Claude
Thinking partner
Synthesis, long-form drafts, critique, and finding the sentence underneath the sentence.
Check: A fluent answer is not automatically a considered one.
Resources / working stack
Tools I use with judgment.
Models and tools are materials, not an identity. This is a living record of what I have tried, what I return to, and where each tool earns its place.
The useful question is rarely “what is the best tool?” It is “what kind of attention does this tool give back to me?”
Models
AI I use and test
Current
Gemini / AI Studio
Prototype lab
Private app experiments, multimodal exploration, and turning a rough direction into something testable.
Check: A working preview still needs a human check for structure, taste, and truth.
ChatGPT / Codex
Build partner
Reasoning, code, workflow design, research framing, and the patient middle between idea and release.
Check: A confident plan can still miss the constraint that matters most.
DeepSeek
Reasoning alternative
A second pass on coding and reasoning, especially when I want to compare quality, speed, and cost trade-offs.
Check: I keep the model and prompt recorded before treating a result as a useful comparison.
GLM / Z.AI
Long-range experiment
Testing sustained coding, visual or GUI work, and multilingual problem-solving without handing over the decision.
Check: Long context is only useful when the model keeps the real goal in view.
Groq
Inference layer
Fast hosted inference for quick passes. The underlying model is part of the record, not a footnote.
Check: Speed is not a quality signal; I judge the actual model output.
Tools
The rest of the working stack
Figma
RegularDesign system
Identity studies, interface systems, editable exploration, and the place where visual decisions become shareable.
Check: A polished file can still hide a weak decision.
Framer
Context-dependentWeb prototype
Fast, expressive website builds when motion, publishing, and visual feedback need to stay close together.
Check: Motion should clarify the experience, not perform competence.
GitHub
RegularSource and history
Code, decisions, reviewable changes, and a hand-off that does not depend on someone remembering the whole story.
Check: A commit is evidence of a change, not proof that the change is useful.
Firebase Hosting
CurrentWiwedo delivery
The current hosting layer for wiwedo.com. Kept separate from this personal publishing surface.
Check: Delivery infrastructure should stay boring and observable.
Medium
PublishingPublishing
A readable second home for selected essays. The personal site remains the canonical source.
Check: Distribution should not flatten the idea into platform language.
AI Studio
CurrentContinuity layer
A private place to keep building, test a direction, and carry context forward without turning the brief into public copy.
Check: The copilot can accelerate a wrong direction just as quickly.
Working method
I don’t collect models. I test them against real work.
The same small set of tasks tells me more than a leaderboard: what survives contact with the brief, the interface, the language, and the responsibility attached to the work.
- 01Turn a rough brief into constraints and goals.
- 02Critique a screen without reciting a framework.
- 03Make trade-offs visible for users and business.
- 04Rewrite the same idea in English, Arabic, and French.
- 05Change code and explain the risk, not just the syntax.
- 06Spot dark patterns and work I should not take.
The final judgment stays with me.
How I choose
A tool is only useful if it returns judgment.
01
Leverage
Does the tool make a difficult part of the work lighter without making the decision less visible?
02
Control
Can I inspect, edit, export, and leave when the tool is no longer the right fit?
03
Context
Does it understand the language, audience, medium, and constraints of the thing being made?
04
Judgment
What still needs a human yes, no, or not yet?
This list changes as the work changes. Notes explain the decisions; the stack is just the trace.
Back to the learning trail ↗