Wow moments with Claude Code

When Code Feels
Like Superpowers

Alexander Gomez · 2026

Saturday · 11:34 AM · Efteling

1:10 — "I've been using AI since the prompting days — back when ChatGPT first came out — and I've watched it evolve fast. Last January I started experimenting with AI Agents. Mid-2025 I began building EMX — the project I'm about to show you — with prompt engineering. The API came together. But I knew the UI was going to take me another couple of months: free time was scarce. Then in February I picked up Claude Code. Agentic Engineering. Suddenly I could move fast — genuinely impressed by what well-planned agents could ship beside me on my own software projects. Within hours the first EMX UI was running. The plan was to demo it Monday morning at the office — show my colleagues of my team what I'd been building. So the build had to be ready Monday. Then I looked at my calendar: the weekend was fully booked. Saturday at Efteling with the family — whole day. I couldn't lose it, and I did not want to push the demo. So Friday night I went to sleep thinking: what if I could SSH from my iPhone into my VPS — and run Claude Code from there? The next morning, this is what was on my phone."

Yes. Kids are in the queue for a rollercoaster my stomach can't handle.
I'm deploying agents from my phone.

Let me rewind.

A photo. A story. A song.
About 4 minutes end-to-end.

visionreads the photo.

composerwrites the song.

imagepaints.

musicplays.

videostitches.

five AI models, working in parallel

For moments.

For memory.

"Papa, kunnen we samen een liedje maken?"

— my son · December

Warm.

808. Brass.

He loved it.

EMX-generated cover for King of the World

Claude Code and I had made a tool that made music my eight-year-old wanted to hear again.

It's on Spotify now. He has an artist page. One single. — that's it.

He's eight. He has a single on Spotify.
I'm now his manager.

Want to see another one made live?

live demo

EMX

emx.firebots.cloud

3:30 — [LIVE DEMO IN CHROME TAB 2 · pre-loaded result] Walkthrough script: 1. "This is the photo I uploaded — a moment from a normal weekend." 2. "This is the emotion pad — I put the dot here for the feeling I wanted." [point to position] 3. "Vision read the photo. Composer wrote lyrics. Image painted a cover. Music played. Video stitched it all together. Four minutes, end to end." 4. "Here's the cover." [show] 5. "Here's the music video." [play ~30s, MUTED — the son's track from earlier is the only musical moment] 6. Return to deck: Cmd+1. DEEP-DIVE (optional): press DOWN ARROW for the architecture diagram — "how it all connects." Use it while the demo is running, or if the room is curious how EMX works. FALLBACK if tab is lost / browser crashed: press DOWN ARROW twice for the static screenshot.

under the hood

EMX — how it all connects

you a photo + a story + a mood

▶

browser emx-web face-api · emotion-pad → BPM

▶

orchestrator BFF Node.js · runs the pipeline

▶

service emx-api Python · vision · compose · cover

▶

ai model Gemini ×3 one call per step

ai service Kie AI model V5 · music, then video

▶

result song + video a finished music video

◀ live progress streamed back to the browser · SSE

one upload → 5 AI calls across 2 providers → a music video

OPTIONAL TECH DEEP-DIVE — reached by pressing DOWN once from the EMX demo slide. Show it while the demo runs, or to answer "how does it actually work?" Walk it left to right: "You give EMX a photo, a story, a mood. The browser reads your face with face-api and turns the mood into a BPM. The BFF — a Node.js service — orchestrates everything. emx-api, in Python, runs three steps — vision, compose, cover — each one a Gemini call. Then Kie AI turns that into a song on model V5, and then a video. And the whole time, progress is streamed back to your browser live over SSE." Land on: "One upload, five AI calls, two providers — a finished music video." Then press UP to return to the demo, DOWN for the fallback screenshot, or RIGHT to continue to slide 18.

fallback

[ fallback static screenshot — EMX result ]

Four minutes. End to end.

Okay. Back to the theme park.

Family selfie at Efteling, Saturday in February

Saturday in February. Pair-programming in a forty-minute queue.

