Daily drivers
Claude Code
My main tool: ~90% of my projects are coded with it.
HamzaElaamely.
I do technology watch: I test AI tools the day they ship and turn them into value for companies.
- 01StudentBusiness IT · ETML-ES
- 02Junior AI ConsultantCIMO · via BS-Team
- 03BuilderInkly · co-founder
About
Three hats, one conviction
Learning by doing, rather than in theory. That is the thread through everything I take on.
Student.
Business IT · ETML-ES
What I study
As much technical work as marketing, accounting and business management.
Why this path
Technical skill alone isn't enough to understand a company. Management gives me the big picture: how an organisation decides, operates and creates value, well beyond the technical layer.
Junior AI Consultant.
CIMO · via BS-Team
What I do
After an internship at CIMO, my interest in AI led me to support them on integrating Microsoft Copilot, in particular building their own agents.
What I'm learning
The AI-focused business analyst role, in the field: translating a business need into a concrete, useful solution.
Builder.
Personal projects
What I do
I don't launch products, I build. Taking an idea all the way to a working product is what makes me understand a tool in depth.
Why
Building shows me, concretely, what each technology can bring to companies, often before they even start paying attention.
Stack
Tools of the moment
What I rely on today. My essentials up front, the rest as backup.
Code & ship.
Where I write and ship all my code, from prototype to project.
Cursor
IDE to run my code-review agents and automate my pull requests.
Codex
Quick website creation, and some tests / comparisons.
Think & write.
My day-to-day assistant: thinking, tasks, writing.
Daily drivers
Claude
My main LLM: writing, discussions, thinking.
Claude Cowork
To delegate and push concrete tasks forward.
Automate.
To make the repetitive run on its own, with an AI layer on top.
Daily drivers
n8n
My automations, with an AI layer on top.
Agents.
The orchestrators that run my AI agents.
Daily drivers
Hermes
The model that runs my AI agents.
Copilot
Agentic in a professional context, within the Microsoft ecosystem.
Mockups & design.
Designing and refining my interfaces before coding.
Stitch
Generating UI mockups quickly from a prompt.
Figma
Refining and structuring my mockups and design system.
Create media.
Generating images and videos.
Gamma
Building presentations and visual decks quickly.
ChatGPT
For image generation.
Higgsfield
Video creation. It bundles the main image and video providers.
Learn & organise.
Learning fast, and keeping a record that organises itself.
NotebookLM
Learning and source research.
Obsidian + Claude Code
A self-managed wiki, Karpathy-style (Obsidian driven by Claude Code).
My models
The right model for the right job
No single model does it all: I pick the right one for each task.
Claude Opus 4.8
heavy reasoning & codeHeavy tasks, advanced reasoning and demanding code.
Claude Opus 4.7
everydayMy everyday model: thinking and discussion.
DeepSeek v4 Pro
agenticThe advanced reasoning behind my agentic workflows.
ChatGPT 5.5
code (Codex)Some coding in Codex now and then, and to try out new Codex features.
Gemini 3.5 Flash
research & docsResearch and documentation: my starting point before continuing on Claude.
Gemma 4
A powerful open-weight model, currently in testing in my setup.
n8n automation
The trio powering my n8n workflows.
- DeepSeek v4 Flash
- DeepSeek v4 Pro
- ChatGPT 4o
Claude Fable 5
Formidable power, now pulled from the market. A tribute to a great model.
Development
How I code, in practice
Not just tools: a complete system to code fast and clean.
Brainstorming.
What I do
Before writing a line, I frame the idea with SuperPower's brainstorming skill: it turns a hunch into a validated spec.
Why
So I don't head the wrong way. Half the work is knowing what to build.
$ claude› /brainstorming «feature idea»✓ spec validated → docs/specs/feature.mdProjects
What I build
Three projects, three grounds to turn an idea into something real.
Project
Inkly.
Booking appointments without getting lost in direct messages.
An app, currently in development, to book and manage appointments without going through Instagram direct messages, where clients always end up getting lost. With a scoring system and a built-in chat.
- Mobile app
- Booking
- Scoring
- Chat
Project
Orato.
Quotes and site tracking, by voice.
A test project: letting independent construction workers create an instant quote and keep their site up to date, using only voice notes. No more language barrier, no more paperwork.
- n8n
- Voice note
- Instant quote
- Site tracking
Project
Obsidian + Claude Code.
A life wiki that organises itself.
