If you grew up playing Pokémon, you remember Diglett. A small, unassuming creature that pops up from the ground, does its thing, and retreats. No drama. No fanfare. Just a round head appearing exactly where it is needed.
Diglett is the perfect metaphor for how AI should work in software products.
The Anti-Pattern: AI as Main Character
Open any product that launched an "AI feature" in the last two years. Chances are, you will find a chatbot bolted onto the sidebar, a sparkle icon that screams "hey, I'm AI!", and a dedicated AI page that tries to justify the feature's existence.
This is AI as main character. It demands attention. It asks you to change your workflow. It creates a new thing to learn instead of making the existing thing easier.
Users tolerate this for about a week. Then they stop clicking the sparkle icon. The chatbot sidebar stays collapsed. The AI page collects dust.
The problem is not the AI. The problem is the visibility.
Surface Area vs. Value
There is an inverse relationship between how visible an AI feature is and how much value it delivers over time.
| Visibility | Example | Initial Excitement | Long-Term Value | |---|---|---|---| | Very High | Dedicated AI chatbot page | High | Low | | High | AI-powered search bar | Medium | Medium | | Low | Auto-categorization of items | Low | High | | Nearly Invisible | Smart defaults and pre-filled forms | None | Very High |
The features that deliver the most value are the ones users forget are powered by AI. They just think the product is "smart."
The Diglett Design Pattern
Here is how to design AI features that pop up where they are needed and disappear when they are not:
1. Trigger on Context, Not on Click
A Diglett feature activates when conditions are right, not when the user clicks a button. Examples:
- When a user pastes a long email into a notes field, automatically offer to extract action items
- When a user creates their third lead in a row, pre-fill common fields based on patterns
- When a user searches for something that returns no results, suggest related items
The key: the AI is responding to user behavior, not demanding user attention.
2. Present Suggestions, Not Takeovers
When a Diglett feature surfaces, it should be a gentle suggestion—a tooltip, an inline recommendation, a subtle highlight. Never a modal. Never a forced interaction.
The user should be able to glance at the suggestion and either accept it with one click or ignore it with zero clicks. If ignoring requires dismissing a dialog, you have built the wrong thing.
3. Learn from Dismissals
This is what separates good Diglett features from annoying ones. If a user ignores or dismisses a suggestion three times, stop showing it. The AI should learn from rejection as quickly as it learns from acceptance.
Most AI features only learn from positive signals (user clicked, user accepted). The best ones learn equally from negative signals (user ignored, user dismissed, user undid).
4. Celebrate the Mundane
The most impactful Diglett features automate boring, repetitive tasks that users do not even think about:
- Auto-tagging incoming data based on content
- Suggesting due dates based on historical patterns
- Detecting duplicate entries before they are saved
- Formatting phone numbers, addresses, and names consistently
None of these are exciting. None of them will make a good demo video. All of them save users hours per week.
Why Invisible AI Is Harder to Build
There is a reason most teams build the chatbot instead of the smart defaults. Visible AI is easier to demonstrate, easier to measure, and easier to get buy-in for.
"Look, we built an AI chatbot!" gets applause in a sprint demo. "We reduced form completion time by 23% using predictive field population" gets a polite nod.
But the polite nod is where the money is. Reduced form completion time means more data entered, more leads captured, more deals closed. The chatbot means another thing for users to learn and probably ignore.
Building invisible AI requires:
- Deep understanding of user workflows. You cannot automate what you do not understand.
- Instrumentation. You need to measure micro-interactions (field focus time, correction rate, abandonment points) that most analytics tools do not track.
- Patience. Invisible AI does not generate impressive screenshots for the landing page. Its value shows up in retention and NPS, not in feature announcements.
The Diglett Test
Before shipping any AI feature, ask yourself the Diglett Test:
- Does it appear only when the user needs it?
- Can the user ignore it with zero effort?
- Does it learn from being ignored?
- Would the user describe the product as "smart" rather than "AI-powered"?
If you answer yes to all four, you have a Diglett. Ship it.
If you answer no to any of them, you might have a Charizard—impressive to look at, but exhausting to deal with on a daily basis.
The best AI is like the best infrastructure: when it is working perfectly, nobody notices it is there. They just notice that everything feels a little easier, a little faster, a little more like the software is reading their mind.
Pop up. Help out. Disappear. That is the Diglett Principle.