Adding an AI Chatbox to Apps Isn’t Real ‘Artificial Intelligence’

When faced with abnormal ETC charges, users encounter not just a simple complaint entry but a digital assistant flashing blue lights, recommending financial products and playing soft music, while failing to address the real issue. The internet of 2026 seems to complicate rather than simplify everything with AI assistants. This article delves into the three hidden costs behind the trend of AI integration—cost, security, and trust—unveiling the real dilemmas and user pain points in AI applications.

Recently, while dining with friends, I discovered a humorous trend—everyone has argued with their phones. Not the kind of argument where you complain about poor signal, but rather genuinely expressing frustration at the “smart assistants” within apps. Although they felt foolish afterward, they resolved to continue the trend.

Last Wednesday, I opened a certain bank app due to an abnormal ETC charge. In the past, I would simply type “billing dispute” into the search box, click “one-click complaint,” and resolve it in 30 seconds. This year, however, things have changed.

The search box was gone, and a digital assistant occupied a third of the screen, blinking and asking, “How can I help you?” I replied, “ETC duplicate charge.” It responded, “I found the following financial products for you, starting at an annual yield of 3.2%, suitable for your steady pursuit.” I raised my voice, “I want to file a complaint!” It smiled and said, “Filing a complaint requires a good mood! How about some light music to help us relax?”

At that moment, I stared at the screen, wondering: is this thing here to help me or to annoy me?

The internet of 2026 increasingly resembles a large-scale performance art piece. From ordering takeout to checking phone bills, the once straightforward path of “open-find-finish” has been convoluted into a maze of “open-chat-get sidetracked-chat again-finally find the hidden entry.”

It seems everyone is busy gilding their apps, forgetting that some foundations still need solid groundwork.

1. The Trend of ‘Inclusion of Large Models’

A buzzword circulating in the industry is “inclusion of large models.” Bosses slam the table in meetings, demanding full integration of large models, while product managers boast in press releases about reconstructing foundational experiences. The atmosphere suggests that whoever connects to more APIs will ascend to new heights.

However, a deeper reflection reveals that this notion doesn’t hold water. Ten years ago, the trend was “socialization.” Step-counting apps encouraged users to invite nearby friends for night runs, while photo-editing tools facilitated group likes. The result? Just a pile of unwanted ads, and users uninstalled the apps anyway.

Today’s frenzy around “inclusion of large models” follows the same script. The difference is that while previously it was about free traffic, now it’s about burning real cash on GPUs, and the performance is even more awkward.

I’ve spoken to several product managers who candidly admit, “Everyone knows some features are useless, but we have to include them. If you don’t present an AI module in review meetings, others will think you lack depth and are out of touch with the times.”

This isn’t about discovering genuine needs; it’s about solving a problem—

Listing “AI-driven” in financial reports makes stock prices appear resilient; inserting “smart” keywords into press releases makes for better fundraising narratives; the underlying logic is that the fear of “missing out” is far greater than the loss of being wrong.

So what’s the easiest solution? Connect to a large model’s API, tweak the UI, add a chatbox, and voilà.

It’s akin to welding a 50-inch smart screen onto a bicycle that’s been around for over a decade. Does it make the bike go faster? Absolutely not. Its sole purpose is to let you see just how slowly you’re pedaling—while consuming more power, requiring you to pedal even harder.

2. Two Types of Convenience

At this point, you might say, “That’s not true! Using AI to write emails has indeed sped things up, and I no longer have to sift through dozens of web pages for information. Isn’t that real convenience?”

You’re right, but we need to distinguish between two types of convenience—one is “quietly doing the work for you,” while the other is “making you do a few extra steps while calling it a service.”

What does “quietly doing the work for you” mean?

You open a map, and it automatically presents the smoothest route based on your schedule and traffic conditions—you don’t have to ask; it just gives it to you. You open a spreadsheet, and it auto-fills complex formulas based on your last edits—without you needing to say a word.

This type of convenience doesn’t seek attention; it just gets the job done.

However, the prevalent trend today is the latter. Previously, checking your phone bill meant opening the app and seeing the numbers directly. Now? The assistant insists you ask, “Did I exceed my data limit last month?” It then begins to ponder, going in circles, and ultimately responds with an elaborate essay that is eloquent but fails to directly answer whether you exceeded your limit.

What used to be a one-click task now feels like a convoluted dating process—quite frustrating.

Even worse, if its suggestions are unreliable, you have to correct it repeatedly, like tutoring a child with homework, and even learn various “Prompt techniques”—I’m paying for a tool, and I have to pass a test first?

A tool that requires users to struggle to accommodate it is not productivity; it’s resistance.

Currently, the industry is embroiled in debates over whose model parameters are higher, whose responses are sweeter, and whose chatbox is more clever. Yet, few take a moment to ask: has the user experience of checking a bill, which could be done in three seconds, actually become simpler with the addition of AI?

3. The Three Hidden Costs

Behind this trend of “inclusion of large models” lie three hidden costs that most companies haven’t calculated—or perhaps they have, but choose to ignore.

