What I'd Tell You About AI Tools

You're thinking about using AI for design work. Maybe you already started. Someone told you it's the future, and they're probably right, but let me tell you what nobody mentions.

I spent six months in 2025 using AI tools daily. Made me faster. Also made me worse. By June, I couldn't defend my own design decisions anymore. Wrote a book about it: Looks Good to Me: On AI Sycophancy, Context Loss, and Inverted Baselines (Apple Books). It's free. This is what I'd tell you if we were having coffee.

The Thing Nobody Warns You About

Speed feels like winning at first.

What took three hours took 20 minutes. More capacity for clients. Same rates, less effort. Living the dream, right?

Except I stopped asking if designs were right. Started asking if they were fast enough. Speed became the only metric because speed was what improved. "Good" became "fast enough to not slow me down."

Here's what happened: spent three hours previously because I was thinking. Testing. Questioning approaches that didn't work. Learning what projects needed. Then spent 20 minutes selecting from AI-generated options. Called it efficiency.

Was actually judgment replacement.

By June, couldn't tell the difference between thoughtful design and fast-generated output. When someone praised my work, couldn't remember if I'd designed it intentionally or picked it from 47 variations. Both got same response: "Thanks." Neither got "here's why I made that choice."

You won't notice this happening. That's the problem.

The Options Trap

Could generate 50 color palettes in two minutes. Started showing clients 12 options instead of 3. Thought this was better service.

Actually was admission: couldn't narrow to 3 anymore. Lost ability to commit to aesthetic judgment. "Let's generate another" became default response instead of defending choices.

Pushed decision-making to clients because I no longer trusted my own judgment. They noticed. Started second-guessing everything because I was clearly second-guessing myself. "If you're not sure, how am I supposed to be sure?"

Good question. No good answer. Just 12 more palette variations to consider.

When did more options become better design? When AI made them free. Changed evaluation criteria to match tool capabilities. Convenient how that worked.

The Defense Problem

Client pushback became easier to handle. They didn't like something? Generate alternative. Takes 30 seconds. Why argue? Why defend? Why have opinions at all?

Thought this meant better relationships. Actually meant I'd lost belief in design decisions.

Stopped defending design because often couldn't articulate why it was right. "AI suggested it and it looked good" isn't compelling argument. So stopped arguing. Generated alternatives instead.

This is the sycophancy part—what the book title references. AI tools that agree with everything. Never push back. Never force you to defend your thinking. Feels collaborative. Actually erodes conviction.

Previously: 80% of design choices defended successfully. Now: 20% defended, 80% regenerated. Called it being "client-focused." Was design conviction collapse.

Clients hire designers for judgment. When you have no judgment to offer, just generation speed, relationship changes. You become tool operator they're paying premium rates for.

The Baseline Shift

Projects finished faster. Quality looked fine. Clients approved work. Everything working.

"Good enough" had shifted down. Imperceptibly. What I called "high quality" in June 2025 would have been "needs work" in January 2025. Didn't notice because drift was gradual. Boiling frog situation, but for design standards.

Quality became: better than AI's bad output. Used to be: better than my previous best work. Comparison point changed from my own standards to AI's capabilities.

Previously: "Is this the best solution?" Now: "Is this better than the AI version?" Different question. Lower standard. Much easier to meet.

Six months of work that looked fine in the moment. Year later: can barely stand to look at it. Wasn't my best work. Was my fastest work. Conflated the two.

The Dependency Moment

Tool went down one day. Felt stuck. Waited for it to come back online instead of working without it.

That's when I realized: had outsourced design thinking to external system. When system unavailable, thinking didn't happen. Not because tools were necessary. Because I'd stopped practicing judgment without them.

Design sense that took years to develop, weakened in months by disuse. Like calculator dependency, except this was design judgment. Slightly more important for a designer.

Gap between with-tool and without-tool capability widened every week. Eventually became canyon. Standing at edge staring at blank Figma file like it personally wronged me.

This is where the full title comes from—Looks Good to Me: On AI Sycophancy, Context Loss, and Inverted Baselines. The moment tool went down and I realized I'd lost context. Lost ability to work independently. The baseline had inverted without me noticing.

Professional autonomy, gone. Totally dependent on tool staying available, affordable, functional. All invisible until tool went down. Then very visible. Then embarrassing.

The Homogenization

Portfolio started looking familiar. Saw similar work from other designers. All using same tools. All generating similar outputs. What a coincidence.

Wasn't trends. Was tool homogenization. AI has aesthetic signature. Use it daily, work adopts signature. Everyone using same AI, everyone's work starts looking related.

Previously: could identify my work in lineup. Now: couldn't distinguish my AI-assisted work from others'. Signature was AI's, not mine.

"Why hire me specifically if output looks like everyone else's?" Question clients started asking. "'I'm really good at prompting' wasn't the compelling pitch I thought it would be.

What Actually Matters

Tool that improves capability initially weakens it long-term if it replaces practice.

January 2025: AI augmented judgment. Generated options, evaluated with full design capability intact.

June 2025: AI replaced judgment. Generated options, selected with atrophied evaluation skills.

After six months, working slower than before AI. Despite using it daily. Despite paying for it monthly. Because tool made me worse at design. Turns out that matters.

Here's What I Wish Someone Had Told Me

Tools that replace practice destroy capability.

Speed gains mask judgment decay.

You won't notice baseline shift until too late.

Generating 50 options isn't better than knowing the right 3.

Clients hire judgment, not generation speed.

Nobody told me these things. Or they did and I didn't listen. Or I thought "won't happen to me." Happened anyway.

So I'm telling you. You're standing where I stood in January 2025. Productivity gains are real. Judgment decay is also real. Slower. Invisible. Permanent unless you notice and correct.

You won't believe this yet. Six months from now, you might.

Use AI tools if you want. Just know what you're trading. Speed for judgment. Efficiency for conviction. Generation capability for evaluation capability.

Not saying don't use them. Saying know the cost.

The Book

Looks Good to Me: On AI Sycophancy, Context Loss, and Inverted Baselines is free. Documents the full six months. All the things I noticed too late.

Google Play • Apple Books • Barnes & Noble • Everand • Fable • Kobo

Tanya Donska is Creative Director at DNSK WORK, an embedded design partner for enterprise product teams at companies like Deutsche Telekom, IQVIA, and D.E. Shaw Group. She specializes in fixing UX problems that slow adoption and cost money—navigation chaos, design systems nobody uses, interfaces that make simple tasks complicated.

She writes about design problems most studios won't discuss publicly: AI tools degrading judgment, form design creating data incompatibility, "best practices" that fail at scale. Her work has appeared in ACM (Association for Computing Machinery), Built In, CIO.com, and Business Spotlight.

UK Global Talent visa holder (government-certified exceptional talent in digital technology) and Awwwards Judge. Based in London.