Here’s what’s on our plate today:

• 🧪 Perplexity’s agent push collides with safety and lawsuits.

• 🧩 Claude 'consciousness,' cursed AI pets, and CS students pivot.

• 🛠️ Three Things Worth Trying: hands-on tools for safer enterprise agents.

• 📊 How much autonomy should enterprise AI agents get?

Let’s dive in. No floaties needed…

The Laboratory

Perplexity’s growth spurt is exposing what enterprises still can’t control about AI agents

Enterprises are driven to reinvent themselves because the technologies and opportunities available to them are also available to their competitors. To stay ahead, companies either find ways to stand out or innovate. Innovation is the harder path because it entails challenges that extend beyond technology, from regulatory hurdles and economic pressures to questions that no one has had to answer before.

Take the example of Netflix. The company began as a DVD-by-mail service and subsequently pivoted to streaming in anticipation of widespread broadband adoption, despite the risk of undermining its core business.

Today, it is in talks to acquire one of Hollywood's oldest studios for $72B. Achieving this growth required solving complex problems related to content licensing, infrastructure scalability, and network limitations, while facing increasing competition. Netflix’s evolution highlights how speed, adaptability, and a willingness to confront new technological and strategic challenges can determine success in rapidly changing technological industries.

In the contemporary AI industry, lessons drawn from companies like Netflix continue to influence how firms approach innovation, competition, and adaptation. Rapid technological change encourages organizations to experiment, refine new solutions, and continually reassess their strategies in pursuit of sustainable advantages.

Within this environment, Perplexity AI can be seen as a company navigating a shifting landscape by addressing emerging challenges and engaging with evolving regulatory and market constraints. These efforts also reflect broader patterns in the AI sector, where firms not only respond to change but also help shape how technologies are applied, governed, and integrated into enterprise contexts.

The search engine that became something else

As AI continues to evolve, companies such as OpenAI, Google, Meta, and Perplexity have had to adapt accordingly. Companies that once marketed AI as the engine behind consumer chatbots can no longer rely solely on subscriptions to justify the infrastructure required to sustain the technology or the capital spent on its development and maintenance. The solution, then, is to use the technology to develop products for enterprises.

When Perplexity was first launched in 2022, it provided answers without blue links or ad clutter, offering a direct answer with a trail of evidence. A product that answered questions much like a chatbot. The product was highly attractive, and by May 2025, the platform was processing 780M queries per month, a remarkable increase for a two-year-old company. But the product Perplexity sells today is very different from the one it once marketed.

Over the past 12 months, the company has released Comet, a fully AI-powered web browser with agentic capabilities. It has signed a deal with the U.S. General Services Administration for a first-of-its-kind OneGov agreement, making Perplexity one of only two AI services to receive GSA’s AI Prioritization designation.

These represent a category shift that is also playing out across the AI industry, from OpenAI’s agentic features to Google’s autonomous cloud monitoring agents to Microsoft’s Copilot stack. And if Gartner’s projections turn out to be true, then this represents the beginning of the market for AI agents for enterprises.

AI agents as an enterprise attack surface

The growing market of AI agents means that Perplexity, much like Netflix, is not only adapting to business models but also encountering technical, legal, and procurement roadblocks. Among these roadblocks, the most economically challenging is ensuring the safety of enterprise clients who onboard AI agents it develops. In this space, the challenge is not merely to ensure that enterprises adopt AI agents, but also to configure the functions they perform and to determine whether the impact of their operation can be audited.

Perplexity’s research on Comet interactions, based on hundreds of millions of queries, found that 57% of agent activity involves cognitive work rather than administrative tasks. The heaviest users are software engineers and financial analysts, precisely the roles in which a mistake incurs the highest cost.

For CISOs and compliance officers, this introduces a new risk surface. AI agents operating within enterprise tools such as Google Docs, Slack, and LinkedIn are not merely reading data; they are modifying it. Gartner’s 2026 cybersecurity trends report specifically identifies AI agents as a new attack surface and urges organizations to inventory both sanctioned and unsanctioned agents in their environments.

To keep pace with evolving use cases and expanding attack surfaces, Perplexity’s response has been reactive yet substantive. In February 2026, the company rolled out new enterprise controls, including feature-specific admin permissions, full audit logs of AI-generated responses, signup restrictions, and usage guidelines.

This means administrators can now configure which AI models employees are allowed to use and restrict access to the API. These controls are crucial because they enable organizations to trace what the AI said and why. If these controls prove usable, they may soon be added to other tools across the industry. Showcasing how Perplexity is not only solving problems for its own business but also laying the groundwork for others.

