Nick Babich

Product designer. Editor-in-chief of @uxplanet

[Nick Babich @101babich]

Why AI Model Effort Customization Is Bad UX

You can find effort customization in both ChatGPT and Claude, and it is truly horrible UX because this control shifts the burden of thinking from the tool to the user.

The whole promise of AI tools is “Just tell me what you want, and I’ll figure out the best way to do it.”

But effort controls basically say: “Before I help, please decide how hard I should think.”

[Nick Babich @101babich]

@zeroxyz Step 4: Fixing visual defects in design

As you probably saw, we have a few noticeable layout issues such as incorrect spacing and element alignment. I'll fix it wth a final prompt.

And now we've got this final page hosted on Zero https://t.co/6ZugjWxDzh

[Nick Babich @101babich]

🪄 It's time to design my personal website, and to do that, I'll use Zero with Codex.

@zeroxyz is a search layer for your AI that gives it access to thousands of tools, APIs, and services. In my case, it will help Codex to build and deploy a real website.

The whole process of creating this page took me about 30 minutes, and I went from a general idea to a real web page hosted by Zero for free (https://t.co/pBQ4zgYCsz)

Let me show you the entire 4-step process step by step.

[Nick Babich @101babich]

Step 3: Adding real content to the page

One thing that you can notice is that the page has very general information about me. Let's fix it with a follow-up prompt, asking Zero & Codex to collect information about me online and use it in design.

The updated design is much better now, as it clearly tells who I am and what I do. Now we can focus on visual polishing.

[Nick Babich @101babich]

Figma Config 2026 Recap for Product Designers

https://t.co/cq073ASfEP

[Nick Babich @101babich]

💡 Figma Config 2026 Recap

Config 2026 was less about UI design and more about general design practice.

Figma is moving away from the idea of “AI replaces the designer” and toward something much more practical: AI as a creative layer inside the current design workflow. At some point, I thought that Figma was trying to build a simple version of the Adobe toolkit for creatives.

The most useful 4 Figma updates for product designers:

1️⃣ Figma Motion: Create expressive animations with timelines and reusable systems.
2️⃣ Shaders: Generate rich visual effects, image fills, mesh gradients, and material-like styles.
3️⃣ Generative Plugins: Prompt Figma AI to build custom plugins tailored to your workflow.
4️⃣ Code Layers: Build and preview coded UI directly on the Figma canvas.

You can see how all features except Code layers work in the attached video 👇

[Nick Babich @101babich]

Claude Code performance optimisation

https://t.co/sOfUPFAlrh

[Nick Babich @101babich]

💡Claude Code Performance Optimization Commands

Claude Code efficiency is not about using fewer commands. It is about keeping the right level of visibility and control.

Use the following 9 commands to improve your Claude Code performance

[/]statusline for always-on visibility.
[/]context to understand what fills the context window.
[/]compact to reduce context size while continuing the same task.
[/]clear to start fresh when switching to a new task.
[/]usage to track cost and usage patterns.
[/]insights to improve your workflow over time.
[/]tasks to manage background work.
[/]memory to keep long-term context clean.
[/]usage-credits to avoid painful interruptions when you hit limits

👇

[Nick Babich @101babich]

Practical guide on what to do if your Claude skill is not working as expected

https://t.co/2uIb65Fgsk

[Nick Babich @101babich]

💡 Troubleshooting Claude Code Skills

Claude Skills are reusable instructions that teach Claude Code how to handle specific tasks efficiently. Skills help tailor Claude Code’s performance to your specific needs and the processes you or your org follow.

If you create a new Claude Code Skill and it is not working as expected, you are likely facing one of these three issues:

1️⃣ The skill is not triggered

This could happen because of the wrong folder structure, incorrect SKILL[.]md naming, formatting issues, or a weak description that doesn’t match the user prompt.

2️⃣ The wrong skill is triggered

This is usually caused by overlapping skill descriptions or priority conflicts between Enterprise, Personal, Project-specific, and Plugin skills.

3️⃣ The skill fails during execution

This often happens because the skill depends on external packages or tools that weren’t clearly defined in the skill description.

