.
2. A bison showed up on the bees question. The renderer matched photos to questions by order, not by name. Wrong photo on 4 of 5 questions.
3. Audio peaks, every week. Splicing and encoding keep pushing the sound over the limit, so now a check runs before every export.
16 versions in 2 days before the first video got approved.
Which would you watch first: classic trivia, Find the Odd One Out, or a retro 8-bit bonus round?
My 9-minute workout video rendered in 13 seconds with plain Python on my Mac's CPU.
Remotion took 56. And turning on the GPU made Python slower.
I tested 4 ways to render the same video:
Python + ffmpeg: 13 s
Revideo: 28 s
Remotion: 56 s
HyperFrames: 85 s
🧵
Why the GPU lost: I split the render across all 10 cores, and they all fought over the Mac's one hardware encoder.
The frameworks lose time drawing every frame in a browser and capturing it.
The full 33-minute version should take about 50 seconds.
What do you use to render video in code? Reply with your stack and your render time.
Follow along, I'm building this in public.
Day 1 of building in public.
I'm a solo builder. I use Hermes Agent to run a team of AI bots for my online businesses: a trivia video channel, a website, and more.
Every day I'll share what I built, what broke, and the templates on GitHub.
Follow along if you're building with agents too.
I hope @google open this to AI Pro subscribers.
We launched the Gemini API CLI. A CLI for calling the Gemini API. Give it to the agent you already use (Claude Code, Codex, Antigravity, Cursor, OpenClaw, Hermes...). It's NOT a new coding agent you can use it too:
1. Generate images, video, speech, music.
2. Analyze PDFs, images, audio, video, YouTube URLs,
3. Transcribe recordings to md / json / srt,
4. Run Deep Research in the background.
Stateless. No memory, no chat mode. Your agent owns the context and the plan, gemini-api does the API call. Build for Agents, it's self-describing. `--help`, `--dry-run`...
Start here: aistudio.google.com/learn/in…
I'll be using this with my current bots setup. Congratulations @NousResearch for this achievement.
With the new Advanced Delegation modes plugin, Hermes Agent's orchestrator can now pick a different model or profile for each subagent from routes you set up. The right model or profile for every job.
hermes-agent.nousresearch.co…
My Hermes Agent today's build:
• Raced 4 video tools on my Mac. Same script, same voice. My own Python pipeline won: 17 sec vs ~35
• Found iCloud folders made renders ~10x slower
• Drafted a contract for my first website client
• Kept my open loops updated
#hermesagent