Perplexity CEO: SMBs Could Drive the Next Economic Boom

Summary

Perplexity CEO Aravind Srinivas argues that AI agents could help smaller, more AI-native businesses become highly productive and contribute more to economic growth. He says Perplexity aims to provide the tools those businesses need, rather than focus only on large enterprise customers.

Perplexity’s product strategy

• From answer engine to agent platform: Perplexity began as an answer engine in 2022, then added multi-step search, Deep Research, browser agents, and Perplexity Computer. Srinivas says agents need web search, browser access, personal context, files and tools, as well as the ability to coordinate different models.
• Perplexity Computer: The product is designed to delegate real computer-based work. Srinivas says it combines model orchestration, search, browser access, and cloud sandboxes while keeping a familiar prompt-based interface. He reports that company revenue is growing fivefold year over year and that enterprise use is accelerating.
• Business use case: He describes a one-person business that reportedly spends over $1 million a year on Perplexity while earning about $3 million by automating newsletters for clients. Srinivas uses this example to illustrate how AI tools can help small companies generate revenue without large teams.
• Model neutrality: Perplexity wants to coordinate across frontier and open-weight models rather than depend on a single model provider. Srinivas argues that advances by model developers can add capabilities and lower costs for Perplexity users.

Local, cloud, and hybrid AI

Srinivas sees a future in which AI workloads are split between data centers and devices controlled by users or businesses.

• Cloud inference is well suited to shared, repetitive workloads and many chat tasks.
• Local inference may become more attractive as agents use longer context windows, personal or business-specific information, and continuous automations. It can also give users more control over sensitive data.
• Hybrid computing lets users choose when work runs locally and when it is sent to cloud models. Perplexity’s portable Computer is described as a fully local setup that can optionally escalate tasks to the cloud. Its Apple silicon approach starts in the cloud and can run sensitive local files on the device.
• Sovereign and on-premise deployments are another possible option. Srinivas says Perplexity is exploring systems where the model, agent runtime, and user sessions run on hardware controlled by a government or organization. The trade-off is that Perplexity would have less visibility into bugs and feedback, potentially making the experience less polished.

Perplexity says it has no plans to build its own data centers. Srinivas argues that existing devices, homes, and offices contain compute capacity that could be used, while data-center construction faces constraints such as power, permits, and grid connections. He says the company intends to work with chip partners including NVIDIA, Apple, and Intel.

AI agents, security, and future research

• Agent safety: Srinivas distinguishes between one main agent using sub-agents and multiple coordinating parent agents, which he considers riskier. He says secure sandboxes, monitoring agents, and threat-detection classifiers will be important.
• Security projects: He mentions open-source tools called Bumblebee for devices and Numbat for servers. They are intended to audit agent activity, check for improper access to sensitive information, and issue alerts based on logs.
• Research agents: Srinivas’s longer-term goal is to build systems that can conduct extended research, test hypotheses, and refine them through experiments. He says current models are not yet sufficiently persistent or capable of autonomous, long-horizon research—especially when results are difficult to verify.

Business and product-design analysis

Srinivas argues that AI products should be judged less by traditional consumer-app measures such as network effects and billion-user scale, and more by depth and recurrence of use: whether users repeatedly rely on the product to do valuable work and earn money.

He also says the bottleneck in many technical tasks is shifting from writing code to deciding what to build—clarifying problems, constraints, specifications, risks, and strategy. He wants AI to help with that reasoning rather than simply produce code.

On product design, he credits Larry Page’s “the user is never wrong” philosophy. As AI capabilities grow, Srinivas says products should become intuitive enough that people do not need to learn prompting techniques, select models manually, or rely on staff to onboard them.

Reviews, guides, or tutorials

No product review, step-by-step guide, or tutorial is included. The discussion is an interview covering Perplexity’s strategy, products, and technical direction.

Main speakers and sources

• Aravind Srinivas — Perplexity co-founder and CEO; primary source for the company’s product, strategy, and technical claims.
• Molly O’Shea — Interviewer and host of Sourcery.
• The video also contains sponsor advertisements for Brex, Harvey, Turing, Zone, VCX by Fundrise, Public, and Deel; these are not part of the main interview.