What blurry text is really saying
When a company with hundreds of millions of users can’t be bothered to render text sharply on a decent monitor, the blur stops being a technical detail and becomes a message. Here is the complaint that made me notice it, posted on Weibo, which has lived in my head ever since:
“Why hasn’t a big company like Baidu, with a product like Baidu Cloud that has so many users, bothered to adapt it for high-resolution screens? It’s like they’re force-feeding users crap; every time I open it, it’s disgusting.”
Disgusting is a much stronger word than the complaint seems to earn. High-DPI support is not a moral failing, and blurry text on a nice monitor is an annoyance, not a crime. But the person writing isn’t reacting to the pixels. They’re reacting to what the pixels tell them: that a company that big weighed the cost of fixing this against them and decided they weren’t worth the afternoon. The product is telling you precisely how much you matter, and you can read it off the screen.
The reflexive explanation is that Baidu is lazy or incompetent. It isn’t. Baidu can ship whatever it decides to ship. The harder version is that not fixing it was the correct decision, and it was correct for reasons that have nothing to do with whether the fix is hard.
Polish is negative ROI when the door is locked
Every product manager knows the thing no earnings call will say out loud: user experience only pays when the user can leave.
Economists have a clean distinction for this. A contestable market is one where a competitor can show up, take your customers, and punish you for treating them badly. A captured market is one where they can’t. In the first, polish is a defensive weapon: every rough edge is a reason to defect, so you sand it down. In the second, that same polish is pure cost with no matching line on the revenue side. You spend three engineer-months making the thing nicer and exactly zero new users arrive, because there was nowhere for them to arrive from.
Baidu Cloud is captured. So is most domestic Chinese software, and increasingly a lot of the American stuff too, not because the code is a moat but because the data is. My files are already in there: contacts, purchase history, a decade of saved state I’ll never sort through. The technical name is switching costs, and switching costs are what prop up every business that treats you badly. The higher the cost of leaving, the worse the service can get before leaving is worth the trouble. Lock-in isn’t a side effect of the walled garden; it’s the point of it, and the deliberately mediocre rendering is priced against your unwillingness to climb out. Scale and captivity tend to arrive together, which is why “big company with millions of users” predicts almost nothing about quality. You don’t tidy the house for guests who can’t leave.
What the PM knows and what the quarter rewards
There is a gap I keep coming back to, a version of the old principal-agent problem. Somewhere inside Baidu is a product manager who knows the high-DPI experience is embarrassing. She works on a good monitor. She has seen the blur, probably filed the ticket twice. She is not the villain; she’s the person best placed to fix the thing and least able to justify doing it.
Because what she can see and what the company pays her to move are two different quantities, and the distance between them is what leaves the blur on the screen. At review time she is measured on numbers that move: monthly actives, conversion, retention, revenue per user. A high-DPI fix moves none of them in a way any dashboard can detect, because nobody churns over blurry text when the alternative is re-uploading their entire life to a competitor. So the fix carries real cost and no legible benefit, and inside a quarterly cycle “no legible benefit” is functionally identical to “no benefit.” The organization isn’t ignoring quality so much as failing to perceive it. The instruments only read revenue, and quality that doesn’t convert doesn’t register.
Multiply that one PM across every team on every captured product and you get the actual texture of modern software: rarely hostile, just systematically indifferent to anything that won’t show up on a chart.
DJI, and two years of nothing
The cleanest example I’ve seen involves the DJI Pocket. There was a forum post about getting footage off the thing, and anyone who has tried it recognized the pain on sight: the transfer workflow is bad in a way that feels almost intentional, as if the friction were deliberate. The post went off. The people who piled into the replies weren’t casuals whining about a minor inconvenience. They were vloggers and NAS hoarders, people who move terabytes on purpose, and they wrote up detailed, genuinely useful proposals for how to fix it. Free consulting from exactly the users most companies would kill to have.
DJI’s response, across roughly two years, was nothing. No fix, no acknowledgment, not even the form-letter “thanks for the feedback” that costs a company a copy-paste and nothing else.
Here is the part that makes it instructive rather than merely annoying: DJI was probably right. The power users screaming about batch transfer are a rounding error. The median Pocket buyer shoots a thirty-second clip, airdrops it to a phone, and posts it before dinner. The Pocket competes on being small, cheap, and good enough for that person, and every hour spent serving the NAS crowd is an hour not spent on the thing that actually sells units. The engineers didn’t fail to hear the power users. They heard them perfectly, priced them, and correctly concluded they didn’t matter. That’s the unsettling part: it wasn’t an oversight, it was arithmetic.
You are a revenue unit, and I mean that technically
“We’re not customers, we’re revenue units in a spreadsheet” is the kind of line that belongs on a mug, so let me try to turn it into a real claim.
A public company does not experience you as a person; it has no organ for it. What it has is a financial model, and in that model you are a small bundle of expected cash flows with a churn probability stapled to the side. That’s not cynicism, just accounting. Everything it knows about you arrives pre-aggregated: cohorts, funnels, lifetime value, retention curves. Your specific afternoon squinting at blurry text does not exist at that altitude. It can’t be seen, so it can’t be weighed.
Which means that when we say a company “doesn’t care,” we’re being imprecise in a way that lets it off the hook. Caring was never on the menu. A corporation is a machine for turning inputs into shareholder value, and it will improve your experience exactly to the degree that your experience shows up as a term in that conversion, and not one inch past it. When you can leave, your annoyance is such a term. When you can’t, the machine never records it, and what it doesn’t record isn’t real. The contempt isn’t personal; it’s structural, which is worse. Being contemptuous takes a person, and there’s no person in here to appeal to or to shame.
The part that’s on us
Which brings me to the part I like least, because it implicates the people doing the complaining, and that’s all of us.
The walls are only as high as we keep them. Switching costs feel like laws of physics, but most of them are habits wearing a costume. The files could be exported; the competitor could be tried for a weekend. Every captured market is, on some slower timescale, still contestable, and the thing holding it captive is that the people inside keep deciding the exit costs more than the insult. In the short term they’re usually right, which is how you end up wrong across years. We rant and screenshot the blur, then open the app again tomorrow because our stuff is in there, and the company reads that correctly: the complaint is noise, the reopening is signal. We are subsidizing our own bad treatment, paying in attention and data and inertia, and the return is a product that got exactly as good as it had to be to keep us.
The Baidu rant was never really about high-DPI support, and I think the person writing it half-knew that. It’s about noticing the arithmetic that has your name in it and realizing you came out on the wrong side of the sum. The blur is honest. It’s the company telling you the truth about the deal you’re in. The only open question the rant leaves behind is whether we’re ever angry enough to make leaving cheaper than staying, and the evidence, so far, is that we are not.