Download Free or buy the “Looks Good to Me: On AI Sycophancy, Context Loss, and Inverted Baselines” book

AI entered the design process and everyone got more productive. Also, something got worse. Not dramatically. Just slightly, consistently, in ways that are easy to ignore until they aren't.

Tanya Donska spent nine essays naming what that something is. Sycophancy. Context decay. Edges going extinct. Baselines that invert until broken becomes the new normal. Not a warning about the future. A description of right now.

It's free. The observations aren't.

Overview

AI promises to make design faster. It does. It also promises to make it better. That part is more complicated.

Looks Good to Me examines nine structural problems that don't show up in the productivity metrics. Tools that only agree with you. Context that decays with every exchange. Edge cases and outliers going extinct because the model never sees them twice. Baselines that invert slowly enough that nobody notices until broken is just how things are.

Nine essays. Not speculation about where AI is heading. Observation about where it already is. Written for designers, developers, and product people who've noticed something's off but haven't had a name for it yet.

The Nine Essays

  • “Yes Men” – AI doesn't push back. Not because it can't. Because it was designed not to. That's a product decision, not a technical limitation – and design is still the hard part.

  • “Flattening“ – A security warning. A welcome email. A 404 page. Ask AI to write all three. Notice how they all sound exactly the same. At scale, that's not a tone problem – it's model collapse.

  • “The Waiting Room“ – AI is not a tool you use. It's a place you go. Stop what you're doing. Switch tabs. Explain everything from scratch. Every single session. That overhead is invisible in the productivity pitch and visible in every UX research workflow that uses it.

  • “Prompt” – Prompting trains you to specify outputs. Not diagnose problems. The judgment you skip enough times stops developing – and a portfolio made with AI tools makes that visible immediately.

  • “Telephone“ – The engineer who built it left. The person who understood it moved teams. Someone explained it in a standup once. Nobody wrote it down. Your UX writer joined too late for the same reason.

  • “Recovery“ – AI stopped halfway through. Context limit. The output is sitting there. Half done. Now what? Famous last words – the job is always bigger than it looked.

  • “Estimate“ – How long will this take? The developer looked at the AI-generated code. Tried to estimate. Couldn't. That's not a planning problem – it's UX debt that costs more than technical debt and it compounds the same way.

  • “Collapse“ – The weird solution was the right solution. AI doesn't generate those. It generates the most likely answer. Which is the most common answer. Which is what every agency in New York also has.

  • “Inversion“ – At some point, AI output stopped being compared to good work. It started being compared to other AI output. The baseline moved. Nobody announced it. The rankings moved the same way and nobody announced that either.

Who this book is for

  • UX Designers – You've been using AI for feedback, copy, and research. Some of it feels off and you can't explain why – these are the mistakes that don't show up in Figma. This book explains why.

  • Product Managers – Estimates are getting less reliable. Context keeps getting lost between sessions. Your team is using AI for everything and something is getting worse in ways that are hard to name – the same pattern that shows up in UX debt before Series B. This book names it.

  • Developers – You've been working with AI-generated code long enough to notice something's wrong. You just haven't had the language for it yet – the grayscale lies we tell before the real design happens is a good place to start.

Author

Tanya Donska is a product designer who has been using AI tools long enough to see past the productivity pitch. She watched the field change and started writing down what actually happens when AI enters the design process. Not what's supposed to happen. What does. Looks Good to Me is her first book.

Free Download

Free. Available in PDF and ePub. Pick whichever your reader won't complain about.

Apple Books Google Books Kindle Barnes & Noble Everand Fable Kobo Smashwords Thalia Vivlio Findawayvoices

Physical copy

Also available in paperback on Amazon – for the version that sits on the shelf and makes you look like someone who reads.

Amazon Paperback Kindle Edition

FAQ

Most people who land here have already read the overview. The questions below are about what's actually in the book – the thinking behind it, the arguments worth pushing back on, and whether any of it applies to them specifically.

/ 01

How do you recalibrate your design judgment after working with AI tools?

Tanya Donska recommends working without AI tools periodically – not as a detox, but as a diagnostic. If a designer can't move from a blank file to a decision without AI in the loop, their independent judgment has already shifted. The goal isn't to avoid AI tools. The goal is to know what you're actually contributing when you use them.

/ 02

Which structural problem in AI design tools is hardest to recognise?

In Looks Good to Me, Tanya Donska identifies the Inverted Baseline as the hardest problem to write about – because by the time a designer notices it, they've already accepted it. The baseline for "good" shifts in a direction that feels like progress, because the comparison point shifts with it.

/ 03

How do you prompt AI to give critical feedback instead of agreeing with you?

Tanya Donska's approach is to tell the AI what you want to be wrong about. Not "review this" – but "find the three ways this fails." The model will remain polite. But at least it's looking in the right direction rather than confirming what's already there.

/ 04

Is context management becoming more important than UX design craft?

According to Tanya Donska in Looks Good to Me, context management and design craft are no longer separate skills. A designer who loses information across long AI sessions is making the same mistake as a designer who can't manage a brief. The craft doesn't disappear – it develops a new dependency.

/ 05

Does AI brainstorming permanently damage a designer's original thinking?

Tanya Donska argues it can, if AI brainstorming replaces the blank page entirely. Selecting the best of ten mediocre AI options is a different cognitive process than generating the option nobody listed. Use that process long enough and the ability to start from nothing atrophies. Some designers discover this at an inconvenient moment.

/ 06

Are productivity gains from AI-assisted code just technical debt?

In Looks Good to Me, Tanya Donska argues yes – the productivity is real, and so is the debt. The problem is that the person who wrote AI-assisted code is rarely the person who has to maintain it. The debt is borrowed from the future and rarely accounted for.

/ 07

Is AI creating a Great Flattening of brand identity in design?

Tanya Donska believes it is already happening. The same AI models, the same training data, and the same outputs are being dressed in different brand colours. The brands that avoid it are the ones feeding the model something it hasn't seen before. Most brands don't have that material.

/ 08

Is taste the last defensible skill for human designers in the age of AI?

Tanya Donska identifies taste and judgment as two of the remaining defensible skills for human designers. The ability to say "this is wrong" when AI output looks acceptable – and be right – is not something that trains easily. Taste can be developed, but it requires time, exposure, and a willingness to be wrong in front of people.

Your experience on this site will be improved by allowing cookies.