Articles by Johnny
-
Secure Sandboxes for Agents
Secure Sandboxes for Agents
Two things are required for agents to execute complex workflows: access to data and the proper infrastructure. While…
16 Comments
Activity
5K followers
-
Johnny Ho shared thisAgentic security is harder than regular security because the same agents that work for you are the same ones that are likely to find workarounds that bypass your controls. Stay safe by adding deterministic detection systems like Numbat around your infrastructure. Numbat is built into each Perplexity's devboxes by default, and coming soon to Perplexity Computer as a product as well.Johnny Ho shared thisToday we’re open-sourcing Numbat, an agent-detection and response layer that is designed to work across agent harnesses. It gives security teams visibility into agent activity, with controls to block selected actions before execution. Numbat works across agent harnesses on desktop, CLI, IDE, and gateway agents. It supports live monitoring, local detection, pre-action blocking, and forensic reconstruction from existing session artifacts. Numbat runs across our systems and sends audit events, findings, and security alerts to Computer. Computer continuously analyzes this telemetry, escalates suspicious behavior, and proposes new on-device detections for Numbat, creating a continuously improving detection flywheel. Numbat is available as a single Go binary for macOS, Linux, and Windows. Read more: https://lnkd.in/gHuVv3uu Available under Apache 2.0: https://lnkd.in/g-XbWrUw
-
Johnny Ho shared thisExcited to announce Projects, which are Perplexity's hubs for agent and human collaboration. Most of my time now lives in Projects, where agents automatically manage their own persistent files and collaborate with native tools. Projects are Perplexity's core for collaboration: as agents become more long-running, they will need more agent-native ways to persist and self-improve: memories, skills, and files. But until today, agents have lacked a central hub for collaboration that allows for a persistent area for knowledge work. That place is Projects, not individual chats or files. Projects come packaged with an set of CLI tools intentionally designed for agents to collaborate across files, projects, and with other agents. Importantly, both individual and project identities are fully functional. Individually-scoped connectors are supported as usual, but so are project-level credentials. A project-level brain agent manages the project and provides objective summaries and memories. Agents do the execution but humans still do the steering with notifications and permissions checks. Internally at Perplexity, the canonical way to work is now to start in Slack (each Project connects with at least one Slack channel), spawn multiple threads of conversation in parallel, accept most first passes, and follow up in Perplexity's apps when refinement is needed. The future of knowledge work is looking bright: work is incredibly efficient with tools like Computer, leaving more time for curiosity and learning. Projects is the next step in the vision that Spaces began for collaboration and persistent storage of context. We're excited to see how you use Projects and we will learn from your feedback to keep pushing the frontier of what is possible when agents and humans work together to build real businesses.Johnny Ho shared thisToday we're launching Projects, an evolution of Spaces. Projects are hubs for ongoing work in Computer. A single place to manage, create, and collaborate on tasks with a shared file system and persistent memory. Each Project has a shared file system. Computer reads, edits, and saves Project files across sessions, so the work builds on itself with each task. When collaborating with teammates in a shared Project, everyone works from the same source of truth. Projects connect to Brain, Computer's self-improving memory system. Between tasks, Brain reviews the Project's files and sessions and updates what it knows, so each task starts with full context from previous work. Projects are available today to all users. Read more: https://lnkd.in/gPRgtkr3
-
Johnny Ho shared thisExcited to be working with Microsoft to bring Computer to all Windows devices. Perplexity Computer on Windows (https://lnkd.in/g6hxVP3j) allows for secure local work and agentic compatibility with the entire Microsoft ecosystem. Thanks also to Shaloo for shouting out the individuals defining the frontier of possibilities here. More coming soon in terms of platforms and capabilities.Johnny Ho shared thisWe are thrilled to share that Personal Computer from Perplexity is now available on Microsoft Windows. With Windows powering more than a billion devices worldwide, users can now run Personal Computer directly on Windows, bringing Perplexity's AI agent platform into the local Windows environment for a seamless, powerful experience. Users can ask Computer to work with their local files, Microsoft 365, and the web in one place. Huge congrats to the amazing Perplexity team- Aravind Srinivas, Johnny Ho, Aleksei Kasharin, Alena Tcareva, Brady Murphy, Fargat Sharipov, Jess Hoffman, Joey Z., Leonid Persiantsev, Matt Kula, Miko Mynttinen, Nik Izrailev, Nikita Podpruzhnikov, Nikolai Krasilnikov, Pawel Stanek, Roman Karasev, Sergei Chusovskoi, Shelwin Sunga, Thom E. & team and the phenomenal Microsoft team- Lydia Stoykovych, Patrick Moulton, Ricardo Martins, James Jarrett, Mike Gaal, Tyler Hagens, Ravi Sudhakar Kambhampati, Tony Imperati, Alex Staddon-Smith , Rajat Gupta & team on this incredible milestone! Click here to learn more: https://lnkd.in/d-S8Q7QH #microsoft #AINative
-
Johnny Ho shared thisFactual accuracy and traceability are non-negotiable values at Perplexity. Citations are built into Perplexity by default. Perplexity Finance tools are built into Computer by default. Computer displays rich citations for every ticker and a full derivation of the calculation.Johnny Ho shared thisQuality of life improvement: Perplexity Computer now displays company logos when financial data is cited in an answer. All data is traceable back to original documents like SEC filings and earnings call transcripts. Derived data is shown with a full chain of formulas and calculations applied.
