To win the AI wars, make your AI invisible
Start using Visor today
- Visualize all your projects and data
- Roll up multiple projects into portfolio views
- Try Visor for FREE
The term “AI-first” is a giant red flag. Unless you are selling AI models like OpenAI orAnthropic, claiming to be AI-first is a great way to signal that you’re more interested in hype than value.
We saw how this played out with the “dot-com” revolution, and it’s going to rhyme this time: customers generally don’t care what technology you use. It’s about whether it improves their life enough to justify changing their behavior. So if you’re pitching your business as an “AI” company, wake up: you are confused about what your customers are actually buying.
What we can learn from the prior hype cycles
eBay, founded in 1995, is one of the dot-coms that became an enduring business. That’s because eBay (originally called AuctionWeb) leveraged the internet to solve a critical marketplace problem: geographic barriers. Compared to a flea market, accessing ebay.com is easier and offers an exponentially wider selection.
Next, consider the mobile revolution. It turns out ordering a ride is way easier when you can tap a few buttons on a location-aware device. That solved the ride-hailing problem dramatically better than calling a dispatch office or hailing a cab on the street. Soon enough, we stopped calling taxis and began getting Lyfts. But we don’t think of Lyft as a “mobile” company; Lyft is a transportation business.
The dot-com and mobile revolutions were massive because the technology solved so many real problems in novel ways. But don’t forget the change in behavior – and costs – these technologies overcame to become commonplace. Stop and think about how much you spend on having your internet-connected phone. I’ll bet it’s close to $1500 per year, maybe more. Not to mention the burden of charging it, encasing it, insuring it (AppleCare, anyone?), and keeping it with you at all times. We don’t even think twice about this anymore: it’s instinctive and natural that we just tote around these thin bricks wherever we go – even regularly into the bathroom. The internet and mobile devices are seamlessly integrated into our lives.
However, not all hype cycles end with lasting changes to how we live, work, and… er, “eliminate.” It’s been years since the peak of the blockchain hype bubble, yet few of us interact in a meaningful way with distributed ledgers, NFTs, and cryptocurrency. That’s because the technology was mainly an alternative to centralization. It turns out that very few things actually benefit enough from a decentralized ledger to justify a widespread behavior change and cost.
Blockchain tech failed to integrate seamlessly into the lives of those outside the bubble. Attempts continue. But with rare exceptions, these are solutions chasing problems.
This conclusion seems trite but is commonly forgotten: customers change their behavior for benefits, not technology.
What’s next for the AI hype cycle?
So where is the AI hype cycle right now? Gartner believes that generative AI is already falling towards the “Trough of Disillusionment.” But right behind it is something still climbing towards the “Peak of Inflated Expectations:” autonomous agents.
There’s a gold rush to replace humans with AI agents. (The more acceptable euphemism is saying “AI partners” will “amplify and augment” humans. But everyone knows where this eventually leads. Right?)
Even in the red hot “AI agent” market, customers aren’t buying AI agents. They’re buying the results. Here’s proof:
Salesforce just announced a downbeat forecast for fiscal 2026, dragged lower by slow adoption of its Agentforce platform. The reason? Customers aren’t getting the results yet. As one industry analyst remarked:
“Given how poor initial generative AI experiments were for many companies, they’re not just writing blank checks until Salesforce shows them Agentforce actually works.” – Rebecca Wettemann, CEO of industry analyst firm Valoir
Fortunately, AI solves workflow automation’s biggest hurdles. It’s just a matter of time to connect the pieces. AI gracefully handles unpredictable inputs; it can reason about how to solve complex problems; and it can generate things like text, images, and even voice. While today’s demonstrations are a little clumsy, it’s easy to see a medium-term future where things that once required a human touch – like customer service – are going to be handled by AI. Don’t believe me? Try a demo here.
Will AI be transformative like mobile devices and the internet?
Within our lifetime, I’m confident AI will eventually rise to the level described by Anthropic CEO Dario Amodei: “When AI systems are better than humans at almost everything. Better than almost all humans at almost everything. And then eventually better than all humans at everything.”
But side-stepping the existential questions this invariably raises, the pertinent question seems to be: will AI seamlessly integrate into our lives like the internet and mobile devices? The answer seems to be clearly yes.
But I just pulled a trick, dear reader. The question was more important than the answer. That question assumes the true test of AI will be whether it “seamlessly integrates into our lives.” But if that assumption didn’t bother you, then you’re onboard. To win at AI, you have to make it either invisible or effortless for your customers.
Imagine an AI that delivers all its benefits without requiring you to change the way you work or live. This is what I call “Ambient AI.” Rather than demanding your attention, Ambient AI operates subtly—optimizing your experience and amplifying your outputs almost invisibly. It becomes an ethereal ally, intelligently configuring systems, streamlining workflows, and achieving results with minimal intervention. In essence, it empowers you by simply making your existing tools work better, without disrupting your natural habits.
By now, you’ve seen the flashy AI that demands your attention and employs you as an accidental prompt engineer. Technologists think putting a big white text box with yellow sparkle emoji on their product invites a world of possibility. It feels like an emperor’s new clothes moment: “Can’t you see the possibility? Aren’t you smart enough to figure out the magical incantations for our genius AI?”
Real users see these as scary white boxes of fear and frustration. It’s a trap. It’s just too soon for this.
