I Asked My AI Agile Team to Change a Button — and Watched It Happen Live

A new walkthrough video. I open the read-only monitor, type one plain sentence asking to change the background color of the buttons on my git-analytics dashboard, and then I just watch. An autonomous SAFe team — Product Manager, Architect, Developer, QA, DevOps — picks the request up and runs it, narrated in plain English, one transition at a time. Everything I do, I do from a seat that can read everything and change nothing. And the same read-only key is below, so you can pull up a chair and watch the live team yourself.

Watch the Walkthrough

Eleven minutes, narrated end to end. (Replace the placeholder link in the embed below with the published YouTube URL.)

What You’re Watching

The window is agentic-nets.com/monitor, booted in read-only mode. The model behind the glass is safe-teams — a complete Scalable Agile team encoded as software: a Product Manager, an Architect, a Developer, a QA engineer, DevOps, a Scrum Master, a Release Train Engineer, and an Operations expert. Every one of those roles is an AI agent. When you arrive, they’re idle, waiting for work.

What makes this safe to put on a public blog is the credential. I’m logged in with a read-only token. It can read every net, every token, and every step the team takes — and it can change nothing. The gateway enforces that: reads pass through, writes hit a wall.

One Sentence, and the Team Takes It

I don’t open a ticket. I don’t write story points. I don’t attend a standup. I type the request the way you’d ask a colleague: “Let’s create a new feature — I would like to change the background color of the buttons on my dashboard of git-analytics.” The read-only Domain Expert takes its single permitted action — it files that request, word for word, into the Product Manager’s intake — and then the team wakes up.

From there you watch it flow, narrated as it goes: the Product Manager turns the sentence into a backlog story; the Architect decomposes it and names the files; the Developer plans the concrete change — its own stated goal, visible in the feed, is to create a dev-plan for the CSS button-color update story — and QA validates it at the commit-or-fail gate, where a git push is the deploy. From “I’d like to change a color” to a planned code change, with no human in the middle translating between the two.

The Walkthrough, Chapter by Chapter

0:00 — A team with no people on it
0:37 — The read-only seat: read everything, change nothing
1:10 — Asking for a feature in one sentence
1:36 — The Domain Expert files the request
2:08 — The work is the diagram
3:05 — A Petri net that talks (the Console Story)
4:05 — Architect → Developer handoff
5:15 — Tracing the request to a backlog story
5:55 — From a color to a commit (the Developer)
7:05 — The CSS button-color dev-plan
8:05 — The commit-or-fail gate (QA)
9:05 — An honest live window
10:30 — Pull up a chair: try it yourself

Why the Key Is Safe to Hand Out

A request filter sits in front of every route. Its rule is brutally simple: if the scope is read-only and the method isn’t a read, reject it.

Always allowed: GET, HEAD, OPTIONS — the entire read surface.
Always blocked: POST, PUT, PATCH, DELETE — returns 403 {"error":"readonly_scope"}.
One narrow exception: the read-only Domain Expert chat — asking a question needs a POST but mutates nothing in the net.

Try It Yourself

Here is a working read-only key. It can read the safe-teams net and chat with its Domain Expert — and nothing else. (It’s rotated periodically; if it stops working, that’s why.)

Open agentic-nets.com/monitor
On the login screen, tick “Read-only access” (it auto-ticks when you arrive from /monitor)
Paste the read-only secret:
07a9af1d663f899f79f08ca56050a977d41472e34cc0dd0f74abe046446f78f9
Watch the Console · Story feed, then open Agents and ask the Domain Expert “What am I seeing?”

Prefer the terminal? The same key drives the API directly — mint a token and tail the live feed:

SECRET=07a9af1d663f899f79f08ca56050a977d41472e34cc0dd0f74abe046446f78f9

TOK=$(curl -s -X POST https://agentic-nets.com/oauth2/token \
  -d "grant_type=client_credentials&client_id=agenticos-readonly&client_secret=$SECRET" \
  | python3 -c "import sys,json;print(json.load(sys.stdin)['access_token'])")

# Snapshot of the last 20 narrated events
curl -s -H "Authorization: Bearer $TOK" \
  "https://agentic-nets.com/api/event-line/safe-teams?limit=20" | python3 -m json.tool

The narration in the video is a single take over the real recording — what you hear is a description of exactly what’s happening on screen, nothing staged or re-cut. The read-only token, the Console Story view, and the gateway’s read-only enforcement are all in production today.

Related reading: From One Sentence to Live Deploy in 2:44 for what a single story does end to end, and Executors and Deployment Architecture for the polling executor that runs Claude Code inside a command transition.