For two years, AI agents were a demo category. A model would book a flight in a keynote video or write a snippet of code while an audience applauded. The week of November 5, 2026, three launches from very different companies pointed in the same direction: agents are moving out of the chat window and into continuous, background, production work.
What actually launched
Perplexity Teammates, announced November 5, 2026, turns the company's AI assistant into always-on colleagues. Each teammate holds a standing role — triaging bugs, tracking a sales pipeline, researching a market — and keeps working on it for weeks, using Perplexity's existing projects, memory, skills, and connected tools. Teammates communicate with human colleagues through Slack, hand work to one another, and run on Perplexity Computer, the company's system for operating agents on a user's device. Access is limited to individual Max subscribers on the web, at $200 per month (about $167 a month billed annually), as a research preview. Enterprise customers are not included by default.
Cursor Live Apps, shipped the same day, closes a different loop. A Cursor cloud agent now takes a plain-language request and scaffolds a web application, provisions a Postgres database when one is needed, puts sign-in behind Cursor SSO, runs deterministic and agentic security checks, and deploys the result to a shareable link. Nothing goes live until the checks pass and a human explicitly publishes. The feature runs on paid plans that include cloud agents, works from new repositories on Cursor Origin or existing GitHub repositories, and is cloud-agent only — it is not supported on local agents.
Airbnb's AI travel agent is the third data point, and the furthest out. In its third-quarter shareholder letter on November 5, the company committed to begin rolling the agent out to guests in the first half of 2027. Rather than answering in a chat window, the agent will assemble a travel-specific interface in real time — maps, photos, comparisons, calendars, itineraries — and help guests plan, book, and manage trips, including refunds, rebooking, and safety support. Airbnb's chief executive, Brian Chesky, has argued that specialised agents will beat general-purpose ones. The company already reports that its AI support assistant resolves about 40% of customer support issues without a human, up from about a third the previous quarter.
From sessions to standing work
The common thread is persistence. A chatbot answers a prompt and forgets the conversation; an agent with a standing role remembers the project, watches for changes, and acts without being asked. Perplexity's design makes this explicit: you stop writing prompts in a session and give a goal instead. Cursor's design does the same for software: the agent owns the whole path from request to deployed application, including the database and the security gate. Airbnb's agent owns the whole trip, not just the search.
For founders, this changes the product question. The winning demo in 2024 was a clever prompt. The winning product in 2026 is a reliable loop: a defined role, durable memory, permission to act, and a way for humans to check the work. Perplexity gives each teammate its own thread and lets colleagues interact through Slack. Cursor keeps a human on the publish button and runs security checks before anything goes live. Both are admissions that autonomy without oversight does not ship.
Distribution moves into existing workflows
The second lesson is about where agents live. None of these three launches asks users to adopt a new destination. Perplexity meets teams in Slack. Cursor meets developers inside the editor they already use, and deploys to a link they can share. Airbnb meets travellers inside the app where they already book stays. The agent is a layer on an existing workflow, not a new app to open.
That has a direct consequence for founders: distribution for agent products increasingly runs through the tools people already use. An agent that requires users to change where they work faces a much harder adoption curve than one that plugs into Slack, GitHub, an editor, or a booking app. The moat is less often the model — Perplexity's default model for teammates is Claude Opus 5.5, a model it does not own — and more often the integrations, the permissions, and the trust that lets an agent act inside a real workflow.
What founders should do now
First, design for continuous operation, not single turns: give the agent a role, memory, and a trigger, and build the review surface before the autonomy. Second, ship inside a workflow people already inhabit. Third, keep a human in the loop at every irreversible step — publishing, spending, booking, sending. And fourth, decide your pricing model deliberately: seat, metered, or outcome-based. Perplexity's $200-a-month plan with metered credits is the most aggressive test yet, and the market has not settled which model wins.
The agent era is no longer a research preview. It is a production question, and the companies that answer it with reliability, oversight, and distribution will win the next cycle of funding.
Angeled.com connects founders with investors.
Source: siliconreport.com
Source: cellcog.ai
Source: vinish.dev
Source: techcrunch.com
Source: skift.com
Source: fourweekmba.com
Read Next
OpenAI Releases Codex Security as Open-Source CLI and SDK
Databricks Unveils Always-On Genie, an AI Agent That Stops Waiting to Be Asked
DeepMind Launches AlphaGenome 2, a DNA Model That Reads Genomes Cell by Cell
ElevenLabs Unveils 13 New Products and Upgrades at Its New York Summit
Google Launches Gemini Decision 1, a Model That Answers Instead of Writing
Runway Open-Sources Praxis-1, Its Robot Model That Learned by Watching Video