-
Johnny Ho shared thisPerplexity continues to hire new grads and residents even as AI accelerates. In our experience, early career engineers and associate product managers are often faster at learning new stacks like agents and evals. Check out First Token 2026 if you are interested or know someone who is early in their career. Also check out www.perplexity.ai/careers to see all roles.Johnny Ho shared thisExcited to speak next Thursday (July 30) at First Token 2026! First Token is Perplexity's flagship event for early-career talent. We believe that exceptionally talented individuals at any stage can thrive in our high-velocity, high-ownership culture. Each year, we select a small and elite cohort of new grads to join our team (we're at an even dozen for 2026!). These individuals enjoy unparalleled breadth and depth of technical challenges, along with a rich network of best-in-the-industry mentors and the opportunity to quickly grow into senior technical and product roles. We have a limited number of invites to extend for students in the San Francisco Bay Area this summer. If you're an intern, researcher, or anyone else interested in solving hard problems in 2026's most interesting technical domains, we'd like to see you at First Token next Thursday. There's even an opportunity to jump straight to the final rounds of our hiring process (click through to the event page for details 🙂). Apply to attend at https://luma.com/e1b05fm1First Token 2026: Early Career Talent @ Perplexity · LumaFirst Token 2026: Early Career Talent @ Perplexity · Luma
-
Johnny Ho shared thisWe’re introducing SPACE, the sandbox technology behind Perplexity Computer. Agents are only as good as the environment they run in. Their workflows are long-lived and stateful. They need a durable infrastructure that is maximally efficient and secure to achieve their full potential. We took SPACE from concept to product in 10 weeks. We’ve been using it internally for two months. Today, it’s live in Computer for all users. Soon, it will be available in Perplexity's API Platform as well.
