How to PoC 105 | A year of agents

How to PoC 105 | A year of agents

May 11, 2026

Tl;DR

Shipping a lifetime in 6M!

Intro

There is live beyond claude desktop and codex app for windows:

winget install Codex -s msstore

Using the latest Agents

Whatever you will choose, make sure to have python and specially node properly installed:

curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash -
sudo apt install -y nodejs
node --version

For python, you can even do it via uv for full pro config:

curl -LsSf [https://astral.sh/uv/install.sh](https://astral.sh/uv/install.sh) | sh
#powershell -c "irm [https://astral.sh/uv/install.ps1](https://astral.sh/uv/install.ps1) | iex"
uv python install 3.12
#uv run script.py

Because uv is written in Rust, it is a standalone executable that doesn’t require Python to run. It can download and manage multiple Python versions (3.10, 3.12, 3.13, etc.) on its own, similar to how nvm works for Node.js.

  1. OpenClaw (ex moltbot, ex clawdbot)

MIT | Your own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞

https://docs.openclaw.ai/start/showcase

  1. https://github.com/nousresearch/hermes-agent

With container setup: https://hermes-agent.nousresearch.com/docs/user-guide/docker and this inspiring video

MIT | The agent that grows with you

https://www.youtube.com/watch?v=jIP0q7HEC0g

mkdir -p ~/.hermes
docker run -it --rm \
  -v ~/.hermes:/opt/data \
  nousresearch/hermes-agent setup
#curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
#hermes model

alt text

ℹ️
I recently formated a Pi for monitoring plants with the camera, and will use it to host these. The setup script made it quick.

Other CLIs

Its been a while since that AI CLI tools post and the vibe coding one with CLI last year:

#get the terminal of your choice ready!
#/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/mkasberg/ghostty-ubuntu/HEAD/install.sh)"
  • Warp

  • OpenCode: which brings some free models (otherwise, BYO API)

#curl -fsSL https://opencode.ai/install | bash
npm install -g opencode-ai
#choco install opencode

alt text

You can also bring commands and skills ofc: https://opencode.ai/docs/skills/

Ive used OpenCode lately to harden my Pi setup script: this will be an excuse to try the chatgpt PRO plan next month, as anthropic doesnt let me use max here

opencode #/models #/connect
ℹ️
Why a Pi again? To harden it and get those agents running at home. No public IP in a VPS, nor messing with my home server.

Agents x Knowledge Management

Dont let AI take away your thinking.

Let AI enhance your thinking.

For this, owning a good context layer is critical!


Conclusions

Expect the compute*tokens/intelligence to keep dropping

Ive heard very good comments about Gemma 4 on one of the new mac air.

You can use this as an opportunity to leverage local AI withing your homelab as I did recently, here.

Whenever you are shifting or force to shift from:

Can you manage the process? thats really over

To: Can you produce outcomes with leverage?

Just reach out:

Whats going on?

AI is coming to the workspace.

And we are about to see which works were more operational and trivial than others: https://github.com/anthropics/financial-services

This thing can make xls/xlsx with DCF ‘models’

Allow me to put model with quotes, bc if you cant bet your ass that it will work, the models sucks

All shaped as plugin > skills »> prompts

ℹ️
Say you have a procedural way to analyze PBI dashboards in .md - You can make a skill and a plugin

MIT | An agentic skills framework & software development methodology that works.

I got to know via https://www.youtube.com/watch?v=utRYkiqGuZI and the intro blog is https://blog.fsck.com/2025/10/09/superpowers/ and latest releases here

Whats Working Whats Not Whats Next

Doing quick PoCs that are useful with almost no effort: *and you get to know new concepts, like XIRR.

