Perplexity just launched Computer for Enterprise, bringing its multi-model AI orchestration platform to corporate customers. The company valued at $20 billion is now asking CISOs to route their Snowflake queries, legal contracts, and proprietary business intelligence through an AI agent that coordinates 20 different models from five competing providers.

Photo by Mateusz Dach on Pexels

The move positions a three-year-old startup directly against Microsoft Copilot, Salesforce Einstein, Google Workspace AI, and Anthropic's Claude Cowork. But Perplexity isn't selling a better AI model—it's selling the orchestration layer that sits on top of everyone else's models.

That might be the most strategically interesting bet in enterprise AI right now.

The Architecture: 20 Models, One Coordination Layer

Here's what Computer for Enterprise actually does: when you ask it to prepare a briefing document pulling from internal Slack, emails, Notion, and the open web, it decomposes that request into subtasks and assigns each to whichever AI model handles that specific job best.

The technical stack:

• Claude Opus 4.6 (Anthropic) for primary reasoning
• Gemini (Google) for deep research
• GPT-5.2 (OpenAI) for long-context recall and web search
• Grok (xAI) for speed-sensitive tasks
• Nano Banana (Google) for image generation
• Veo 3.1 (Google) for video

Each session runs in an isolated Firecracker virtual machine—the same microVM technology AWS built for Lambda. That architectural choice matters: it means one customer's session physically cannot access another's data or infrastructure. For regulated industries evaluating the product, that isolation layer is table stakes.

Photo by Tara Winstead on Pexels

But the orchestration engine is the real product. Perplexity's internal data tells the story: a year ago, 90% of queries routed to just two models. By December 2025, no single model commanded more than 25% of usage. The implication: as AI capabilities fragment across providers, the coordination layer becomes more valuable than any individual model.

The Distribution Strategy That Built Slack

Computer started as an internal Slack bot at Perplexity before it ever reached consumers. According to Dmitry Shevelenko, the company's head of business, employees couldn't stop using it:

"Our finance team automated how they run accounts receivable. Everybody on our enterprise sales team automated how they create proposals. And that all came from cross-pollination—employees observing each other's queries in shared Slack channels and realizing what the technology could do."

That organic spread addresses enterprise AI's thorniest problem: adoption. When employees see how colleagues use the tool in shared channels and can reply with follow-up questions, learning happens without formal training programs. The transparency of shared Slack becomes its own onboarding mechanism.

The Slack-first distribution playbook worked for Slack, Dropbox, and Zoom—all consumer products that became enterprise staples through bottom-up adoption. Perplexity is betting the same pattern works for AI agents.

The enterprise features announced this week include:

• Direct @computer mentions in Slack channels
• Business-grade connectors for Snowflake, Datadog, Salesforce, SharePoint, HubSpot
• Purpose-built workflow templates for legal, finance, sales, customer support
• SSO/SAML authentication, SCIM provisioning, SOC 2 Type II certification
• Zero data retention options
• Usage-based billing (organization credit pool with admin controls)

Non-Technical Workers Querying Snowflake in Plain English

The Snowflake and Datadog connectors—exclusive to enterprise customers—may be the most quietly disruptive feature. They let non-technical employees query corporate data warehouses in plain English, eliminating the bottleneck of waiting for data teams to write SQL.

Shevelenko described his own experience that morning: asking Computer what percentage of users who ran a query also purchased additional credits beyond their subscription. "I'm not technical, so I don't know how to write SQL. I wouldn't have known where to look that up. If I had to ping one of our data scientists—even though I'm an executive—it probably still would have been a few hours before I got a response.

Here, within a minute, I got a full analysis."

Another case: a sales rep needed a customer reference from a B2B media company. Previously, that meant posting in Slack, hoping the right person saw it, hoping they replied. Instead, the rep asked Computer to identify the company's most loyal B2B media customers and the account reps managing those relationships. Three recommendations delivered in minutes.

This pattern—collapsing multi-step, multi-person workflows into a single natural-language query—is the core enterprise value proposition. It's not a chatbot that answers questions. It's an agent that does work.

Photo by Lukas on Pexels

The Competitive Positioning: Orchestration vs. Lock-In

Microsoft has embedded Copilot across Microsoft 365. Salesforce has Einstein. Google keeps expanding Gemini's reach. Anthropic's Claude Cowork targets similar enterprise use cases—but relies exclusively on Anthropic's models.

Perplexity's argument: its multi-model orchestration is a structural advantage. Rather than locking enterprises into a single vendor's AI ecosystem, Computer selects the best model for each subtask automatically.

Shevelenko positioned it as "the tool that combines every other tool," but he also took a direct shot at the open-source alternative: "There's really no solution out there that is this comprehensive and has security. You obviously have stuff like OpenClaw, which has an open library of skills and connectors, but it is a security nightmare."

The reference to OpenClaw—the popular open-source AI agent framework with powerful capabilities but minimal enterprise governance—draws the competitive line. OpenClaw runs locally on users' machines. Computer runs entirely in the cloud with enterprise-grade audit trails and administrator controls.

But Perplexity's multi-vendor strategy carries risk. The orchestration layer depends on continued API access to models from Anthropic, Google, OpenAI, and xAI—all competitors to varying degrees. If any provider restricts access, raises prices, or degrades performance for third-party callers, the value proposition weakens.

Market Timing: The Agentic AI Inflection Point

Shevelenko made a claim about timing worth examining: Computer couldn't have worked even three months ago because underlying agent models—particularly Claude Opus 4.6—weren't capable enough to power reliable multi-step orchestration.

"I don't think Computer could have been as powerful as it is if we had launched it even three months ago," he said. "You have to have the right harness at the right time. Every technology we've built laddered up to this moment, where you finally had a smart enough agent model to orchestrate through that harness."