The psychology of user experience design
My interest in user experiences officially began with my undergraduate thesis on the psychology of user experience design. I researched how the human mind receives information from the world around us, borrowing time from Princeton’s most renowned psychology professors. The result was a deeper appreciation for how well-designed graphic user interfaces (or “GUIs”) can improve cognition by adapting to the ways our brains process information. For example, knowing how the cones in our eyes receive photons and how our brains process images can improve the way designers lay out information on a page.
In 2023, my company Visor began exploring “AI-powered fields” as a concept for enabling customers to calculate information without writing formulas. The research dramatically changed my perspective on the trajectory of this technology. The research participants were widely befuddled by the options we presented: pick a model, choose a temperature, write your prompt. Even as we simplified it to just the prompt “What do you want to calculate”, they didn’t know what it could – and couldn’t – do. It was like we had just time-warped them in front of an IBM 3270 mainframe terminal with a flashing phosphor green cursor.
I quickly realized why the GUI with buttons and organized layouts became so important vs. typing commands in a terminal: the GUI advertises exactly what it can do via buttons. You don’t need to guess whether it can print (and how to coax it to do that). You just find the button and click it. Satisfying!
There will come a time where AI becomes so capable that humans will not need to “prompt engineer” to get value. Humans will be able to communicate naturally, effortlessly. Even then, I foresee a seamless experience combining the best parts of a GUI and a conversational experience (whether by text or voice). Certain parts of a GUI deliver too much value in clarity and efficiency to ever fully get replaced by a text- or voice-only conversational experience.
Would you really prefer having to listen to a server at a restaurant read out the entire menu to you vs. just seeing the menu yourself, with headings and sections like “Appetizers” and “Drinks”? Is it really 10x easier to have to type (or speak) “I just met the Acme team and they are interested in moving forward with a fifty-five thousand dollar purchase” vs. typing the number in a field?
But today’s AI isn’t quite at the level required to enable natural conversational exchanges for accomplishing work tasks, leading to disappointing experiences and disillusioned customers. Nobody is going to switch to working this way just because it’s new. It has to be better. How ironic that in 2025 everyone thought it would be revolutionary to give users a new kind of terminal. Didn’t we learn this lesson already in the ’70s?
Where is AI the answer to a real problem?
Based on these research findings, Visor continued AI explorations but decided (much like Salesforce’s customers) to only invest when AI solved something better than any other tool. Ideally, it would be something that required no change in user behavior.
That day came when we decided to solve the hardest problem our customers experience on day 1: configuring our flexible product to suit their needs. Based on our analysis, this is where the vast majority of our users hit a wall. We’re calling it “Smart Templates,” and you’d barely know it was AI except for the data processing disclosures.
The challenge with selling horizontal, flexible products via a product-led growth strategy is that the product can only deliver value if customers configure it properly. It’s like selling LEGOs, except the customer only buys if they like what they construct with your building blocks.
Offering templates is an ideal solution for most products to bypass the setup, but templates are hard for Visor because they need to work with integrated data. How do you set up a board view for a customer if you don’t know what fields they have to organize the board? I simply couldn’t stomach building a “Template Setup Wizard” where customers have to pick the fields Mad-Libs style from Jira or Salesforce. What a drag! It defeats the one-click expectation set by the term “Template.”
It turns out, AI is the perfect tool for solving this problem for our customers (and Visor). With our new Smart Templates, customers just pick a template from a list. Then our AI automatically configures Visor for them based on this template and what it knows about their integrated apps. It’s like magic, and it’s a perfect example of Ambient AI in practice. The user does not need to become an involuntary prompt engineer. With Visor’s new Smart Templates, customers just click a button, and they get that template.
The best AI is the kind you barely notice
There will come a time when SaaS applications cease to exist. AI workers will not need these structures to organize and execute work. And humans will interact with their AI counterparts in some form of hybrid experience combining GUI and conversation. But there’s an important phase upon us right now: the transition, where our SaaS apps remain critical.
In my vision for this transition, AI is the beating heart that brings existing products to life. Imagine an email client that writes your emails, a code editor that fills the next line, and… Oh, wait. That’s already exactly how successful AI is working its way into our daily lives.
Code completion assistant Cursor went from $4 million in annually recurring revenue to $48 million between April and October of 2024. They 12-tupled in size in less than a year. Their killer feature? The AI predicts what software engineers will want to type next and allows you to just press the tab key to accept it. As a customer of Github Copilot, Microsoft’s product in this domain, I can confirm it works surprisingly well. It’s become second nature. Ambient AI sells.
The goal of Ambient AI is for products to feel less like a tool and more like an extension of your own mind – or perhaps more like an interaction with something sentient.
If SaaS apps may eventually go extinct in their current form factor, what will replace them? Here’s one proposal: rather than making a layer of dynamic AI agents that interact with our static SaaS apps (which are basically just databases), another possible future is the SaaS apps becoming as intelligent and autonomous as the agents.
It stands to reason that the AI economy may reward specialization like the human economy does. A persuasive sales AI might not be ideally suited to GAAP accounting, and vice versa, so these will probably come from different vendors. SaaS vendors may be able to save themselves from obsoleteness by embracing a future where they become specialized agents, operating as part of a digital workforce. Either way, the customers will decide what best solves their problems. They’ll reward the vendors that offer the highest value-to-cost ratio, keeping in mind that behavior change is among the highest costs out there.
So consider skipping the in-between stage where you think of yourself as AI-first and bludgeon your users with whiz-bang AI sparkle emojis, befuddling text boxes, and massive behavior changes. Stay focused on doing job they hired you to do – better. Because if they can, they’ll eventually hire something that’s better – and maybe cheaper.
