I have been looking for work since May. LinkedIn is my primary tool for that search, and sometime in late July I opened it to do something I have done a hundred times: type a job title, apply a few filters, scroll through results. Instead, I got a prompt box. No filters, dropdown for experience level, remote toggle. Just a text field asking me to describe what I was looking for. Well, I typed "Marketing Manager" and got back a mix of roles that had nothing to do with what I asked for.

I was not alone. LinkedIn's comment threads filled with people having the same moment. On August 14, the company quietly made it easier to switch back to classic search, after enough complaints, including from me, made the backlash impossible to ignore. It was not the rollout of a carefully planned feature,but a company reacting in real time to a problem it had not anticipated, or had anticipated and discounted. That reaction tells you more than the feature itself.

The distinction companies do not announce

When a company integrates AI into an existing product, it is making one of two choices, and it almost never says which one out loud.

The first is enhancement: AI is added on top of the experience users already have. A new tab, an optional panel, a "try AI" button. The original experience stays where it was. Users who want the new thing can find it; users who do not can continue as before.

The second is replacement: AI becomes the new default. The old version may technically still exist, usually as a toggle buried somewhere in settings, but the experience most users have has fundamentally changed. Getting back to what you knew is now your (big) problem.

LinkedIn chose replacement and described it as enhancement. The platform insisted classic search was still available. What it did not say was that most users would land on the AI version by default, that the filtering logic they had built habits around no longer worked the same way, and that finding the toggle to switch back required knowing it existed.

The frustration was not that AI search was bad, though for many users it was. A tool people relied on changed underneath them without warning, at a moment when the stakes for using it correctly were real. Looking for work is not a casual activity. The search precision that keyword filtering provided was not a preference, but a functional requirement.

Twitch said the quiet part out loud

The LinkedIn story would be easy to frame as a UX mistake, a rollout that moved too fast without enough user research. That framing lets the company off lightly. A harder, more honest reading is that this was the expected outcome of a deliberate choice.

On August 12, 2026, Twitch enabled a setting called "Training for Generative AI" for every account on its platform, without advance notice, without an email, without a pop-up. The setting allowed Amazon to use streamers' live broadcasts, archived videos, clips, chat messages, and channel images to train generative AI models. It was on by default. Streamers who objected had to find the option in Security and Privacy settings and disable it manually.

When Twitch's chief product officer Mike Minton was asked during a livestream why the setting was not opt-in instead, he answered plainly: "If it was opt-in, nobody would opt in. That's honestly the answer" (Forbes / 404 Media, 2026).

That sentence is worth sitting with. It isn’t a mistake or a poorly chosen phrase, but it’s a company executive describing, with unusual candor, the logic behind a design decision that affects millions of people. The feature was made opt-out because opt-in would have failed. The desired adoption numbers would not have materialized. The product investment would not have shown returns. So the default was set to on, and users who objected were invited to find the switch.

The circularity is precise. Twitch knew its community did not want this. It provided a mechanism to register that objection. And it designed the mechanism so that most of the community would never use it. The platform is aware its community objects, provides a way to register that objection, and simultaneously designs the mechanism so that most of the community will never use it.

LinkedIn's logic is structurally identical, even if less explicitly stated. An AI job search that users could opt into would not have generated the usage data needed to justify the investment. A default that replaced familiar behavior guaranteed engagement, even if that engagement was primarily people trying to find their way back to what they had before.

The demand question nobody is asking

What connects these cases is not bad execution. It’s a prior question that is rarely asked before the build begins: does the user actually need this?

The academic literature on technology adoption has a framework for this. Rogers' diffusion of innovations theory, developed across five decades of research, identifies perceived usefulness and relative advantage as the primary predictors of voluntary adoption. People adopt new technologies when those technologies solve a problem they have, faster or better than what they had before. (Rogers, 2003)

The inverse of this is the organizational trap that LinkedIn and Twitch fell into: building a feature because the technology exists, because competitors are moving, because the investment has already been made, and then manufacturing the adoption data that would normally precede the build. Default-on settings and forced replacements are how you generate usage numbers for a product that would not have been chosen freely. They are also how you discover, after the fact, that the problem you solved was not the problem your users had.

AI-powered natural language job search solves a problem that most experienced LinkedIn users do not have. Experienced users know how to use filters. They have learned that "marketing manager remote" plus a location toggle produces reliable results. The thing they would benefit from isn’t a more conversational interface. It is better data on which job listings are genuine, which companies are actively hiring, and which applications are actually reaching human eyes. None of that was part of what LinkedIn shipped.

What immaturity actually looks like

There is a version of this story that flatters the companies involved: they moved quickly, they experimented boldly, they iterated in response to feedback. That version is not entirely wrong, but it omits the part that matters most. The people who were most affected by these decisions were not consulted before they were made. Their existing behavior was treated as a problem to be solved rather than a signal to be understood.

Users who relied on LinkedIn's job search filters were not using a legacy system that needed modernizing. They were using a tool that worked. The replacement was not an upgrade, but an experiment conducted on people who had not agreed to be subjects.

A mature approach to AI integration starts from a different place. It asks what problem the user is trying to solve, whether the existing tool is already solving it adequately, and what specifically AI can do that the current solution cannot. It tests the new thing alongside the old thing, not instead of it, and it measures adoption under conditions where users have a genuine choice. It treats low voluntary adoption as information about the product, not as an obstacle to be designed around with defaults.

The LinkedIn rollback and the Twitch backlash are both forms of market feedback. The feedback is not that AI is unwelcome, but is that features designed to guarantee adoption rather than earn it tend to produce exactly the reaction that justifies opt-out being the only option offered.

‍

Photo by Alex Green: https://www.pexels.com/photo/upset-young-black-guy-covering-face-with-hand-while-working-remotely-on-netbook-5699826/

‍

References

• Rogers, E. M. (2003). Diffusion of Innovations (5th ed.). Free Press. Link
• Merkuri, K. (2026). Why LinkedIn's New AI Job Search Has Users Fuming. The Helpful Tipper. Link
• Tassi, P. (2026). Twitch Admits "Nobody Would Opt-In" To AI Training. Here's How To Opt Out. Forbes. Link
• gHacks. (2026). Twitch Adds Opt-Out for Amazon AI Training After Enrolling All Creators by Default. Link
• TechTimes. (2026). Twitch Streams Feed Amazon AI by Default. Link

‍