AI Journey

I use AI the way I used Lean — as a tool, with a human reviewing every line.

Every project is a chance to learn where these tools genuinely help and where a human touch still matters most. I'm building in the open — this page tracks the toolbox, the milestones, and what I'm learning next.

How AI fits into the process

The same four steps, every time.

1 Idea (you)
2 Draft (ChatGPT / Claude)
3 Build in Cursor
4 You review, then it ships

"If I cannot explain the change, it does not go live."

Tools I'm exploring

A rotating toolbox of AI assistants I use for writing, coding, and problem-solving.

Daily
Claude Anthropic

What it is

Anthropic's AI assistant, built for careful reasoning, writing, and code — with a large context window for working across many files at once.

What I use it for

My main pair-programmer. Architecture decisions, writing and refactoring code, and most of the actual building on this site.

Daily
ChatGPT OpenAI

What it is

OpenAI's general-purpose AI assistant — one of the most widely used chatbots for writing, research, coding help, and everyday questions.

What I use it for

Everyday writing, brainstorming, and quick research — usually the first place I go with a loosely-formed idea.

Daily
Grok xAI

What it is

xAI's AI assistant, built into X (Twitter), known for a more direct, less filtered tone and awareness of real-time trends and conversations.

What I use it for

A second opinion when I want a blunter, less hedged take — or to cross-check how a different model reasons about the same question.

Daily
Cursor Anysphere

What it is

An AI-native code editor (a fork of VS Code) with chat and autocomplete built directly into the development workflow.

What I use it for

In-editor pair programming — fast iteration on layout, CSS, and JavaScript without leaving the code.

Referee
Gemini Google

What it is

Google's multimodal AI model, integrated across Google's products, capable of working with text, images, and code together.

What I use it for

Cross-checking Claude's reasoning on trickier problems and getting a genuinely different perspective before committing to an approach.

Daily
Microsoft Copilot Microsoft

What it is

Microsoft's AI assistant, built on OpenAI's models and woven into Windows, Edge, and Microsoft 365.

What I use it for

Comparing suggestions against more traditional, Microsoft-centric tooling — useful when a project touches that ecosystem.

Daily Local
Ollama Ollama, Inc.

What it is

An open-source runtime for running large language models entirely on your own hardware — no cloud, no API calls.

What I use it for

Running Gemma 4 (12B) locally for anything involving private or sensitive data, and as a fast daily driver that never leaves my machine.

Local lab

Mac Mini · Ollama · Gemma 4 (12B) — sized to run comfortably on the hardware in front of me, not in someone else's data center. Client files never leave this desk.

Claude, ChatGPT, Grok, Cursor, Gemini, Microsoft Copilot, Ollama, and their respective logos and trademarks are the property of their owners — Anthropic, OpenAI, xAI, Anysphere, Google, Microsoft, and Ollama, Inc. They're shown here purely to identify the tools I personally use; this does not imply any affiliation, sponsorship, or endorsement by those companies. Links open each provider's own site in a new tab.

My journey so far

A few milestones along the way.

August 2026

Handing real tasks off to Grok Bot

This is xAI's agent platform, not the Grok chatbot — worth keeping straight. Instead of asking it a question and reading the answer, I hand it an actual task and walk away. It runs on its own persistent cloud desktop, logging into the same sites and tools I'd use myself, and only checks back in when it needs my sign-off. I've started running a few in parallel and showing one a workflow once so it can repeat it — genuinely useful for the boring recurring stuff like inbox triage, expenses, and reproducing bugs. Still early and in beta, bundled into the paid SuperGrok and Cursor plans, but it's shifted how I think about what an "AI assistant" even is.

June 2026

Going local with Ollama

Picked up an Apple Mac Mini and started experimenting with local AI models — no cloud required. Ollama running Gemma 4 (12B) has become a genuinely fast, capable daily driver, running entirely on my own hardware. Just as important: it opens up a way to work with private or sensitive data entirely within my own, fully controlled environment, instead of sending it off to a cloud AI service.

Since May 2026

Deepening prompt-engineering skills

Now focused on getting more deliberate about how I work with AI — better prompts, better judgment about when to trust the output, and figuring out where it fits into a real product workflow.

See the post

January 2026

Building with Cursor and Claude

Brought AI into my actual workflow — using Cursor and Claude to speed up front-end development and explore ideas faster. This very site was built with AI assistance.

See the post

Mid 2025

Cross-checking with Gemini, Grok & Copilot

Started comparing how different models reason through the same problem, learning to spot where each one's strengths and blind spots show up.

See the post

June 2023

First conversations with ChatGPT

Started experimenting with AI for everyday writing, brainstorming, and researching — mostly out of curiosity about what it could actually do.

What AI doesn't get to do

The trust rules behind every project, no exceptions.

Prompt → result

An illustrative example, not a real transcript — three draft directions, then the human last mile.

Prompt: "Write a 12-word eyebrow for a locksmith in Bad Kreuznach."

Draft A

"Trusted metalwork craftsmanship for homes and businesses across Bad Kreuznach."

Draft B

"Precision locksmithing and steelwork, proudly serving the Bad Kreuznach region."

Draft C

"Railings, staircases, steel — local locksmith craft built to last generations."

Edited final line

"Balcony railings, staircases, and stainless steel — built in Bad Kreuznach, for the long run."

Currently learning

Prompt engineering, AI-assisted debugging, and knowing when not to use AI at all.

Favorite so far

Claude and Cursor, for how naturally they fit into an actual coding workflow.

Using it for

Faster front-end iteration, drafting copy, and exploring ideas before committing to one.

Changelog

Updated whenever a tool, model, or habit changes.

What you can hold me to.

  • I will tell you if a page was drafted with AI.
  • I will not send your content, photos, or customer lists to a cloud model without asking.
  • Every line is reviewed by me before it goes live.
Get in touch