Who Owns Your Attention? TikTok, Recommendation Algorithms and Human Agency


Have you ever opened a video app for one specific reason and looked up much later, unable to explain where the time went? A recommendation feed makes it effortless to move from one interesting clip to another. The deeper question is what happens to our intentions when an interface continually makes the next choice easier than leaving.

Man in glasses and a checked blazer sits holding a phone beside a plain wall.

In 2020 I translated Jia Tolentino’s New Yorker essay on TikTok and attention into Russian for VerticalVideoPortal. It was a long, memorable piece, rich in scenes and reporting. Rather than reproduce it, I want to revisit one of the underlying questions from my perspective as an engineer and creator: how do we benefit from personalized discovery without surrendering the direction of our time?

Frictionless is not the same as neutral

The short-video interface is remarkably low-friction. Full-screen content starts without searching; a swipe moves to another selection; feedback arrives almost immediately. These design decisions help people discover unfamiliar art, jokes, languages and knowledge. They also make stopping an active choice.

Recommendation algorithms can learn from patterns of behavior, but behavior is imperfect evidence of desire. Watching a disturbing clip does not necessarily mean you wanted to see more of the same. Replaying an explanation may mean confusion rather than fascination. A system can accurately predict a next action while misunderstanding what a user considers worthwhile.

The metrics do not know your purpose

Imagine a student opening a feed to find one explanation of an electric field. The app may show a relevant demonstration followed by unrelated entertainment. If the student stays, retention may rise. But did the student achieve the learning goal? From the user’s perspective, there is an important difference.

I work with AI and education, and I believe this distinction should shape our metrics. Watching time is one signal. Understanding, remembering, practicing and applying a concept are different outcomes. A product that measures only attention can optimize the wrong thing very effectively.

The creator faces the same tension

Creators are constantly encouraged to increase retention, produce faster hooks and avoid any moment that might make a viewer leave. Some of that is good storytelling practice: make the subject clear, remove unnecessary filler and respect the audience’s time.

But when every creative decision is governed by a prediction of the next swipe, a video can lose its point of view. Complexity disappears. Repetition wins. A creator may gradually learn to perform whatever works instead of developing what they actually have to say.

I think of this as a question of authorship. You can use data to make your communication clearer without allowing data to dictate the beliefs, relationships or stories you are willing to put into the world.

A practical attention contract

  1. State the intention. Before opening an app, know whether you want to learn, relax, research or connect.
  2. Use explicit feedback. Follow useful creators, indicate disinterest where supported, and curate what you can.
  3. Separate discovery from deep work. Save a useful idea and return to it in a calmer format rather than endlessly browsing adjacent clips.
  4. Notice the exit. If the next video no longer serves the original intention, leaving is a successful decision.
  5. Evaluate the result. Ask what you gained, made or understood—not only how much time passed.

What this means for AI agents

Recommendation systems are beginning to connect with assistants that can search, schedule and take actions. That raises the stakes. A feed choosing the next video can cost attention; an agent choosing a purchase or sending a message can create financial or social consequences.

Human agency therefore needs to be designed into the product. We should be able to inspect reasons, grant limited permission, approve consequential actions and change our minds. Automation should not make the user’s purpose harder to see.

I do not oppose recommendations. I love how a thoughtful system can introduce us to a place, an artist or a scientific idea we would never have found on our own. But discovery should increase the range of choices available to us, rather than quietly decide what we are for.

Why I care as a creator

My English project is called AI • People • Culture. It includes technology, education and storytelling because these subjects belong together. A useful algorithm is an engineering achievement; the lives and purposes of people affected by it are equally part of the story.

The question I now ask of any engagement-optimized product is simple: when the session ends, does the person feel more capable of living deliberately—or merely more likely to come back?

Reading