The market data supports this timing thesis:

• 100% of surveyed enterprises plan to expand agentic AI use in 2026 (CrewAI survey)
• 65% already use AI agents in production
• Organizations have automated 31% of workflows on average
• Global agentic AI market projected to grow from $9.14B in 2026 to $139B by 2034 (Fortune Business Insights)
• 40% of enterprise applications will feature task-specific AI agents by 2026, up from <5% in 2025 (Gartner)

Perplexity is entering the enterprise agent market exactly when demand is inflecting upward and underlying models have crossed a reliability threshold.

The Pricing Model: Usage-Based vs. Per-Seat

Computer's enterprise pricing breaks from the per-seat licensing that dominates enterprise software. Enterprise Max subscribers get a per-user credit allotment. Organizations can purchase additional credits when usage exceeds that baseline. Administrators control spending at employee, team, or company levels.

The rationale: cost varies dramatically by task. Generating a video costs significantly more than a text briefing document (roughly 15 cents of compute). Shevelenko estimated a typical briefing that would take a chief of staff five hours costs about 15 cents in compute. "I feel pretty good about the ROI there."

The usage-based approach makes sense for workloads ranging from one-off research queries to multi-day campaign automation. But it creates a different challenge: enterprise procurement teams prefer predictable costs. Usage-based billing can trigger budget shock when adoption spreads faster than expected.

The financial stakes are substantial. Perplexity hit approximately $148 million in annualized revenue by mid-2025, with internal projections targeting $656 million annual recurring revenue by end of 2026—requiring roughly 230% growth. Computer for Enterprise, priced to scale across entire organizations, is clearly designed to close that gap.

The Trust Question: Three Years Old, Asking for Crown Jewels

The biggest obstacle isn't technology—it's trust. Perplexity is three years old and asking chief information security officers to route sensitive Snowflake data, legal contracts, and proprietary business intelligence through its platform.

Shevelenko addressed this directly: "The story of Perplexity in the enterprise is—we have tens of thousands of enterprise customers and only six people on our enterprise go-to-market team. It is because the leaders at these enterprises use Perplexity themselves. There's actually a lot of trust that we've created because people have used the product personally."

He drew on his board role at Lazard, the investment bank: "Financial services, a lot of very sensitive data, regulatory scrutiny. They wouldn't have appointed me to the board if they didn't care a lot about AI transformation. I have a deep empathy for the constraints, especially in regulated industries."

That bottom-up adoption pattern—executives discover as consumers, then advocate for company-wide deployment—mirrors how Slack and Zoom became enterprise infrastructure. But those companies sold relatively simple, well-understood products. Perplexity is asking organizations to trust an AI agent with access to their most sensitive data systems, running autonomously, making decisions about which external AI models to route queries through.

The security architecture—Firecracker VM isolation, SOC 2 Type II compliance, zero data retention options, audit logging—answers the technical concerns. Each enterprise will need to evaluate against their own risk tolerance.

What Happens When the Orchestration Layer Becomes the Product

The most provocative question: is Computer an interface layer that sits on top of existing enterprise software—or a replacement for it?

If Computer can query Salesforce, generate reports from Snowflake, manage campaigns through marketing APIs, and build internal tools on the fly, how long before enterprises question why they need those underlying platforms?

Shevelenko offered a measured answer: "I think the world of productivity software contains multitudes. Many organizations will have an all-the-above strategy. They'll keep their existing vendors, but they will also encourage internal teams to use Perplexity Computer to fill the gaps."

He pointed to viral Bloomberg Terminal demonstrations: "To be clear, you cannot one-shot Bloomberg. Bloomberg has all kinds of things you can't rebuild with Perplexity Computer. But there are certainly gaps in Bloomberg Terminal that you can make better with Perplexity Computer. And there's certainly a lot of people that can't afford a $35,000-a-year piece of software that can get a $200-a-month version through Consumer Max."

That framing—gap-filling and democratization, not displacement—is diplomatically smart. Perplexity needs Salesforce, Snowflake, and SharePoint as connector partners even as Computer potentially reduces time employees spend inside those platforms.

It's a classic platform-layer tension. Every successful middleware product eventually threatens the applications it connects.

The Orchestration Bet

Perplexity is betting that no single AI provider will dominate every capability, and that enterprises will increasingly demand the ability to route tasks to the best available model for each job. The company's own internal usage data supports this: model distribution shifted from 90% concentration to no single model commanding more than 25% of usage in a year.

If that trend continues—if AI capabilities keep fragmenting across providers rather than consolidating—then the orchestration layer becomes the most valuable piece of the stack. Not the models, but the intelligence that coordinates them.

That's the $20 billion bet. A three-year-old company thinks it can out-orchestrate Microsoft, Google, and the model providers themselves by sitting on top of everyone else's infrastructure.

The last company that successfully pulled off this playbook was Salesforce. They built a layer on top of existing enterprise systems, made themselves indispensable through integrations, and eventually became more essential than many of the systems they connected to.

Whether Perplexity can replicate that trajectory depends on execution, trust, and whether the multi-model orchestration thesis holds. But the timing—entering the enterprise market exactly when agentic AI demand is inflecting upward and models have crossed a reliability threshold—suggests they picked the right moment to make the move.

---

Continue Reading:

• The Real Cost of AI Transformation: Atlassian's $225M Bet
• Why High-ROI AI Teams Work 7x Faster
• When Your AI Investors Are Also Your Customers: OpenAI's $110B Round

---

Have thoughts on multi-model orchestration strategies? Connect with me on LinkedIn, Twitter/X, or Facebook.---

Want to calculate your own AI ROI? Try our AI ROI Calculator — takes 60 seconds and shows projected savings, payback period, and 3-year ROI.

Continue Reading