#git clone https://github.com/JAlcocerT/poc
cd ./poc/stocks-simple/web
podman build -t stocks-simple . #after 4min build...
podman run -d --name stocks-simple \
  -p 3000:3000 \
  -e ALLOWED_DEV_ORIGINS=192.168.1.18 \
  -v ./transactions.csv:/data/transactions.csv:ro \
  -v stocks-cache:/data/.cache \
  stocks-simple 

Because you saw sth was not right via https://aegis-freedom.pages.dev/ already

How big will be the PoC graveyard by the end of this year?

How many cool data projects with nice charts will never see the light?

alt text

Not sure, we are still working in semantic silos:

While human requirements are loosie loosie and people working at FMCG are still doing human RCA:

git clone /poc
cd ./poc/libg/why-postmortem
#npm run baml:generate #yea, this time it uses BAML
npm run dev #

why why why to actions PoC

http://localhost:5173/

I made it initially as skill

/postmortem:blameless-check

These 2 share the same dna and shaped as per this brd

LayerPM CopilotWhy-Postmortem
DomainForward-looking — define the right productBackward-looking — learn from the incident
ContextBusiness goals, personas (Sec/Legal/DevOps/Finance), domain triggers, success-metric rulesorg.yaml, services.yaml, teams.yaml, severity-matrix.yaml, enums.yaml, glossary.md, history.md
FunnelWhy Engine → Persona Critique → Edge Cases → SMART Metrics → Traceability MatrixTimeline → Fishbone (6 cats) → 5-Whys per branch → Blameless Draft → Action Items
  • Blameless Culture: This is the most important part. If the RCA finds “John forgot to click a button,” the root cause isn’t John—it’s “The system allows a single point of failure without a confirmation step.”

  • The “Action Item” Rule: Every RCA must result in at least one ticket in your backlog (Jira/Asana/Linear). If there’s no ticket, the RCA didn’t happen.

  • AI for Patterns: Once you have 10 postmortems, feed them all to an AI and ask: “What is the recurring theme across our last 10 failures?” This is where you find the “Grand Root Cause” of your team’s friction.

  • https://github.com/nexu-io/open-design

Apache v2 | 🎨 Local-first, open-source alternative to Anthropic’s Claude Design. ⚡ 19 Skills · ✨ 71 brand-grade Design Systems 🖼 Generate web · desktop · mobile prototypes · slides · images · videos · HyperFrames 📦 Sandboxed preview · HTML/PDF/PPTX/MP4 export 🤖 Runs on Claude Code / Codex / Cursor / Gemini / OpenCode / Qwen / Copilot / Hermes / Kimi CLI.

#npm install -g @openai/codex #https://github.com/openai/codex/
codex

Codex cli can now create images, search the web and automate repeatable workflows via the exec command bye n8n :,)

Thanks to the references inside open-design, I got to know another skills:

npx skills add https://github.com/alchaincyf/huashu-design --skill huashu-design

And yep, it makes even better slides as html: can you make a Presentation slides : 1920x1080 HTML deck, can be used as a PowerPoint presentation for the poc we have just built?

npx decktape generic deck.html deck.pdf --size 1920x1080

node .claude/skills/huashu-design/scripts/render-video.js \
--url "file:///C:/Users/j--e-/Desktop/poc/libg/why-postmortem/demo-video.html?record" --duration 32 --out demo-video.mp4

node .claude/skills/huashu-design/scripts/render-video.js why-postmortem/demo-video.html --duration=32

With Huashu design you can easily create a brand kit and generate not only pptx and video.

But also infographic, one pagers…

ℹ️
To demo them publically w/o disclosing your domains you can always use tailscale funnels
htop #btop 
iotop #iftop #ctop

Keep Stop Start - Doing


FAQ

Apps that Inspire

  1. DocuSign DocuSeal

  2. Tiinyhost

Whats interesting?

That you can get inspired by these webs UI/X with google stitch nowadays.

Inspiring apps: all those pwa like that just…work

Full-Stack DApp Boilerplate for Substrate and ink! Smart Contracts

How about building like they’ve done?