Perplexity has achieved one of the most remarkable growth stories in AI - scaling its valuation nearly 30x in just 18 months, from around US$500 million to around US$15 billion.
In a crowded world of copilots and AI apps, it's not just riding the wave of foundation models - it's best known today as a fast-growing consumer app, but it's also building an enterprise-grade product layer that users return to and increasingly rely on for work.
As Head of Enterprise Product, Frank te Pas is focused on building tools that don’t just answer questions, but become embedded into core enterprise workflows. In this conversation, Frank shares how Perplexity is thinking about differentiation, what "AI-native" product teams really look like, and the underrated metrics that help enterprise PMs stay on track.
Q: What do you look for when hiring product managers in the AI space - especially when many don’t come from AI backgrounds?
Frank: The speed of change in AI is so high that velocity itself becomes a competitive advantage. To keep up, I believe product teams must have four key traits:
Principle-based decision making: In a space where capabilities shift every month, PMs need clear product principles to guide decisions. This clarity helps avoid being reactive
Strong orchestrators, not bottlenecks: Engineers and designers are often highly product-minded. PMs should act as the glue, facilitating momentum rather than slowing it down
Clarity in chaos: AI introduces new ambiguity. A strong PM can take an overwhelming set of possibilities and distill it into a clear product direction
Creativity: With AI unlocking faster build cycles, the ideas you pursue matter more. Creativity becomes a key differentiator as the barrier to execution drops
Q: What does it mean to be an "AI-native" product manager?
Frank: Being AI-native means two things. First, you use AI yourself - to automate lower-value tasks like transcription, enablement materials, and reporting - so you can focus on higher-order strategy and customer discovery.
But more importantly, you understand where the role is going: PMs will soon manage fleets of agents. These AI agents will conduct research, draft content, and complete product ops tasks - while checking in for human oversight. The PM becomes a conductor.
Eventually, I think we’ll see role convergence: engineers, designers, and PMs will blur into "product builders," empowered by AI. In this world, creativity and orchestration matter more than traditional role boundaries.
Q: Enterprise products typically have slower feedback loops than consumer. What metrics do you track to know you're on the right path?
Frank: Unlike consumer products, enterprise teams can’t rely on instant feedback. The stakes are higher - enterprise users depend on these tools to perform their jobs, which means reliability, consistency, and trustworthiness aren’t just nice-to-haves; they’re non-negotiables. As Perplexity launches deeper into the enterprise, there’s more it can do to meet that bar - ensuring the product is robust enough to support mission-critical work, even in high-stakes environments. Here’s how we handle it:
Define hero queries: Start by identifying 50–100 representative use cases your product must excel at - a proxy for real-world customer value
Build a scoring rubric: For each, define an ideal answer and evaluation method
Set a go/no-go threshold: For Perplexity, a typical launch bar might be hitting 80% quality on this eval dataset
Track post-launch evals: Continue scoring real-world usage to monitor quality, especially when making changes to models or UX
This lets us measure progress even before we ship.
Q: Everyone talks about deep workflow integration. What does that actually look like to you?
Frank: To me, the core principle is: Meet users where they do their work.
Don’t force users to adopt new suboptimal behaviors. Embed into their natural workflow and aim to be best-in-class at just one of the steps in their broader process. Then earn your way into more.
For example, if a workflow has 10 steps, start by owning one - but do it better than any other tool. From there, expand to three steps, then more. Always reduce friction on both entry and exit.
Take a common enterprise use case like investment or market research. Step one might be Perplexity helping users synthesize the latest web information around a sector or company. A natural expansion would be to then compare that research with internal theses or proprietary insights from internal knowledge bases - validating or contradicting assumptions. Finally, Perplexity could enable users to export that output into structured formats that integrate easily into the next step of the workflow.
That’s how we design at Perplexity - by delivering value at critical moments and integrating seamlessly with the enterprise stack.
Q: How do you decide which workflows or steps to go after first?
Frank: We use two filters:
High-impact pain points: Where are customers spending lots of time on frustrating or inefficient processes?
Product edge: Where can we uniquely deliver better outcomes based on our capabilities?
For Perplexity, that means:
High-quality web search: That's been our focus from day one - building a proprietary index optimized for grounded, source-rich results tailored to AI search
Intuitive, collaborative UI: Our team is obsessed with user experience. We aim to make every interaction feel simple, fast, and team-friendly
Orchestration: We're doubling down on making Perplexity not just a Q&A layer, but an action-oriented assistant that can handle a broader set of tasks end-to-end
Model-agnostic infrastructure: We ensure that users always get access to the best available models in the market, without being locked into a single provider.
These strengths help us shine in workflows like research, content generation, and light action-taking. But we also know where not to go - like fully replacing PowerPoint or Figma.
Q: With so many copilots and AI wrappers out there, how does Perplexity stand out?
Frank: I see three differentiators:
Answer quality: At the end of the day, users care about whether the answer they receive is actually helpful, accurate, and trustworthy. We obsess over delivering the highest-quality answers, every single time. That’s what keeps users coming back
Advanced orchestration layer: A lot of the competitive intensity is shifting away from just building tools to how those tools are orchestrated. We focus on enabling action-oriented workflows - combining capabilities like advanced search, internal knowledge access, and code execution - and intelligently selecting the right tool for each task. The goal is to turn fragmented inputs into cohesive, high-impact workflows. That orchestration layer is where the next level of differentiation is happening
Model-agnostic infrastructure: We believe in using the best model for the task, not being tied to one provider. That’s why we can support emerging models like DeepSeek almost instantly. Our users don’t need to pick between different models - we route intelligently to the model that delivers the best result, and we update constantly as new innovations emerge
Taken together, these differentiators let us evolve faster, deliver better answers, and avoid platform lock-in. That’s the core of our advantage.
Q: Final thought - will Perplexity pick the best model for me?
Frank: Yes. That’s exactly what happens. We dynamically select the right model for your query. Simpler tasks may use faster models; deeper research calls for something more powerful.
This flexibility helps users get the best output balancing depth and speed.