When brand identity and algorithms pull in opposite directions

Sometime around 2014, the phrase "Pinterest perfect" stopped being a compliment. It had drifted into something more like an accusation — shorthand for a domestic idealism so polished it made ordinary life feel inadequate. The irony was genuine: a platform built around the pleasure of collecting things you liked had accumulated a cultural reputation for making people feel bad about what they had. The brand had become its own distortion.

The design response James Hurst developed with Made Thought moved directly at that distortion. Rather than smoothing Pinterest's visual identity into something more aspirational — more of what the platform had already been blamed for — the work moved the other way. A collage aesthetic, deliberately rough at the edges. Images overlapping without resolution. The visual language of a mood board mid-thought, not a mood board finished and framed. The brand would signal: this is what inspiration actually looks like before it becomes a decision. Mess as corrective. Incompleteness as honesty.

It was a persuasive idea. And it raises a question that has only grown sharper since: what happens to a brand identity built around productive accident when the platform itself is getting better and better at removing accident from the equation?

Pinterest introduced Lens, its visual search tool, in 2017. The capability it pointed toward — machine learning systems that could look at what a user responded to and surface more of it, faster, with less friction — has only deepened in the years since. The logic is straightforward and defensible: people find things they love more quickly, the platform becomes more useful, engagement follows. Algorithmic personalisation is a genuine service.

But there is a structural tension inside that service. The collage aesthetic was built on a theory of inspiration: that the thing you didn't know you were looking for is often more valuable than the thing you came to find. Serendipity is not a flaw in the discovery experience — it is the experience. The rough edge of the mood board, the unexpected adjacency, the image that shouldn't work next to the other image but somehow does — these are not decorative features of creativity. They are, for many people, how creativity actually moves.

Recommendation engines are optimised against a different theory. They are built to reduce the gap between what you've expressed and what you're shown next. The better they get, the narrower that gap becomes. Which means the better Pinterest's algorithm gets at predicting what a user wants, the more the experience of using Pinterest starts to resemble a mirror rather than a window. You see yourself reflected back, refined and confirmed, rather than encountering something genuinely outside your existing frame.

Spotify ran into a version of this publicly. Its 2023 home screen redesign moved further toward algorithmically predicted listening and away from open browsing — and the discussion that followed was less about the music people were finding and more about the music they suspected they were no longer finding. The complaint wasn't that the recommendations were bad. It was that they were too coherent. That something had been lost in the tidying.

James' second book, Weird the Normal / Normal the Weird, frames a related problem at a structural level. The argument is that neither move — estranging the familiar, or stabilising the strange — works alone. Too much normalisation stagnates. Too much disruption burns out. The actual skill is the switching between them: keeping the pulse alive rather than landing permanently on one side. What the book describes as a discipline is, in practice, a resistance to resolution.

Algorithmic recommendation systems are, almost by definition, resolution engines. They are built to close the loop between preference and supply. Applied to a platform whose brand identity was designed to resist resolution — to stay in the productive middle of a mood board that hasn't decided what it is yet — the question becomes genuinely interesting. Can those two things coexist? Or does one eventually win?

There are ways to hold the tension. A recommendation engine could be tuned to introduce deliberate friction — to surface something adjacent rather than identical, to widen the loop rather than tighten it. Some platforms have experimented with serendipity as a feature rather than a failure mode, though whether that intervention can survive the pressure of engagement metrics is another question. The business case for surprise is harder to make in a quarterly report than the business case for time-on-platform.

The collage aesthetic said: we believe inspiration is messy. The algorithm says: we believe we know what you want. Both statements can be true. But they are not both true at the same time, in the same interface, for the same user.

What James built with Made Thought was a brand argument — a position the platform took about its own nature. Brand arguments are not self-executing. They require the product to keep making the same argument, at the level of the experience, every time someone uses the thing. If the experience gradually stops making that argument — if the mess resolves into prediction, the collage into a feed that knows you too well — then the visual identity becomes something closer to a promise the platform is quietly walking back.

Which is worth sitting with, not as a problem to solve, but as a question worth staying curious about: when the brand and the algorithm are pulling in different directions, which one is telling the truth about what the platform actually believes inspiration is?