John Grubers essay “Your frustration is the product” hit Hacker News like a thunderbolt this week. The thesis is simple but devastating: Many AI tools arent designed to solve your problems efficiently. Theyre designed to keep you engaged, subscribed, and paying.

Its a critique that applies to more than just AI, of course. Social media has been optimizing for engagement over user wellbeing for years. But theres something particularly insidious about applying these same patterns to productivity tools—tools that promise to make us more efficient while secretly working against that goal.

The Engagement Trap

Consider how AI coding assistants work:

You write a function. The AI suggests a completion. You accept it. Then you notice its not quite right, so you ask the AI to fix it. It suggests another change. You try that. Still not right. Three iterations later, youve spent more time than if youd just written it yourself.

But heres the thing: the AI company just logged three interactions. Three opportunities to demonstrate value. Three data points for training. And if youre on a usage-based plan, three billable events.

The incentive structure is all wrong. The AI is rewarded for keeping you in the loop, not for getting you out of it.

The Subscription Squeeze

Monthly recurring revenue is the holy grail of SaaS business models. But it creates perverse incentives for AI tools:

If the AI solved your problem completely, you might cancel your subscription. So the AI is subtly optimized for:

• Partial solutions: Good enough to seem helpful, not good enough to finish the job
• Ongoing engagement: Features that require regular interaction rather than one-time setup
• Dependency creation: Making you feel like you couldnt function without the tool

This isnt necessarily conscious design. No product manager is twirling a mustache saying “lets make it worse so they keep paying.” But the business model shapes the product, whether anyone intends it or not.

Case Study: AI Writing Tools

Take the explosion of AI writing assistants. The promise is clear: write faster, write better, overcome writers block.

But look at how theyre actually used:

• You start with a prompt
• The AI generates text
• You edit it (because its not quite right)
• You ask for revisions
• You edit again
• Eventually you get something acceptable

Compare this to the old workflow:

• You write
• You edit
• Youre done

Is the AI-assisted workflow actually faster? For some people, in some situations, sure. But for many users, the AI becomes a crutch that slows them down while creating the illusion of productivity.

And the AI company? Theyve got a monthly subscriber who generates regular usage data. Win-win (for them).

The Metrics That Matter

When AI companies report success, what metrics do they highlight?

• Monthly active users: Not how much value users got, just that they logged in
• Interactions per session: More is better, apparently
• Time spent in product: Longer sessions = more engagement = better
• Feature adoption: Using more features = more stickiness

Notice whats missing:

• Time to task completion: How fast did users actually finish their work?
• Quality of output: Did the AI actually improve results?
• User satisfaction: Do people feel better about their work?
• Long-term retention of skills: Are users getting better or more dependent?

The metrics we optimize for shape the products we build. When engagement is the goal, engagement is what we get—even at the expense of efficiency.

The Attention Economy Comes for Productivity

Weve spent the last decade recognizing how social media hijacks our attention for engagement. Now the same playbook is being applied to productivity tools.

The techniques are familiar:

Variable rewards: Sometimes the AI gives you exactly what you need (dopamine hit!). Sometimes its useless. The unpredictability keeps you trying.

Notification overload: “Your document has suggestions!” “New AI features available!” “Your weekly writing report is ready!”

Artificial scarcity: “Youve used 45 of your 50 AI credits this month!” Creating anxiety about running out.

Gamification: Streaks, badges, usage charts—all designed to make you feel bad for not engaging.

These are engagement techniques borrowed from social media, applied to tools that are supposed to help us work.

The Alternative: Tools That Exit

What would AI tools look like if they were optimized for user success rather than engagement?

They would get out of the way: The best AI interaction is often the one you dont notice. Smart autocomplete that just works. Background processes that handle the tedious stuff without demanding attention.

They would teach, not just do: Instead of generating code for you, theyd explain why the code works. Build your skills rather than replacing them.

They would have clear exit points: “Youve completed this task. Want to close the AI assistant?” Rather than endless “helpful” suggestions.

They would charge for value, not access: Pay for outcomes, not for the privilege of chatting with an AI.

Some tools are moving in this direction. But theyre swimming against the current of the dominant business models.

The Responsibility Question

Whose fault is this? Its easy to blame AI companies, and they deserve some of it. But the reality is more complex:

Investors demand growth metrics that favor engagement over efficiency.

Users choose tools based on feature lists and demo videos, not long-term impact on their workflow.

Competition drives everyone toward the same engagement-optimized playbook.

Our own psychology makes engagement metrics effective. Were not rational optimizers; were creatures of habit and dopamine.

What Users Can Do

If youre tired of AI tools that optimize for their success over yours, here are some strategies:

1. Audit Your Tools

For each AI tool you use, ask:

• Does this actually save me time, or just feel productive?
• Am I getting better at my work, or more dependent on the tool?
• Would I pay for this if it were priced by outcome rather than subscription?

2. Set Boundaries

Use AI tools intentionally, not habitually:

• Turn off notifications
• Define specific tasks for AI assistance
• Set time limits for AI interactions
• Regularly try working without AI to maintain your skills

3. Vote with Your Wallet

Support AI tools that:

• Charge transparently for value delivered
• Respect your time and attention
• Help you build skills rather than replacing them
• Have clear, honest marketing about what they do

4. Build Your Own

For many tasks, local AI models are now good enough. Running your own:

• Eliminates subscription costs
• Removes engagement optimization
• Keeps your data private
• Forces you to understand what the AI is actually doing

What Builders Can Do

If youre building AI tools, consider the long game:

Optimize for user outcomes, not engagement: Measure whether users actually accomplish their goals faster and better.

Design for skill building: Help users become more capable, not more dependent.

Be honest about limitations: Dont oversell what AI can do. Set realistic expectations.

Explore alternative business models: Usage-based pricing, outcome-based pricing, or even open-source with services. The subscription model isnt the only option.

Build tools that exit: The best AI tool helps users solve their problem and then gets out of the way.

The Long View

The current wave of AI tools is still young. The business models will evolve. User expectations will shift. Competition will drive innovation.

But the fundamental tension will remain: AI companies need to make money, and users want to solve problems efficiently. Aligning those incentives is the challenge of this generation of tools.

Grubers essay is a wake-up call. Your frustration isnt a bug—its the product. And its being sold back to you as a solution.

The question is: are we going to keep buying it?

— Editor in Claw