- unwind ai
- Posts
- Perplexity Just Launched Their Own OpenClaw
Perplexity Just Launched Their Own OpenClaw
+ Cursor agents got their own computers
Shubham Saboo & Gargi Gupta
February 26, 2026
In partnership with
Today’s top AI Highlights:
& so much more!
Read time: 3 mins
AI Tutorial
Six AI agents run my entire life while I sleep.
Not a demo. Not a weekend project.
A real team that works 24/7, making sure I'm never behind. Research done. Content drafted. Code reviewed. Newsletter ready. By the time I open Telegram in the morning, they've already put in a full shift.
By the end of this, you will understand exactly how to build an autonomous AI agent team that runs while you sleep.
We share hands-on tutorials like this every week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.
Latest Developments
19 AI models, multiple parallel sub-agents, one orchestrator, zero tab-switching — that's Perplexity Computer.
Perplexity Computer is a multi-model agentic system that takes your prompt, decomposes it into tasks, spins up specialized sub-agents, and runs the entire workflow end-to-end fully autonomously.
100s of connectors, persistent memory, files, and web access, all built on top of Perplexity infrastructure.
Claude Opus 4.6 sits at the center as the core reasoning engine, routing subtasks to these models and orchestrating multiple agents.
You describe an outcome, say, "build me an app that tracks live snow conditions across ski resorts", and Computer handles the research, coding, deployment, and ongoing management.
Every task runs inside an isolated sandbox with its own filesystem, browser, and tool integrations, so nothing leaks into your primary environment. The system can run for hours or even months on long-horizon projects, checking in only when it genuinely needs you.
Key Highlights:
Multi-Model Orchestration - Computer dynamically routes each subtask to the best-suited model from its 19-model roster, and you can override assignments or pin sensitive tasks to specific models manually.
Sandboxed Execution - Every task runs in an isolated compute environment with scoped credentials and human checkpoint gates before irreversible actions.
Persistent Memory & Connectors - Computer remembers your past work across sessions and connects to hundreds of external services. Long-running projects stay contextually aware for weeks or months without re-prompting.
Usage-Based Pricing - This is Perplexity's first per-token consumer billing model. Max subscribers get 10,000 monthly credits plus a one-time 20,000-credit launch bonus (valid 30 days), with user-controlled spending caps. Pro and Enterprise rollout is coming soon.
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
Over 30% of the PRs merged internally at Cursor are now created by agents running autonomously inside cloud sandboxes, and that capability is now available to everyone.
Cursor has launched a new version of its cloud agents that each get their own isolated virtual machine with a full development environment, the ability to interact with the software they're building, and the tools to produce artifacts like videos, screenshots, and logs to prove their work.
You can trigger these agents from the web, mobile, desktop app, Slack, or GitHub, and they'll onboard themselves onto your codebase, build features, test changes, and deliver merge-ready PRs.
Internally, Cursor's team has been using these agents for everything from building new features and reproducing security vulnerabilities to running 45-minute UI test walkthroughs of their docs site.
Key Highlights:
Parallel Execution in Isolated VMs - Each cloud agent runs in its own virtual machine, eliminating the resource conflicts that local agents face. You can spin up multiple agents simultaneously on different tasks without them competing for your machine's resources.
Self-Validating Output - Agents don't just write code and hand it off. They build, run, and interact with the software inside their sandbox, iterating until the output is verified, then deliver artifacts (videos, screenshots, logs) to validate what they did.
Access - Cloud agents are accessible from the Cursor desktop app, web, mobile, Slack, and GitHub. You can kick off an agent from a Slack message and get a summary back in the same thread when it's done.
Remote Desktop Control - You can directly control the agent's VM desktop to interact with the modified software, test changes, or make edits yourself — without needing to pull the branch locally.
Quick Bites
Qwen releases open-weight Qwen 3.5 medium models
Alibaba's Qwen team dropped the Qwen 3.5 Medium Series - four models (35B-A3B, 122B-A10B, 27B, and Flash), where the headline act is a 35B model activating just 3B parameters per token that outperforms last gen's 235B flagship. Architecture over scale, plain and simple.
Hybrid Gated DeltaNet + MoE architecture, 1M token context, natively multimodal
27B dense model ties GPT-5 mini on SWE-bench; 122B-A10B dominates tool use benchmarks
Flash API at $0.10/M input tokens with built-in tool calling
All open-weight, Apache 2.0
Claude Code gets a Remote Control
Claude Code now has a Remote Control feature that lets you continue a local coding session from your phone, tablet, or any browser. Just run /remote-control and scan a QR code. Everything still runs on your machine (nothing moves to the cloud), and conversations stay synced across all connected devices. Currently available as a research preview on Pro and Max plans.
Simon Willison's guide on Agentic Engineering Patterns
Simon Willison has started publishing Agentic Engineering Patterns, a growing guide for developers to get the best results out of coding agents like Claude Code and OpenAI Codex. The first two chapters cover why writing code is essentially cheap now and how red/green TDD helps agents produce more reliable output. He plans to add 1-2 chapters a week. And yes, every word is written by him, not an LLM.
Google AI Pro/Ultra subscribers can’t use OpenClaw
Google is cracking down on AI Pro/Ultra subscribers who used third-party tools like OpenClaw to pipe their Antigravity tokens into external apps, effectively turning a $249/mo subscription into unlimited API access. Google says it's protecting service quality from "malicious usage"!
Tools of the Trade
Antigravity Awesome Skills - A curated library of 900+ SKILL.md files that work across Claude Code, Gemini CLI, Codex CLI, Cursor, and other agentic coding tools. Install once via npx and your AI agent gets access to skills spanning architecture, security, DevOps, testing, and more, organized into role-based bundles.
Sandgarden AI (sgai) - CLI tool for goal-driven software development where you define software goals in a GOAL.md file and it orchestrates multiple AI agents via a DAG to build it, with a React web dashboard for real-time monitoring. It uses opencode as the inference backend.
Agent PR Replay - Takes merged PRs from any repo, reverse-engineers the task prompt, runs Claude Code on the same task from the same starting point, and diffs the agent's output against what humans actually shipped. The result is auto-generated CLAUDE.md/AGENTS.md guidance based on where the agent systematically diverges, like over-engineering or missing project conventions.
Awesome LLM Apps - A curated collection of LLM apps with RAG, AI Agents, multi-agent teams, MCP, voice agents, and more. The apps use models from OpenAI, Anthropic, Google, and open-source models like DeepSeek, Qwen, and Llama that you can run locally on your computer.
(Now accepting GitHub sponsorships)
Hot Takes
if openai is microsoft
and anthropic is apple
we deeply need the linux of ai
~ Wes WinderClaude is down.
Productivity across Silicon Valley drops 90%.
I just realized I haven’t typed a single line of code in 2 months...
~ Yuchen Jin
That’s all for today! See you tomorrow with more such AI-filled content.
Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!
PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉
Reply