I store my technology watch, my documents and my tweets in Obsidian. With Claude Code, following Karpathy's idea, I turn it into a knowledge base that my AI maintains and links on its own through the links between .md files. The result: Claude Code always has the right context.
- Obsidian
- Claude Code
- Markdown
- Watch
Exploration
My path through AI
Two years of watching the field: from a simple LLM that answers to agentic AI, all the way to the harness that monitors my code quality.
1
August 2024Cursor · early days
DigitAll Service
My first AI-built website, in the very early days of Cursor. The result was modest, but that is where I developed a taste for building with these tools.
November 2024Cursor
Rules, sub-agents, MCP
I follow the arrival of rules, sub-agents and MCP (released in November 2024). With each new feature, I fold it into how I code to gauge what it really brings.
December 2024Ollama
Open-source models
I test open-source models as soon as they ship, locally, to compare their strengths. It is also a credible alternative in terms of cost and privacy for a company.
Early 2025n8n
Automation, with an AI layer
I move into automation and add an AI layer on top. Several prototypes come out of it, including Orato.
5
Mid-2025Microsoft Copilot Studio
AI in the enterprise
Discovered at CIMO: building agents directly inside the Microsoft ecosystem. A different approach to AI, more integrated and more governed.
6
2025Inkly
A real development system
With Inkly, I structure my entire development system: brainstorming plugins, MCP on the database, validation rules, sub-agents that check the code.
7
Late 2025OpenClaw · Hermes · DeepSeek
Agentic AI
I dive into agentic: OpenClaw first, then Hermes and DeepSeek as the models powering my agents. The foundation of my future Life OS.
2026Obsidian · Claude Code
Karpathy & the self-organising wiki
I follow Karpathy closely. His Obsidian-driven-by-Claude-Code method, for a wiki that organises itself, made a strong impression on me.
Watch
How I keep watch
Sifting the noise for the signals that matter, then testing without delay.
01
Sources
X and a handpicked set of YouTube channels, which I check every day: Anthropic, ClaudeDevs and Figma on X, Matt Pocock and AI Engineer on YouTube, plus people like Karpathy. I follow specific sources rather than the trend feed. Source quality is everything.
02
Filtering the noise
Most of what ships is noise or a rehash. I quickly set aside anything that adds nothing, keeping only the genuine developments.
03
Spotting
I look for what shifts things: a new model, a new way of working, a tool that opens a possibility. That is where I stop.
04
Testing
And above all, I put it to use immediately, on a real case. Real use is how you measure a tool's worth, beyond the promises.
Vision
AI, in Switzerland, now
Everyone talks about AI. Few people really say how to make the most of it. Here is how I read it.
Does AI really save you time?
AI is sold as a time-saving machine. The reality is more nuanced. The time it saves, we have often already spent elsewhere: setting it up properly, checking its answers, fixing its mistakes. Not to mention the cost, rarely counted honestly.
That doesn't mean it fails to deliver. It means we still judge it poorly, forgetting all the invisible work behind a result that “works”.
How far can we trust AI today?
I see it as a final-year apprentice. On some tasks it is already autonomous and impressive. On others, it still needs guidance, review, correction. It isn't ready to go solo, but it learns fast when it is well supervised.
The real value, then, isn't in the tool alone, but in what you build around it: agentic systems, workflows designed for a precise use. And that isn't improvised. You have to test, run POCs, accept a few mistakes before it truly runs.
Why is keeping watch a craft of its own?
To make the most of it, you have to keep pace. Tools ship every week, models change, yesterday's best practices are already outdated. Keeping the good, dropping the rest, validating what holds in production: it is a full-time job.
And that is precisely the time most teams don't have. They have a business to run, not a technology watch to carry out on the side.
What card can SMEs play?
In Switzerland, we stay cautious about AI, and I understand it. Nobody wants to take risks on sensitive topics, and large groups already have the resources to ship to production.
But I am convinced SMEs have a real card to play. Less on customer service, heavy in governance and testing, and more on internal automation: the kind that eases teams' daily work and gives them back time for what really matters.
Concretely
What if you let an enthusiast take care of it?
Keeping watch, testing the tools, validating what works and putting it in place: that is my everyday. Enough to free up time for your team, without the risk of carrying it all in-house.
Let's work together →Contact
Let's talk
We don't all come here for the same reason. Tell me what brings you, and I'll tailor my reply.
You're here for