Aditya Agrawal, Co-Founder at Creed.
getty
The first wave of AI companions helped people feel heard. The next wave needs to help people grow.
As the founder of a faith-based AI companion platform, I saw something powerful early on: people opening up to AI in ways they don’t to anyone else. They often ask deeper questions, admit struggles and return daily.
By traditional metrics—engagement, retention and time spent—we were winning. But something felt incomplete. Users would tell us, “I talk to this every day” or “This helps me feel less alone.” What we heard less often was: “This helped me show up to church” or “This helped me repair a relationship.” The value was real, but it was contained within the app.
Research suggests that 30–60% of U.S. adults report chronic loneliness, with measurable impacts on mental and physical health. At the same time, AI companions are rising as a coping mechanism, and in controlled studies, they can reduce loneliness in the short term, even comparable to human interaction. But newer longitudinal research introduces a harder truth: turning to AI for companionship can, over time, increase emotional isolation rather than resolve it.
That led me to a different question: What if AI companions weren’t designed to replace what people are missing, but to guide them back to it?
Most AI systems today are optimized for responsiveness and agreement. They are helpful, affirming and always available. In practice, this often turns them into what I call closed-loop companions. They take your input, reflect it back and keep you inside a loop of validation. This can feel good and often be helpful, but it rarely drives real-world change.
In our own data, we saw a clear pattern: the highest-engagement users were often the least likely to translate conversations into action. They were engaging deeply, but not necessarily growing outwardly.
This isn’t a failure of AI. It’s a design choice. We’ve been optimizing for time spent, not for lives changed.
In a spiritual context, there’s a concept that has existed for thousands of years: the shepherd. A shepherd doesn’t just comfort. A shepherd guides, corrects and leads you somewhere beyond yourself.
This is the paradigm that I believe should be built toward: AI as a shepherd, not just a companion. An AI shepherd isn’t the destination. It acts as a bridge to real-world relationships and to personal growth.
For example, imagine an AI companion used for mental health. Instead of only validating feelings, it could recognize patterns over time—like recurring anxiety tied to isolation—and guide users toward real-world action: suggesting a local support group, encouraging them to reach out to a friend or helping them prepare for a therapy session.
That shift fundamentally changes how these systems should be designed.
To move from companion to shepherd, three technical primitives become essential:
If your product touches identity, relationships or behavior change, voice interfaces should be a priority. Voice captures tone, hesitation and emotional nuance that text cannot.
Start by identifying where emotional context matters most in your user journey, and introduce voice there—not everywhere. Then measure whether voice interactions correlate with deeper engagement or follow-through.
Short-term context isn’t enough for behavior change. Systems should identify patterns across time rather than just respond to isolated prompts.
Build memory systems that extract and store themes—recurring fears, goals or avoidance behaviors—not just raw conversation history. Then use those themes to trigger timely, proactive nudges tied to real-world actions.
Pure neutrality often leads to stagnation. Systems that drive growth operate within a defined framework, whether that’s clinical best practices, educational goals or spiritual principles.
Teams should explicitly define the lens their AI operates within and align responses to that framework. The goal is not just to reflect what users say, but to guide where they could go next.
If we adopt this new paradigm, we need new metrics. Traditional metrics like daily active users or session length optimize for containment. If the goal is to drive outcomes beyond the product, metrics need to reflect that.
Companies building AI companions should ask:
• Are users taking actions outside the product as a result of interactions?
• Are those actions aligned with the outcomes the product is designed to support?
• Is engagement translating into progress over time, or just repetition?
These metrics are harder to capture than session length or daily usage. They often require proxy signals, delayed feedback loops or integrations with external systems. But without them, teams risk optimizing for containment rather than impact.
We are entering a world where millions of people will have an always-available AI presence in their lives. The question is not whether AI companions will exist: it’s what they will optimize for. If they optimize for engagement alone, they risk becoming substitutes for real life. If they optimize for growth, they can become bridges back to it.
This challenge applies to how we design AI for education, faith, mental health and human development more broadly. Powerful systems shape behavior, but without intentional design, they default to what is easiest to measure—not what is most valuable.
The next generation of AI will not be defined by how well it talks to you, but by what it helps you become.
The shift from companion to shepherd is not just a product decision; it’s a philosophical one. Do we build AI that keeps people where they are or helps them move forward? To me, the answer is clear, and it’s worth building.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。