Podcast Summary: The Last Invention is AI

Episode: Perplexity Max Debuts Multi-AI Agent Tool
Date: March 1, 2026
Host: The Last Invention is AI

Episode Overview

This episode explores the debut of Perplexity Max’s multi-AI agent tool—a cloud-based platform that orchestrates 19 different AI models to independently execute complex workflows. The discussion delves into Perplexity's product evolution, strategic pivot towards specialized professional users, their thoughts on business models, and how their innovations compare with key competitors (OpenAI, Google, Anthropic).

Key Discussion Points and Insights

1. Perplexity Max: Multi-Agent Tool Explained

• Exclusive Access: Available to Perplexity’s top-tier ($200/month) subscribers, targeting enterprise and professional use-cases rather than mass adoption.
  • “So what their highest tier gets access to…is their Perplexity Max plan because it’s also running entirely in the cloud.” [02:00]
• Cloud-Based Security: Emphasizes reduced local device risk, unlike OpenClaw and other local solutions.
  • “they say that it can avoid a lot of security risks that have happened to local device level agents like OpenClaw” [02:15]
• Unified AI System: Markets itself as a tool that “unifies every current AI capability into a single system”.
  • Handles complex, multi-step tasks (data gathering, analysis, visualization), spawning “sub-agents” to tackle subtasks.
• Demo Postponed: Perplexity canceled a live demo after finding last-minute software flaws, highlighting the intense pace and pressure in AI releases.
  • “This is kind of brutal. I mean, it really just shows how breakneck we are right now with a lot of these software.” [04:10]

2. Perplexity’s Market Position and Strategic Evolution

• From Search Engine to AI Orchestrator:
  • Started as a search engine similar to Google AI, now repositioning as a user/Internet intermediary through products like the Comet browser.
  • “One of those is Comet. It’s an entire web browser…positioning themselves as a new layer between users and the Internet.” [07:18]
• “GDP-Moving” Target Audience:
  • Shifts focus away from mass user counts toward professionals making high-impact (economic) decisions.
  • “We’re not actually on a mission to get as many users as possible.” [09:50]
• Business Model Pivot:
  • Previously experimented with ads, but abandoned them in late 2025 due to trust and answer accuracy concerns—opposite of ChatGPT’s new embrace of advertising.
  • “They said that the model risked kind of undermining user trust and answer accuracy. Well, that’s interesting because ChatGPT is now about to jump into this…” [08:15]

3. Technical Uniqueness: Multi-Model Routing & Deliberation

• Dynamic Model Selection:
  • Utilizes a mix of third-party LLMs (like GPT, Gemini, Claude), automatically optimizing for task type, speed, and cost.
  • “Their system right now can automatically choose what model based on cost and performance.” [17:40]
• “Court” Deliberation Feature:
  • Sends a question to all top models, aggregates responses, and has models deliberate for the best answer.
  • “It pitches the question to like all the top AI models…has them all deliberate between each other what the best response is.” [13:35]
• Workflows by Domain:
  • Gemini Flash for visual tasks, Claude Sonnet for code, GPT for medical research.
  • Also integrates modified open-source Chinese LLMs for cost efficiency.
  • “Visual outputs…most frequently handled by Gemini Flash, software engineering tasks were done by Claude Sonnet and medical research was all given to GPT…” [15:00]
• Transparency Critique:
  • Some criticism regarding disclosure of model usage, particularly open-source Chinese LLMs, but Perplexity claims secure and transparent handling.

4. Recent and Upcoming Developments

• Infrastructure Investments:
  • Built its own AI-optimized search index, reducing reliance on third-party APIs (e.g., Google, Bing).
  • “Perplexity also said that they have built their own AI optimized search index…” [12:00]
• New Benchmarks:
  • Introduced “Draco” for complex research, claiming to outperform Gemini, especially in deep research tasks.
  • “They recently introduced a new benchmark for some complex research research tasks. They’re calling it Draco…” [11:50]
• Upcoming Launches:
  • Comet browser coming to iOS.
  • First developer conference slated for March to support third-party API adoption.
  • “Comet browser…slated to arrive on iOS next month…planning on hosting their first developer conference in March…” [21:15]
• Business Metrics Shift:
  • Prioritizes revenue over daily query counts; some community complaints about stricter rate limits on free/paid tiers.

5. Competitive Landscape and Perplexity’s Speed

• Quick to Market:
  • Often beating OpenAI, Google, and Anthropic in releasing groundbreaking features.
  • “They are, it feels like, faster than a lot of the other frontier companies. They beat OpenAI before OpenAI was able to release their Atlas browser…” [24:00]
• Strategic Focus:
  • Moving well beyond search; pursuing being a versatile AI execution engine and Internet intermediary.
  • The host expresses ongoing interest in following Perplexity’s developments.

Notable Quotes & Memorable Moments

• “Perplexity is jumping on the open claw bandwagon, or at least their own flavor of this…a new agent tool that can take 19 different AI models all working together…” — Host [00:15]
• “This is a cloud based computer using agent. It can execute complex workflows independently. It's orchestrating 19 different AI models and it can even spin up what they are saying are sub agents to tackle, you know, specific sub problems.” — Host [02:30]
• “We're not actually on a mission to get as many users as possible.” — Perplexity Executive, quoted by Host [09:50]
• “I think this is just them kind of following through. And I like, I will give Perplexity credit, which is that they are, it feels like they're faster than a lot of the other frontier companies.” — Host [24:00]

Timestamps for Important Segments

• 00:15 – Introduction to Perplexity Max multi-agent tool
• 02:15 – Security and cloud-based orchestration
• 04:10 – Demo cancellation & the pace of AI product cycles
• 07:18 – Perplexity's evolution: from search to browser & enterprise layer
• 08:15 – Advertising model comparison: Perplexity vs. ChatGPT
• 09:50 – Perplexity’s user base strategy & executive quote on audience focus
• 11:50 – Introduction of the “Draco” research benchmark
• 13:35 – “Court” feature: models deliberate best answer
• 15:00 – Domain-specific model routing (visual, code, medical)
• 17:40 – Automated routing and transparency in LLM usage
• 21:15 – Comet browser/iOS and plans for developer conference
• 24:00 – Perplexity’s speed and competitive standing

Tone and Takeaways

The episode adopts an analytical yet enthusiastic tone, acknowledging Perplexity's calculated shift towards a premium, professional market and lauding its technical innovations and speed to market. The host positions Perplexity as an agile, potentially underappreciated competitor in the AI agent race—worth ongoing attention for industry watchers.