I’ve bought more vinyl records, physical books, and 4K Blu-rays in the past year than in the previous decade combined. Not because the content is different. Because I spend time with it differently. A vinyl record takes twenty minutes a side. A book requires hours. A disc means committing to a movie from opening titles to end credits. I’m less inclined to skip between things when I use these formats.
I still want software that works efficiently. Bad friction wastes time. Useful friction protects judgment. The problem is that software usually cannot tell the difference.
For years, the entire direction of software has been to remove the space between wanting something and having it. One-click buying. Instant streaming. Same-day delivery. Autofill, auto-play, auto-complete, auto-generate, auto-summarize. Fewer steps. Fewer questions. Fewer places where a human has to stop and decide.
Much of that was good. Broken forms, pointless approvals, and duplicated data entry should be fixed or removed. Bad search, missing files, platform lock-in, unclear ownership, and bureaucracy that protects nobody should all be treated as bugs.
But when friction disappears everywhere, other things disappear with it: review, deliberation, ownership, memory, source checking, taste, consent, privacy, accountability, and the moment where someone in the room asks, “Should we?”
As routine work becomes faster, deliberate review may account for more of the time and cost of doing it well.
Bad friction and useful friction
Bad friction is re-entering the same production data into five systems.
Useful friction is having one person verify the call sheet before it goes out.
Bad friction is waiting three days because nobody knows who approves the deck.
Useful friction is making the approval owner explicit before the client sees it.
Bad friction is manually hunting through old folders because search is broken.
Useful friction is keeping source files and version history so the team can prove what changed.
Bad friction is adding legal review to everything by default because nobody wants to think.
Useful friction is adding legal review when rights, likeness, confidential material, or client exposure are involved.
A delay is useful only if it serves a purpose. Waiting for an unidentified approver or repeating work lost by a broken form does not improve the result.
Useful friction does something. It protects quality, safety, rights, privacy, accountability, or memory. It creates a deliberate pause before risk. It gives the work an owner. It leaves evidence.
The bad kind should be automated out of existence. The useful kind should be designed on purpose.
AI makes the distinction unavoidable
AI removes friction from generating, summarizing, searching, drafting, editing, coding, and publishing. That is useful. I use it. I want the machine to handle routine work where it can: first drafts, cleanup passes, search, comparison, transcription, formatting, rough code, alternate phrasing, early concept work, and other small tasks that take up much of a day.
The mistake is treating every removed step as progress.
AI makes it easier to publish without understanding, summarize without reading, generate without sourcing, imitate without permission, decide without context, accept confident wrong answers, and skip the human who owns the consequence. A fluent answer can obscure missing reasoning. It can make a deck look client-ready before anyone has checked whether the plan is feasible. It can make a legal-looking paragraph before anyone has checked whether the claim is true. It can make an image look usable before anyone has asked who owns the style, likeness, reference, or source.
Review the output according to how it will be used.
If AI drafts the brief, a human still owns the promise. If AI summarizes the research, a human still checks the source. If AI generates the image, a human still clears the rights. If AI writes the code, a human still tests the behavior. If AI produces the answer, a human still has to decide whether the answer belongs anywhere near the real world.
The value is not in preserving every manual step. The value is in knowing which steps require a person to evaluate the result.
Platforms remove friction too
Platforms remove friction from distribution. That is also useful. A producer can find a director on Instagram. A VFX supervisor can discover an artist on Vimeo. A writer can test an idea on LinkedIn. A small tool can find its first users because a post traveled farther than the homepage ever would have.
That makes platforms useful for finding an audience.
That convenience can also make it easy to become dependent on them. They make it easy to publish instantly somewhere you do not control, build an audience you cannot move, accept default formats, let the feed decide what gets remembered, and trade ownership for convenience.
I wrote the longer version of this in Build on Your Own Land: use platforms for reach, but keep the canonical work, archive, URLs, and audience somewhere you control. That is another version of useful friction. Owning the domain, keeping the markdown file, backing up the media, saving the source, maintaining redirects, and preserving an exportable list all add steps.
They are not glamorous steps. They are not the fun part. They are the reason the work survives when a platform changes rules, search gets worse, an API shuts down, or a company decides your archive is not worth maintaining.
Those extra steps preserve your ability to move the work elsewhere.
Production already knows this
Production people understand useful friction because the job is full of real-world consequences. A shoot day is expensive. A rights problem can prevent a campaign from running. A skipped handoff conversation can cause a week of corrections in post. A beautiful mockup can become a budget problem if nobody asked how it gets made.
Generating a deck does not complete pre-production. The team still needs to agree on what the proposed work requires, clear the references it uses, and identify the approved version.
Production requires reviews of feasibility, safety, budget, and rights, along with records of what was approved.
These steps can be annoying and still be necessary. A call sheet records information the crew needs to coordinate safely. Version history shows what changed. A documented client approval confirms which version the client accepted before the team proceeds.
Resolving a disagreement, revising an inadequate version, and questioning a default all take time. Removing those steps may speed up delivery at the expense of the work.
Cameras like the ARRI Alexa became production standards partly because they make the right decisions easier and the wrong ones harder. They do not win by chasing every possible feature. They win by supporting intentional decisions under real production pressure. That reliability is part of their value.
Why I choose slower formats
When a quick version is easy to produce, people may value evidence that someone took time to consider the details.
I’m not the only one noticing. Bookstores have value even while Amazon dominates retail. Film photography keeps finding people while phone cameras have never been better. Theaters charge premium prices for movies that will eventually stream at home. People pay more to do certain things the hard way. The additional effort can be part of what they enjoy.
A handwritten note in a world of AI-generated messages. A meal cooked from scratch when delivery is fifteen minutes away. A photograph composed on film when you could spray a hundred digital frames. A website you built yourself when a template would have been faster. These choices cost something, and the cost is what makes them register.
The process alone does not establish quality. A website built with AI can still reflect careful decisions. A film photo can still be lazy. A handwritten note can still be manipulative. A long process can still produce mediocre work. Time and effort do not guarantee a good result.
It is only valuable when it makes the work better, clearer, safer, more accountable, or more yours.
The friction test
Before removing a step, ask:
- Does this step protect quality, safety, rights, privacy, or accountability?
- Does it make someone think before creating risk?
- Does it preserve memory or evidence?
- Does it clarify who owns the decision?
- Does removing it save time without creating hidden cost?
- Is this friction serving the work, or only protecting a habit?
- What breaks if we remove it?
- Who pays when it breaks?
If the step protects nobody, remove it. If it only exists because a tool is bad, fix the tool. If it hides ownership, redesign the process. If it turns one person’s fear into everyone else’s delay, name the decision owner and move on.
But if the step is where the source gets checked, the client approves, the rights get cleared, the version is saved, the safety issue is caught, the editor pushes back, the producer asks what this costs, or someone finally says “wait, should we?” then be careful.
Removing that step may remove a necessary check.
Keep the steps that improve the result
I expect careful review to become more valuable as generating a plausible first version gets cheaper. That does not justify every delay. Keep a step when you can explain what it protects or improves, and remove it when you cannot.
The useful question is whether the faster process still produces work someone has checked and is willing to take responsibility for.