👇

[Nick Babich @101babich]

💡“Black box” problem in AI automation

Workflow mapping is the first thing you do when you create a new AI agent.

This task seems like a simple exercise in theory, but in practice, many organisations have poorly defined & poorly documented workflows.

In such workflows, business management often knows only the input and the output: they know what starts the process and what the final result should look like, but what happens between the two is often unclear.

A typical “black box” situation.

And this is where many businesses make the wrong assumption. They think AI can automatically uncover this black box for them.

But even the best AI tools for workflow automation, like Akai from Deel (https://t.co/AYMX0HDTcq), still ask you to show the workflow. You have to walk through the workflow once or twice, describe what you want in plain language, and show what happens at each step: which systems are used, who makes decisions, where approvals happen, what exceptions exist, and what data moves from one place to another.

Only then can AI map every step, learn your systems, build the required connectors, and create the workflow. Without this walkthrough, you simply cannot create proper automation.

That’s why, in my experience, the hardest part of AI automation is not always the AI part; it’s connecting the dots and understanding what is really happening between input and output.

Because if your workflow is a black box for humans, it will be a black box for AI too.

[Nick Babich @101babich]

📕Complete guide for optimizing MEMORY[.]md for Claude Code projects

https://t.co/zkvqsJBAtl

[Nick Babich @101babich]

💡Second most important file in Claude Code project: MEMORY[.]md

Most Claude Code users know about CLAUDE[.]md, but not everyone is familiar with MEMORY[.]md.

While CLAUDE[.]md tells Claude how to operate, MEMORY[.]md tells Claude what to remember.

In product design projects, it stores long-term project knowledge:
✅ Product decisions
✅ Technical lessons
✅ UX findings
✅ Constraints
✅ Failed approaches (super important, as it helps AI avoid trying things that didn’t work in the past)
✅ Important historical context

Think of MEMORY[.]md as persistent project memory for Claude Code and review & optimise it from time to time. The key is to keep it high-signal: store only long-term knowledge, explain the reasoning behind decisions, and remove outdated noise regularly.

Well-optimized MEMORY[.]md helps Claude avoid repeating mistakes and makes every new session feel less like starting from scratch.

👇

[Nick Babich @101babich]

📕 Complete guide for creating optimal Claude Code setup for product designers

https://t.co/PT9Njfdv70

[Nick Babich @101babich]

💡 Ultimate Claude Code Setup for Product Designers

The best setup isn’t about prompts; it’s about creating a design-aware workspace where Claude understands your product, UX standards, design system, and workflow.

My 5-step setup:

1️⃣ Create a proper project structure. Give Claude access to docs, design files, tokens, components, and product context.
2️⃣ Write a strong CLAUDE[.]md. Define the role, product context, design principles, system rules, workflow, and expected output.
3️⃣ Create Skills. Turn repeatable design tasks into reusable workflows.
4️⃣ Create specialized subagents for automating tasks. UX Reviewer, Design System Guardian, Accessibility Reviewer, QA Tester, etc.
5️⃣ Connect only the MCP tools you need. Too much context creates noise. Less context noise = better output quality.

👇

[Nick Babich @101babich]

📕 Practical examples for 7 tips for Claude[.]md optimisation:
https://t.co/AcOVefcjWg

[Nick Babich @101babich]

💡 7 Advanced CLAUDE[.]md Tips for Claude Code

CLAUDE[.]md is not just documentation; it’s the behavioral operating system for Claude Code.

A good CLAUDE[.]md helps Claude understand:
✅ what the project is
✅ what rules it must follow
✅ what it should avoid
✅ what success looks like

7 tips I recommend:
1️⃣ Keep CLAUDE[.]md under 200 lines
2️⃣ Optimize the first 30 lines (put the most important info upfront)
3️⃣ Separate hard rules from preferences
4️⃣ Add anti-patterns
5️⃣ Define what success looks like
6️⃣ Use imports for deeper context
7️⃣ Use scoped CLAUDE[.]md recursively

👇