If a CSS fix takes 40 minutes in a queue —
what could a whole app take?

live demo · audience prompt

Vibe Deployer

vibe.firebots.cloud

3:30 — [LIVE DEMO IN CHROME TAB 3 · audience prompt] Walkthrough script: 1. "Shout out a one-line app idea — something small, something fun." [wait 5s for response] 2. If silence after 5s, pivot: "I've got one." [type the locked backup prompt from /opt/wow-long/DEMO-PROMPTS.md] 3. Type the prompt. Narrate as Claude works: "composing the HTML… writing the JS… deploying…" 4. When deployed, click the URL. 5. Show it works — do one interaction (click a button, see the response). 6. Return to deck: Cmd+1. DEEP-DIVE (optional): press DOWN ARROW for the architecture diagram — "how it all connects." Use it while the deploy is running, or if the room is curious how Vibe Deployer works. FALLBACK if WebSocket fails or deploy takes > 90s: press DOWN ARROW twice for the pre-recorded deploy screenshot. Narrate: "And that's exactly what would have happened — here's one I made earlier."

under the hood

Vibe Deployer — how it all connects

you a sentence describe an app

▶

chat ui vibe-chat nginx · the chat page

▶

relay vibe-ws Node.js · the relay

▶

workflow n8n the deploy workflow

▶

ai model Claude writes the whole app

result a live app + a QR to share

◀

host vibe-apps nginx · serves it live

◀ n8n writes Claude's code to disk

one sentence → Claude writes the app → live and shareable in about 2 minutes

OPTIONAL TECH DEEP-DIVE — reached by pressing DOWN once from the Vibe Deployer demo slide. Show it while the deploy runs, or to answer "how does it actually deploy?" Walk the loop: "You type a sentence into vibe-chat — that's just an nginx page. It posts to vibe-ws, a small Node.js relay, which starts an n8n workflow. n8n calls Claude, and Claude writes the whole app — the HTML, the CSS, the JavaScript. n8n writes that code straight to disk, and vibe-apps — another nginx — serves it live. You get back a URL and a QR code." Land on: "One sentence in, a working app out — live and shareable in about 2 minutes." Then press UP to return to the demo, DOWN for the fallback screenshot, or RIGHT to continue.

fallback

[ fallback — pre-recorded Vibe Deployer result ]

live, on the internet, right now

From a sentence. In about 2 minutes.

Terminus tmux · claude

~ tmux attach

$ claude

● SSH from phone

● tmux session

● Claude Code agent

● iOS keyboard

✓ read · write · ship

Terminus · the studio in my pocket

The studio is wherever I am.

Here's what I checked having breakfast this morning.

live · my calorie tracker

OpenClaw

on my VPS · via Telegram

1:30 — [LIVE DEMO IN CHROME TAB 4 · Telegram Web] Walkthrough script: 1. "This is OpenClaw — an agent I programmed and deployed entirely with Claude Code. I use it every day to track what I eat." 2. "Here's today's log." [scroll to today] 3. "Watch this." [type one item from /opt/wow-long/DEMO-PROMPTS.md — e.g. 'one coffee'] 4. Wait ~2-3s. Agent replies with parsed entry + running total. 5. "Counted. Logged. Done. From bed." 6. Return to deck: Cmd+1. DEEP-DIVE (optional): press DOWN ARROW for the architecture diagram — "how it all connects." Use it while you demo, or if the room is curious how OpenClaw works. FALLBACK if Telegram Web won't load: press DOWN ARROW twice for the screenshot.

under the hood

OpenClaw — how it all connects

you a food photo or just a description

▶

chat app Telegram you message the bot

▶

gateway OpenClaw Node.js · agent gateway

▶

skill calorie-tracker a skill I wrote

▶

ai model Claude sees the food, counts it

result your daily total replied in Telegram

◀

log calorie-log.json one line per meal

◀ the agent saves what Claude counted

a photo of a meal → Claude counts it → logged and replied, in seconds

OPTIONAL TECH DEEP-DIVE — reached by pressing DOWN once from the OpenClaw demo slide. Show it while you demo, or to answer "how does it work?" Walk it left to right: "I send a food photo — or just a line of text — to a Telegram bot. OpenClaw, an open-source agent gateway running on my VPS, picks it up. It runs a skill I wrote myself — the calorie-tracker. The skill asks Claude to look at the food and estimate the calories and protein. The agent writes one line into a plain JSON file, and replies in Telegram with my running total for the day. No n8n, no database — just an agent and a file." Land on: "A photo of a meal in, my running total out — counted and logged in seconds." Then press UP to return to the demo, DOWN for the fallback screenshot, or RIGHT to continue.

fallback

[ fallback — OpenClaw Telegram chat ]

Counted. Logged. Done. Over breakfast.

in one year

Three things
Claude Code changed for me.

I wrote an nginx config while my kids ran around the playground.

I made a cinematic hip-hop song I couldn't have made alone.

I shipped production from the back of a rollercoaster queue.

