The Comet Incident: How AI Products Lose User Trust Through Pricing Changes

A Feature Change That Sparked User Backlash

A Reddit post titled How You Lose Customers sparked widespread discussion about AI product pricing and user experience. The subject was Comet, the browser product from Perplexity, and the controversy centered on a change to its Assistant feature.

According to the official explanation: "Computer now handles the actions you ask Assistant to perform in Comet Desktop — from opening pages and filling out forms to completing tasks across tabs." What appeared to be a capability upgrade received a very different reception from actual users.

The user's criticism was sharp: "I watched Computer burn through credits for nothing. I watched the chain of thought as Comet crawled around web pages trying to figure out what to do. Absolutely not. What Comet does on web pages is not worth these tokens, Computer isn't good enough at it, and this bait and switch is unacceptable."

The Real Cost of AI Agents — and the Experience Gap

Behind this complaint lie two structural problems common to AI Agent products today.

Token Consumption vs. Perceived Value

Browser automation agents work by repeatedly reading page content and generating a chain of thought to infer the next action. This process is inherently token-intensive. Users watch the agent "crawl" across pages, trial-and-erroring its way through tasks — and every reasoning step burns through their paid credits.

When the task is simple (filling out a form, switching tabs), users instinctively ask: why did this tiny thing cost so many credits? Worse, if the agent ultimately fails to complete the task, every credit spent is a total loss. This combination of high cost and low success rate is the fastest way to drive users away from an Agent product.

Overpromising on Capability

Agent products are often marketed with the vision of "just give the command and let AI handle the rest." But in reality, complex cross-page and cross-tab operations remain a significant challenge for current models in terms of planning and execution. When there's a visible gap between marketing expectations and actual performance, user disappointment is amplified — because they've wasted not just time, but real money spent on credits.

Why "Bait and Switch" Destroys Trust Most

What angered this user most wasn't that the feature was poor — it was what he called the "bait and switch." He used a pointed analogy:

"Dear customer, thank you for purchasing your Whizmobile last year! We're pleased to announce that accelerating over 60 miles per hour will now require a GoFast subscription. We've added $25 in credits for your next 30 days of driving, after which the subscription will continue at 5 cents per mile over 60 mph. We're so glad to have you as a customer!"

The analogy cuts right to the heart of a particular business pattern: taking a core capability users already had and already paid for, repackaging it as a new service that requires additional payment or credit consumption. It's like buying a car and then finding out that pressing the accelerator past a certain speed costs extra.

For users, this kind of change breaks the most fundamental value contract. When people buy a product, they have clear expectations about what they'll get. When a company unilaterally moves existing features into a new billing framework — even with a small compensatory credit top-up — it reads as a hidden price hike, or outright deception.

The Pricing Dilemma for AI Products

From the company's perspective, these kinds of adjustments often come from genuine pressures. Large model inference is expensive, and the compute cost of Agent features far exceeds that of ordinary conversations. When a free or fixed-quota model can no longer cover real costs, shifting to usage-based pricing becomes almost inevitable.

The question is how to make that transition. This incident offers several lessons worth reflecting on for every AI product team:

• Transparency first: Users need to clearly understand roughly how many credits each action costs, and why. Opaque credit-burning is the fastest way to breed distrust.
• Don't touch existing entitlements: Re-billing users for capabilities they already paid for is what triggers trust collapse. New billing should apply to new capabilities — not erode existing promises.
• Success rate is a prerequisite for charging: Usage-based pricing only holds up when the agent can reliably complete tasks. If users are charged even when it fails, they'll feel like they're paying for the company's technical immaturity.

In the AI Era, Trust Is More Fragile Than Features

This Reddit complaint might look like a single user venting about a browser feature. But it reflects the pricing growing pains the entire AI product industry is going through. When model capabilities aren't fully mature and costs remain high, any change to a billing strategy risks crossing the line of user trust.

For AI products that aim to build for the long term, the real lesson may be this: technology can be iterated, but trust, once lost, is nearly impossible to recover. Before packaging features into subscription tiers and credit systems, companies need to honestly ask themselves — are you creating value for your users, or are you just reselling them something they already had?