D3 Weekly — Jul 26 to Aug 02, 2026

This Week’s Focus

The most striking through-line this week wasn’t a single tool or launch. It was the growing recognition that our information architecture — the invisible skeleton of every product — is the real bottleneck for AI. As one piece argued, every hallucination, wrong answer, and poor retrieval trace back to IA projects that never got funded. We’ve spent years optimizing surfaces while starving the structures underneath.

The second theme is fatigue. Not just AI fatigue, but the specific exhaustion of being the human in the loop. When the code starts writing itself, the reward function shifts from building to supervising, and that’s a different kind of cognitive load. The pieces this week weren’t asking whether AI can do the work. They were asking what happens to the people who used to do it.

Design

Information architecture is the foundation AI is starving for

Sidebar.io

Inspiration

Every AI problem you chase — hallucinations, wrong answers, poor retrieval — traces back to the IA projects you never funded. A sharp argument that the boring work of structure is now the highest-leverage design work available.

Components are dead

Heydesigner

Inspiration

Design is how it tastes, 9 designers open up about AI-assisted coding, and the design process meets the agent process. A weekly roundup that captures the industry’s slow realization that component libraries were a means, not an end.

Rauno Freiberg’s field note #2

Sidebar.io

Inspiration

How grid density, context menus, and custom widgets alter information architecture. A masterclass in the micro-decisions that define how users actually navigate complex surfaces.

AI and Technology

The human-in-the-loop is tired

Sidebar.io

Inspiration

On reward functions, dopamine, and what it actually feels like when the code starts writing itself. The most honest piece this week about the psychological cost of supervising machines that work faster than you can think.

Evaluating agents beyond the first prompt

Sidebar.io

Inspiration

Single-turn scores overstate reliability. Regressions, not missing features, are the real bottleneck. Essential reading for anyone shipping agentic features — the first prompt is the demo, the tenth is the product.

A practical guide to reducing token spend

Sidebar.io

Inspiration

Building a Swamp Workflow reduced real world token usage of an AI Agent by 8x and improved execution time by 2x. Concrete, actionable, and the kind of cost discipline most teams will need as agent usage scales.

Drawing the Mona Lisa with GPT-5.6, Claude, Gemini, and Grok

Sidebar.io

Inspiration

Four frontier models, a blank canvas, and colored pencils. Every stroke, dollar, and output tracked as they tried to draw the Mona Lisa. A fascinating benchmark that reveals more about each model’s process than its output.

UX Strategy and Process

This CPO regrets that product management exists | Tom Verrilli (CPO of Whatnot)

Lenny’s Newsletter

Inspiration

Tom Verrilli explains why fewer, more senior PMs doing real IC work outperform any ratio-driven org structure. A provocative take that resonates with the design world’s own reckoning with process overhead.

Your people get AI. Get out of their way.

UX Collective

Inspiration

The bottleneck isn’t adoption, it’s permission. A direct argument that the people closest to the work already know how to use AI — leadership just needs to stop being the constraint.

Jakob’s law: How to apply it as AI collapses surfaces into one chat box

UX Collective

Inspiration

Users spend most of their time in other products — and now that other product is a chat interface. A practical look at how Jakob’s Law applies when every surface starts to look the same.

Battling AI fatigue as a designer and developer: a practical guide

UX Collective

Inspiration

The exhaustion is real, and it’s not just about tool overload. A grounded guide to managing the cognitive load of constant AI integration without burning out.

Looking Ahead

The through-line for next week is clear: the conversation is moving from capability to accountability. The pieces that landed this week weren’t asking what AI can do — they were asking what it costs, who bears that cost, and what structures we need to build so the humans in the loop don’t become the loop. Watch for more work on IA as the foundation for AI, and for the first serious attempts to measure the cognitive toll of supervision work.