I said three wow moments.
There's a fourth.
And I use it almost every day.

live · YouTube → analysis · /verify · /roadmap

YoutubeSummarizer

on my VPS · via Telegram

2:00 — [LIVE DEMO IN CHROME TAB 5 · Telegram Web] Walkthrough script: 1. "I send a YouTube URL — get back the analysis." [show last analyzed video in the chat] 2. "And I built two commands on top of it that I use constantly:" 3. "/verify — fact-checks the claims in the video against the web." 4. "/roadmap — turns the video into a learning plan I can actually follow." 5. [paste queued URL from /opt/wow-long/DEMO-PROMPTS.md + send /verify] 6. [Wait ~30-55s — narrate during the wait:] "While that's running — this whole pipeline is Codex doing deep research on my VPS. Real web tools. Real sources." 7. Show the /verify result when it arrives. 8. (Optional) Quickly show a recent /roadmap output from the chat history. 9. Return to deck: Cmd+1. DEEP-DIVE (optional): press DOWN ARROW for the architecture diagram — "how it all connects." Use it while /verify is running, or if the room is curious how YoutubeSummarizer works. FALLBACK if /verify takes > 60s OR Telegram fails: press DOWN ARROW twice for the screenshot pair.

under the hood

YoutubeSummarizer — how it all connects

you a YouTube link + a command, maybe

▶

chat app Telegram you message the bot

▶

orchestrator n8n the YoutubeAnalyser flow

▶

step 1 yt-transcript grabs the captions

step 2 · summary Fabric AI patterns analyse it

step 2 · commands yt-research /verify · /roadmap

▶

ai model Codex gpt-5 · searches the web

◀ the finished analysis is sent straight back to your Telegram chat

a link in → an agent reads the whole video → an answer back in your chat

OPTIONAL TECH DEEP-DIVE — reached by pressing DOWN once from the YoutubeSummarizer demo slide. Show it while /verify runs, or to answer "how does it work?" Walk it left to right: "I send a YouTube link to a Telegram bot. That triggers an n8n workflow — the YoutubeAnalyser. First it calls yt-transcript, a small service that grabs the video's captions. Then, depending on what I asked: a normal summary goes to Fabric, which runs AI analysis patterns over the text. My two commands — /verify and /roadmap — go to yt-research, which runs Codex with live web search, to fact-check the video or turn it into a learning plan. The finished analysis lands straight back in my Telegram chat." Land on: "A link in, a real analysis out — an agent reads the video so I don't have to." Then press UP to return to the demo, DOWN for the fallback screenshots, or RIGHT to continue.

fallback

YoutubeSummarizer /verify + /roadmap screenshots

[ fallback — /verify + /roadmap outputs ]

An agent that reads YouTube so I don't have to.

Old setup: a laptop on a desk on a weekday.
New setup: a phone deploying agents in a queue on a Saturday.

Monday. Your innovation Sprint.

The gap isn't skill — it's the first prompt. Pick one ticket, describe it out loud, and let an agent start.

Speaker by the sea using Claude Code on iPhone

Soon, engineers will deploy agents from anywhere we are.

Thank you

GitHub@orchestratedbyalex

LinkedIn/in/lexgomez

When Code FeelsLike Superpowers

Yes. Kids are in the queue for a rollercoaster my stomach can't handle.I'm deploying agents from my phone.

Let me rewind.

A photo. A story. A song.About 4 minutes end-to-end.

Warm.

808.

Brass.

He loved it.

He's eight. He has a single on Spotify.I'm now his manager.

Want to see another one made live?

EMX

Four minutes. End to end.

Okay. Back to the theme park.

If a CSS fix takes 40 minutes in a queue —what could a whole app take?

Vibe Deployer

From a sentence. In about 2 minutes.

The studio is wherever I am.

Here's what I checked having breakfast this morning.

OpenClaw

Counted. Logged. Done. Over breakfast.

Three thingsClaude Code changed for me.

I said three wow moments.There's a fourth.And I use it almost every day.

YoutubeSummarizer

An agent that reads YouTube so I don't have to.

Old setup: a laptop on a desk on a weekday.New setup: a phone deploying agents in a queue on a Saturday.

Monday. Your innovation Sprint.

Thank you

When Code Feels
Like Superpowers

Yes. Kids are in the queue for a rollercoaster my stomach can't handle.
I'm deploying agents from my phone.

A photo. A story. A song.
About 4 minutes end-to-end.

He's eight. He has a single on Spotify.
I'm now his manager.

If a CSS fix takes 40 minutes in a queue —
what could a whole app take?

Three things
Claude Code changed for me.

I said three wow moments.
There's a fourth.
And I use it almost every day.

Old setup: a laptop on a desk on a weekday.
New setup: a phone deploying agents in a queue on a Saturday.