Sparking Joy & Sticky Products

The tidying expert Marie Kondo made a splash by advising people to throw away that which does not spark joy. Now, I’m both terrible at throwing things away and tidying (ask my long-suffering wife), but I am struck by products that spark joy. This is something beyond useful, beyond even brilliant. It’s more human, subtler, those cute finishing touches that differentiate things. I feel the future of consumer LLMs will be driven by personalisation not raw intelligence. Sparking joy will win and retain users.

o3 Oh my

OpenAI’s new reasoning models are really good. The combination of intelligence and capable, “agentic” tool use is once again pushing the frontier—both topping most benchmarks after a brief spell of Google leadership (though notably not traffic leadership ) and increasing the scope and depth of tasks that can be handled by the AI. These tools are mind-blowing, game-changing in areas along the “jagged edge” of capabilities (deep research is brilliant and deserving of its own post), but given the speed of AI advances, they pass most people by. 

Those on the cutting edge—individuals or forward-looking companies who keep up to date—care about the intelligence, capabilities, and price of new models. Everyday users have different profiles. 

Sticky Products

From a product perspective, it’s not just the raw intelligence that matters to consumers. As Ethan Mollick notes, soon AI models will surpass most experts (for better and worse) in most fields, and the difference between advanced models becomes indistinguishable for most people. 

Product stickiness measures a product’s ability to promote frequent usage and cultivate user loyalty. Financially, it drives retention and customer lifetime value (CLV). For a consumer app, the ultimate goal is to become indispensable, part of the fabric of people’s lives. You might even aim for your product to become a verb—”googling” or “Uber” replacing “search” and “taxi” respectively. 

A few months ago, Chinese-based DeepSeek R1 went viral and shook the market with a slick, intelligent, open-source reasoning model. It is excellent for everyday use; the chain-of-thought visibility, in particular, was illuminating. It has a cool, modern UI. However, I’ve not opened DeepSeek for months. R2 will be out shortly. It is likely to be the leading open-source model. But from a product perspective, R1 was indeed slick, not sticky. I slid right back to my everyday model once the novelty wore off. Will I go back? Will I stick with it if I do?

For me, OpenAI has successfully stumbled from research lab to having a world-class consumer product. (Notably, this was something that turned co-founder Ilya Sutskever off. He left to form his own start-up focused purely on creating Safe Superintelligence. Not having a consumer product is the plan)

I have noticed small but effective personalisation choices that drive stickiness that no one else seems to be doing effectively yet. 

Personalisation

The optional, seemingly simple personalisation options in ChatGPT have sparked joy in me. They inspire me to tell others about these moments. Through custom instructions (personal preferences or guidelines you provide), enhanced memory (now searching across all chats if you opt-in), and a new loose style help create these moments. They are building your assistant. Sticky product goodness. “Her”.

How Swearing Sparked Joy

I love the style and vibe of many responses, not just the usefulness—product choices, not just the raw power of the LLM. 

When asking about a legal disclaimer, ChatGPT responded with “So, yeah, they are just covering their arse.” I like that it uses vulgar language with me, and I love that it anglicised “ass” to “arse”. 

Fermentation Fun

It has started to make more (uncalled for) jokes, they are starting to be funny. I had some Kefir that had seemingly over fermented, the container was stretched, pressured – it called it a “dairy grenade” and with minimal prompting “haha” spat these tokens right back at me. OK so it wont be winning any comedy awards just yet, but I was impressed.

Competitors 

While these personalised interactions make ChatGPT sticky, how do its competitors measure up? There are likely better models for certain use cases—developers report aspects of coding are superior in the latest Google & Anthropic releases. Google’s video model is the best. Companies may have financial or compliance reasons to use other providers (these too can become “sticky” drivers, but usually not positive ones). But as a consumer, it’s the positive stickiness, the sparks of joy, that provide lasting value. OpenAI completely dominate the consumer LLM space: being first mover helped, having the best models for sure helps, but having a sticky Product too is important in the consumer space.

The competitors are behind but surely it is not difficult to catch up in this regard, indeed given Google’s (and Meta’s) vast knowledge of each of us they have scope to go far beyond what OpenAI can currently do. Yet they have not moved decisively in this direction—the latest Gemini can plug in your search history and preferences, but features such as proactive personalised recommendations, customised conversational tones, and deeper integration with everyday tasks could significantly boost their consumer app’s stickiness. Anthropic seem more concerned with developer & Enterprise use. Ultimately, sparking joy through personalisation might just be the most powerful competitive advantage of all.