Start here · Agents Pro ~9 min read Free survey

Perplexity agents: beyond search.

Perplexity started as a 'Google with citations' search engine. That's not what it is anymore. In 2026 it's a full agent platform: Comet is a real AI browser, Spaces are persistent research workspaces, Deep Research produces analyst reports, Model Council compares answers across frontier models, Labs builds working apps from prompts. This free lesson is the map.

The mental model

Perplexity went from search to deliverable.

The old paradigm: search for information, read the results, synthesize in your head. Perplexity v1 was that — but with citations.

The new paradigm: describe a deliverable, get back a deliverable. Comet does it in a browser. Labs builds a finished app. Deep Research produces a finished report. Spaces holds the accumulated knowledge for ongoing projects.

Each agent is a different shape of the same shift: from 'help me find' to 'give me the thing.' This is the practical edge Perplexity has on bigger competitors.

Workflow 01 Comet

1

Comet — the AI-native browser

Comet is a full browser (iOS, Android, Mac, Windows) built around the premise that you usually want to do something with a page, not just read it. Now running on Claude Opus 4.6 underneath.

The prompt that works

What Comet doesComet examples: • While reading any article: "Find 3 sources that disagree with this and summarize their argument" • While on a product page: "Compare this to [competitor] including hidden costs" • While in a form: "Fill this in based on the info I just pasted" • Autonomous: "Find the cheapest flight from MSP to SFO on Friday morning, returning Sunday evening. I'll confirm before booking." A browser that does work, not just displays pages.

Best use cases

  • In-page research without tab-switching
  • Comparison shopping with informed analysis
  • Multi-site autonomous tasks
  • Cross-system actions (email + browser + calendar)
Time savings: Multi-site errands: 1-2 hours → 10-15 min supervision.

Workflow 02 Spaces

2

Spaces — persistent research workspaces

The problem with one-off AI searches: they disappear when you close the tab. Spaces solve it — persistent workspaces that hold sources, custom instructions, and conversation history. Can be shared with a team.

The prompt that works

What Spaces enableUse Spaces for: • Ongoing competitive intelligence ("Competitor X" space, accumulated over months) • Account-specific research (one Space per top-tier customer) • Multi-week research projects (you and teammates contributing) • Industry intelligence shared across product marketing Custom instructions per Space focus the AI: format preferences, source quality requirements, what to skip.

Best use cases

  • Sales-team competitive intelligence
  • Multi-week research projects
  • Team-shared research with accumulated context
  • Reusable research templates
Time savings: Reusing accumulated context: 10-15 min/follow-up search.

Workflow 03 Deep Research + Model Council

3

Deep Research and Model Council — high-stakes answers

Deep Research is the long-running analyst mode, now running on Claude Opus 4.5. Model Council asks the same question to multiple frontier models simultaneously — and shows you where they agree and disagree.

The prompt that works

High-stakes answersDeep Research: "Research the state of vertical AI in legal tech, mid-2026. Produce a 12-slide deck covering top players, traction, pricing, market sizing." Get back a finished presentation. Model Council: "Is this claim accurate based on current information?" Asked to ChatGPT, Claude, Gemini, and others simultaneously. Disagreement is signal — that's when you investigate further.

Best use cases

  • Industry analysis reports (Deep Research)
  • Pre-publication fact-checking (Model Council)
  • Multi-model verification for important claims
  • Strategic decision support with cross-model perspective
Time savings: Real research report: days of analyst work → 30 min.

Workflow 04 Labs

4

Labs — autonomous long-running builds

Labs is Perplexity's most ambitious agent. Describe an app or a deliverable; Labs builds it. Working web tools, presentations, spreadsheets, dashboards, even small sites.

The prompt that works

What Labs buildsA Labs build: "Build me an interactive web tool that lets users paste a resume and a job description, then produces: top 5 skills present in both, top 5 skills missing, suggested resume rewrites, and a 1-10 match score. Clean dark-mode design. Mobile-friendly. Shareable URL." Labs builds it. You get a working app, not a tutorial on how to build one.

Best use cases

  • Internal tools for repetitive tasks
  • Demo prototypes for stakeholder review
  • Comprehensive multi-day research deliverables
  • Customer-facing utilities and calculators
Time savings: Idea → working prototype: weeks of dev → an hour.
The shift

Perplexity 2026 isn't a better search engine. It's an answer-and-deliverable platform — five different agents, each one designed around 'give me the result' instead of 'help me find information.' For research-heavy work, it's quietly become one of the most useful AI products in the market.

Want each Perplexity agent walked through?

Pro members get 12 hands-on Perplexity lessons: Comet browser deep dive, Spaces for collaborative research, Deep Research with deliverable outputs, Model Council for high-stakes claims, Labs for autonomous builds, plus the patterns that get the most out of each one. Founding members lock in $9/mo forever (vs. $19 standard).

Back to Perplexity Mastery → Try a prompt in the playground →

What you can do now

  • You know the five Perplexity agent surfaces (Comet, Spaces, Deep Research, Model Council, Labs)
  • You understand the shift from search to deliverable
  • You've seen concrete examples of each agent's value
  • You can identify which agent would matter most for your work
  • Next step: Pro lessons walk you through each one with real workflows
Pro
Up next in Perplexity Mastery

Lesson 02 · Perplexity basics — the fundamentals

First hands-on Pro lesson: the Perplexity fundamentals — search modes, citation patterns, focus modes, and the prompt patterns that get the most out of the base product. See the track →