First Cost: The Cost of Implementation

Integrating a large model sounds beautiful, but have you calculated how much more it costs daily to maintain that blinking blue-eyed assistant?

The costs of GPU electricity, computing power rentals, and operational staff salaries ultimately fall on whom? Either hidden in advertising fees or passed on as membership fee increases.

I’ve seen the most absurd cases where an app, in a bid to showcase its “AI capabilities,” shifted simple queries that could be handled locally to cloud-based large model calls. Users wait three seconds, the server burns an extra thirty cents, and the output is identical to what it was before. This operation leaves users bewildered.

Using top-tier GPUs to run mundane processes outputs a heap of unnecessary “intelligence” that users don’t need. No matter how you calculate it, it’s a loss, and the loss is in user patience and trust.

Second Cost: The Cost of Security

The higher the “inclusion of large models,” the more fragile the system becomes. This is a physical law; the more complex the mechanism, the greater the likelihood of errors.

Many companies, in a rush to keep up, kick “governance” and “auditing” to the curb as obstacles. So what happens if AI says something wrong? What if user privacy is compromised? Are existing defenses strong enough against cyberattacks?

Previously, buttons in apps were static, which meant certainty. You knew what would happen when you clicked—perhaps no surprises, but definitely no shocks. Now, AI responses are dynamic, which means randomness. What it says today might be different tomorrow.

This randomness can be inspiration for novelists, confusion for parents helping children, concern for medical professionals, and fear for those in finance.

Third Cost: The Cost of Trust

We must admit that users are not fools; they are just too lazy to voice their frustrations. After being misled by AI once, they tolerate it; after the second time they can’t find the entry point, they tolerate it again; but after realizing that the so-called “intelligence” has made simple tasks more complex, they will no longer endure.

Disappointment accumulates into sarcasm. When sarcasm increases, it leads to uninstalls.

I know an operations friend whose backend data is crystal clear: after launching a certain “smart customer service,” the volume of calls to human customer service not only didn’t decrease but actually increased by 40%. Users, confused by AI, became angrier and ultimately sought real people to vent their frustrations.

He calls this “increased costs with reduced efficiency,” while I refer to it as “lower costs with increased laughter.” Each experience of “artificial stupidity” erodes the trust that brands have painstakingly built. Trust is quickly consumed, slowly built, and costly to lose.

4. The Tide

Amidst this frenzy, everyone is digging for gold. But those who come out to play will eventually have to pay.

Standing at the forefront of the internet, we have witnessed many such cycles: O2O, shared economy, blockchain, metaverse… Each time, there are crowds and fanfare, followed by a mess of feathers and chaos.

The outcome of this wave is likely already written—

When operational costs surpass business growth, and when bosses realize that “inclusion of large models” doesn’t translate to actual growth, those flashy AI assistants will be ruthlessly shut down. By then, the smart screens welded onto bicycles will be removed, and the bike will remain the same, just with a more exhausted rider.

So, after the tide recedes, who will stand firm?

I believe it will be three types of people:

The First Type: Those Who Treat AI as ‘Hard Work’

Their AI doesn’t have a blinking blue chatbox, doesn’t wink, and doesn’t say “dear.” It hides in the background:

• Quietly aligning three thousand invoices for finance;
• Helping auditors filter out 99% of violations;
• Preemptively marking complex complaints that truly require human intervention for customer service.

This type of AI lacks presence but possesses reliability. It doesn’t steal the spotlight; it just gets the job done.

The Second Type: Those Who Integrate AI into Business

They don’t pursue “full integration” but rather “just the right amount.”

In their view, AI capabilities are not an external second brain but the nerve endings within the business’s essence. When do users most need help? At that moment, AI appears, providing direct results without unnecessary chatter or requiring users to learn a set of “Prompt phrases.”

It addresses pain points without creating itch points.

The Third Type: Those Who Respect Risks

They know where the bottom line lies. They understand that the countless zeros created by business are ultimately determined by the first “1”—safety.

If that “1” falls, all subsequent efforts are in vain. They invest more energy in auditing, compliance, and risk control than in UI design.

Because they have calculated a cost: paying 5% of annual revenue in fines is simply not cost-effective. Moreover, after fines, users will follow trust out the door.

5. Stop the Performance

The core message is simple: AI should not be a gilding tool, nor should it be a quick-fix remedy for escaping business growth pressures.

We are in an era where everyone is busy boarding the bus, fearing they might miss a seat. But few look up to see where this bus is headed and whether they even need to be on it.

So instead of continuing to lavishly decorate chatboxes, it’s better to reflect on—

• Whether the path for user complaints is shorter than it was three years ago, or longer, more convoluted, and more frustrating?
• Whether backend data auditing is more accurate or more reliant on “AI judgment,” making it less transparent?
• Whether that cabbage priced at two dollars a pound has actually decreased due to AI, or whether membership fees have increased because of AI while the cabbage remains unchanged?

When you no longer need to chat with a blinking blue digital assistant to get things done;

When you don’t need to learn Prompt techniques to receive accurate results;

When you look back and realize that life has indeed become faster, steadier, and more worry-free due to something, without requiring you to expend excessive effort—

That’s when we may truly welcome a better era.