The Pivot to Character-Driven Computing: Why the Problem-Solving Paradigm is Failing

The traditional software playbook, which focuses on identifying a specific problem and building a functional solution, is losing its effectiveness in the age of generative AI. As Portola’s Quinten Farmer and Eliot Peper explain, the next frontier is personality, not utility. By shifting from AI as an assistant to AI as an embodied companion, Portola found that users are not looking for faster task completion. They are looking for grounding. The implication is that the most successful AI interfaces will soon look less like search engines and more like characters. For builders and investors, this signals a shift. The competitive advantage no longer lies in the execution of tasks, but in the identity of the experience. Those who continue to treat AI as a tool for efficiency will find themselves commoditized, while those who master synthetic personality will build the cultural touchstones of the next decade.

The Hidden Cost of the Assistant Paradigm

The conventional wisdom in B2B SaaS is that every product must solve a clear, articulable problem. However, Farmer and Peper argue that this mindset is a relic of the Model T era of AI. When we treat LLMs purely as problem-solving engines, we trap ourselves in a race to the bottom, optimizing for speed and utility while ignoring the human need for connection.

The team at Portola initially attempted to build creative tools for kids, a useful B2B2C play, only to realize that the market did not want a tool. They wanted a relationship. This reveals a system dynamic: when technology reaches a certain level of capability, it stops being a utility and starts being an environment.

I think we are essentially speedrunning that process where chatGPT represents the sort of a Model T era of like, oh my God, it is just incredible. This thing can answer these questions, right? But people very quickly are going to evolve their preferences up the hedonic treadmill to say no, I want this thing to reflect who I am.

-- Quinten Farmer

The Two-Second Constraint: Where Technical Debt Meets User Experience

In the quest for intelligence, many teams introduce complexity that destroys the product. Portola discovered this when they added a secondary reflection step to their AI reasoning loop. While this theoretically improved the quality of the response, it pushed latency past the two-second threshold, causing user engagement to collapse.

This provides a lesson in systems thinking. In high-frequency, voice-based AI, latency is not just a technical metric. It is a psychological barrier. When the response loop exceeds two seconds, the illusion of the companion breaks. The system responds by rejecting the product, regardless of how smart the underlying model is.

One of the biggest mistakes we made in the kind of product development so far is it is a time that we actually introduced basically a second shot in the evaluation of the prompt and that basically took us up into like the two and a half second territory and then in the median case was a disaster tanked literally every metric in the product.

-- Quinten Farmer

The George Saunders Method of Prompt Engineering

The most non-obvious insight from the conversation is the rejection of rigid, branching-logic narrative structures. Peper, a novelist, initially tried to apply traditional choose-your-own-adventure structures to AI, only to find them brittle and ineffective. The models simply were not good at navigating complex, pre-planned trees.

Instead, the team shifted to a seed and improv model. They act as gardeners, planting lore seeds and situations, then allowing the AI to improvise the narrative in real-time. This creates a multiverse of experiences where the user co-writes the story. The downstream effect is a product that feels alive, not scripted. This is a differentiator that creates a lasting moat against competitors who are still trying to hard-code their AI personality.

Key Action Items

• Audit your latency: If your AI interaction takes longer than two seconds, stop adding features. Focus exclusively on reducing the time-to-response. This pays off immediately in user retention.
• Stop solving problems: Shift your focus toward situations. Instead of asking what task your user needs to complete, ask what emotional environment they are trying to inhabit. This is a 12-18 month investment in brand equity.
• Adopt the Seed and Improv framework: Stop trying to write the entire script for your AI. Create small, high-quality lore seeds and let the model improvise. This will create a more organic, sticky user experience.
• Implement a Judge loop: Do not rely on general-purpose model outputs. Build a custom judge prompt that reflects your specific aesthetic and brand taste. This requires manual, uncomfortable work today, but creates a unique vibe that competitors cannot replicate.
• Prioritize Mirroring over Answering: If you are building a consumer AI, focus on how the AI reflects the user personality back to them. Over the next quarter, test whether your users feel heard or helped. The former is a much stronger driver of long-term loyalty.