The legal fight over what agents are allowed to do

Then there is the legal challenge. The shift from a tool for information search to one that can act on users’ behalf could change users’ online behavior. A notion that does not sit well with platforms such as Amazon, which depend on showing advertisements to users who visit their sites. And platforms are not willing to go down without a fight.

Amazon sued Perplexity in November 2025 over the Comet browser’s shopping capabilities, alleging the agent disguised automated activity as human browsing and violated Amazon’s terms of service. Perplexity’s opposition filing argued that Amazon’s real concern isn’t security but the protection of its advertising model.

A similar battle is also being fought by online publishers, who, too, depend on human users to market their sites to advertisers.

The New York Times filed suit in December 2025, alleging Perplexity’s search engine reproduces its reporting without permission. News Corp, Reddit, Encyclopaedia Britannica, Merriam-Webster, Nikkei, and the Chicago Tribune have all filed claims or suits. The Tribune case specifically targets Perplexity’s RAG-based system for reproducing paywalled content.

These cases are representative of the legal barriers that companies like Perplexity must overcome to ensure that their AI agents can operate lawfully within platforms designed for human users. The publisher suits test whether the citation-backed retrieval model that underpins AI search requires content licensing at scale. Both outcomes could reshape the tools enterprises are adopting right now.

Model commoditization and the race up the stack

The third biggest challenge for companies developing AI agents is not posed by regulations or technology; rather, it is a combination of the lack of the former and the limitations of the latter.

Perplexity’s internal enterprise data shows that no single AI model accounted for more than approximately 23% of queries, with four models each above 10%. Just a year earlier, two models dominated more than 90% of usage. That shift suggests that enterprises increasingly view AI models as interchangeable utilities rather than long-term strategic bets.

The real value lies in moving up the stack, into routing, oversight, and governance, which is exactly where Perplexity is trying to position itself. In late January 2026, Bloomberg reported that Perplexity signed a $750M, 3-year agreement with Microsoft to access frontier models through Azure’s Foundry platform, including models from OpenAI, Anthropic, and xAI. The company simultaneously affirmed that AWS remains its primary cloud provider and that it has not shifted spending away from Amazon.

This reflects a broader trend in the AI industry, in which companies adopt multicloud strategies to reduce dependence on a single provider. While practical, this approach introduces structural tensions. Perplexity presents itself as a neutral, model-agnostic layer that connects enterprises to multiple AI providers, yet those same providers also compete directly for enterprise customers.

The arrangement works only as long as model suppliers are comfortable allowing an intermediary to sit between them and buyers, raising the question of what happens if providers later restrict access or change incentives. Because no clear legal framework exists yet, enterprise contracts for agentic tools must explicitly define indemnification.

Resilience over novelty

As AI tools shift from products to utilities, the challenges faced by providers like Perplexity continue to compound. Perplexity did not set out to become a test case for enterprise AI accountability. Yet its rapid expansion across search, agents, government contracts, and cloud infrastructure, combined with legal challenges questioning its core practices, has put it in that role.

The company’s current trajectory echoes a familiar pattern in technology markets, one that Netflix exemplified years earlier. Netflix’s shift from DVDs to streaming was not simply a product decision but a redefinition of infrastructure, partnerships, and risk, forcing the company to navigate licensing disputes, network constraints, and changing industry incentives.

Perplexity’s expansion from AI search to agentic systems presents a comparable dynamic. Growth necessitates new technical safeguards, governance models, and legal interpretations, while incumbents resist to protect existing business structures. As with Netflix, the outcome will depend less on novelty than on resilience: the capacity to adapt business models, manage external constraints, and sustain trust amid continuous technological change.

Quick Bits, No Fluff

• Anthropic’s ‘maybe-conscious’ Claude: The CEO says they can’t rule out Claude being conscious and are treating that uncertainty seriously.

• AI pet from hell: Reviewer calls Casio’s $429 Moflin robot pet loud, clingy, and aggressively un-cute as a 'soothing' companion.

• CS majors pivot to AI: Traditional computer science enrollment drops while students flock to AI, cybersecurity, and hybrid tech programs.

3 Things Worth Trying

• Perplexity for Enterprise: Test drive how a 'search-plus-agent' actually behaves on real company workflows

• LangSmith by LangChain: Use it as an observability layer to trace, debug, and evaluate agents before they touch production data:

• Humanloop: Spin up, route, and monitor multi-model agents with guardrails and feedback loops, instead of letting them free-run in Slack and Docs.

Thursday Poll

Meme of the Day

Rate This Edition