I used to push pixels. Now I push back.

Twenty years ago I was pushing pixels for whoever paid the invoice. Today I get paid to have opinions, which my younger self would have found both suspicious and deeply unfair.

This is not a lessons-from-two-decades listicle. There will be no "ten things I wish I knew". It is about one thing that changed, and a lot of things everyone assumes changed that did not.

The short version of the long version

I have worked inside agencies (Cygnis Media, Big Drop Inc, XS Studios), inside product companies (Day One Technologies, Envase Technologies, WiseTech Global, Teamfront), and independently, taking founders from a blank canvas to a shipped product.

Along the way: freight dispatch and transport management, field service software, fintech, home care, a platform for trading card game players, and lately agentic AI systems. On paper that reads like someone who could not decide what they wanted to do. In practice it was the same job wearing different uniforms.

What did not change the job

The tools changed constantly. Sketch became Figma. Static comps became interactive prototypes. Design systems went from a PDF someone printed once to a living library the developers build against. And now AI drafts half my grunt work before I have opened the file. This year I shipped whole applications through a workflow where design and frontend are one activity, with no handoff at all.

None of that touched the actual job. It changed the speed of the job, which is a different thing, and I will come back to why speed matters more than it sounds.

What did change the job: the brief

Early on, someone handed me a brief and I made it look good. That was the whole contract. The thinking had been done upstream by people who were paid to think, and my role was to make their thinking presentable.

Now I am usually the one deciding what the brief should have said in the first place. Four recent examples, none of which is a pixel decision:

On a home care platform, I mapped the care chain by role before designing a single screen: agency coordinators, clients, families and caregivers. What came back was one failure seen from four seats. The visit had no shared record. That turned three feature lists into one lifecycle, and every screen after it had a reason to exist.

On a landing page for an import logistics business, sixty-three questions went to the client across two rounds, covering what the company sells, who it is for and how to reach it. None came back answered. The design question stopped being "how do we present this business" and became "what is a page allowed to say about a business that has told nobody anything".

On a platform for card game players, the decision that shaped everything was refusing to build a feed. The live game session became the centre, because hosting the game was the one thing the incumbents structurally could not do.

Across six field service products, the call was to rebuild all six against one design system rather than restyle each on its own terms. Slower to start, and the only version where the debt stops regrowing.

Every one of those decided what the pixels would be for, before a pixel existed.

The pixels still matter

I want to be careful here, because "I am strategic now" is the most tiresome sentence in design. I have not graduated from the craft and I do not trust designers who say they have. On that logistics page, 142 contrast ratios were computed and published, not eyeballed. Opinions without craft are just meetings.

What changed is the order. The opinion comes first, then the craft, and the craft is how you find out whether the opinion was any good.

The part nobody warns you about

When the brief was someone else's, a wrong call was theirs. Now it is mine.

That is the real price of being trusted to decide, and it is not a comfortable one. On the home care platform I argued that compliance data is only as good as the conditions it is captured in, then designed the whole verification flow around the tired end of a caregiver's shift. I never observed that moment myself. The decision still stands up, and it is still the gap I would close first. Being trusted to decide means owning what you did not get to check.

The only protection I have found is having done this long enough to know which kinds of call I tend to get wrong, and building the checks in before I need them.

Why speed makes this matter more

AI made production cheap. A screen that took days takes an hour. That sounds like it should make judgement less important, since you can just try everything.

It does the opposite. When building a screen is the cheap part, the judgement that matters moves entirely to deciding what not to build, and the tooling gives you no help with that at all. A generated screen is always prettier than a blank one. Knowing when blank is the honest answer is the job now.

The throughline

Twenty years across freight dispatch, field service, fintech and now agentic AI systems, and the throughline was never the industry. It was earning the right to be listened to before the pixels show up.

The pixels still have to be good. They just stopped being the point.