-
Johnny Ho shared thisWe’re giving back to the open source and open weights communities by open sourcing our deep and wide research benchmarks. We’ve seen a lot of value in evaluating our products on open benchmarks and hope every company can do the same. The challenge with today’s models is that performance is highly dependent on the context and harness. Every organization deserves to know their own end to end performance rather than overindexing on self-reported benchmarks. Hope this work encourages and helps others build their own benchmarks and evals as well.Johnny Ho shared thisWe’re open sourcing WANDR. WANDR is an internal benchmark we built and used for building deep and wide research capabilities inside Perplexity Computer. https://lnkd.in/gfCh95F3 Wide-and-deep research requires two capabilities. 1) Agents must search broadly enough to find all qualifying entities 2) Agents must investigate deep enough to support every claim with evidence. WANDR represents these requirements as hierarchical, independently verifiable records. It consists of 500 research tasks that require 170,495 source-backed records across three tiers of difficulty. WANDR tests how well agents discover large sets of entities and verify specific facts about each one. It provides a dense, interpretable eval signal that reveals whether an agent fails, and where. The pipeline also doubles as a semi-automated factory for training data. WANDR is constructed from de-identified production use cases covering people’s day-to-day research tasks like competitive research, due diligence, literature review, market analysis, product comparison, talent sourcing, and more. Instead of grading against a gold solution, WANDR re-fetches every cited page and checks each claim against the underlying evidence. This allows certain tasks to contain time-varying facts. Soft scores give partial credit. Hard scores require every component for a member to be complete and correct. The benchmark tasks and evaluation harness are available at: https://lnkd.in/gvaYBDdG
-
Johnny Ho shared thisAt Perplexity, over the last month, we’ve been focused on a quality sprint, polishing out small papercuts. One heavily requested feature now available: account switching. On Mac and mobile, you can now add multiple accounts and switch between them in a single click. This makes it easy to keep your work and personal environments separate without repeatedly authing. Try it under Account → Add account. Web support is coming soon. Prosumer products like Perplexity straddle the gap between personal and professional use. Though Computer was originally designed for professionals, we've also seen unexpected surge of usage of Computer for personal use cases. We've seen demand for personal use cases like personal finance (https://lnkd.in/gPsBywDp), personal health, shopping, travel, and personal side projects grow rapidly, especially while on the go. Now you can keep work and personal isolated but still only one click away. Check out the full changelog at https://lnkd.in/gdiv-3Qh
-
Johnny Ho posted thisIt's undeniable that LLMs and agents have completely transformed the software engineering role. At Perplexity, however, we've seen that other roles have changed just as quickly as the engineering function. Interacting with code and engineering tools is quickly becoming mandatory for all roles at Perplexity, especially product, UXR, and design. A simple prompt like "research the codebase for the latest" improves the accuracy of answers in a way that was not previously possible without manual work. "What is the project status" is now a fully automated question with a single ground truth answer. Underneath the hood, agents like Computer power tasks like research, context gathering, and execution all through code and engineering tools. As projects and experiments spin up more quickly than ever before, workplace agents (like Computer's Brain agent) now own the organizational ground truth (shared memories and repos) and teach humans, rather than humans owning documents and manually onboarding coworkers. What previously were considered engineering tools, like git and developer environments, are becoming general work tools. CLIs are now universal tools for all knowledge workers. Computer's Search as Code architecture even equates knowledge work tasks (search and research) with coding. When anyone can do both with products like Computer, what is the divide between working and coding? Is code owned by a person or a team, or is code owned by agents that work across organizations? These are difficult questions as products evolve over the next year. More to come in a series of posts about how specific roles have changed at Perplexity. Stay tuned.
Johnny Ho replied to a comment
1mo
Perplexity Computer now agentically writes files to the project and automatically scans through them to generate memories.
-
Johnny Ho liked thisJohnny Ho liked thisDecagon is powering customer support for some of the biggest brands in the industry: Delta Airlines, Ticketmaster, Deutsche Telekom, American Airlines. A lot of support related questions require accurate online information and we’re happy to power that with Perplexity’s search. More: https://t.co/sw0hx7QHPj
-
Johnny Ho liked thisJohnny Ho liked thisWe are doubling down on making Perplexity Computer more useful for financial researchers and analysts. Computer now connects to new licensed data sources, including Dun & Bradstreet, Guidepoint, IBISWorld, and 20+ others. An analyst can ask Computer a question and get an answer built from the firm's own licensed sources without APIs or separate logins to work through. Every figure traces to the source record it came from. Teams doing high-stakes work would know exactly which data set produced which number. To connect, select a provider in Settings → Connectors and sign in with an existing license. Available now to all Perplexity users. https://lnkd.in/gVSR-Z4k
-
Johnny Ho liked thisJohnny Ho liked thisOpenrouter just published independent benchmarks of web search providers. Perplexity's Search API topped the charts across the board on all axes -- quality, speed, and price. Why? Part of it is the index. 400B+ unique URLs and growing. But the real reason is that the same API serves hundreds of millions of queries a day inside our own products, Perplexity Computer and Perplexity Ask. If search quality is bad, we're the first to hear about it from our users. If search is token-inefficient, our margins suffer. If search is slow, our product deteriorates. TLDR: There's a lot of noise in the agents space. Buy primitives from people with skin in the game. Full results: openrouter.ai/benchmarks
-
Johnny Ho liked thisJohnny Ho liked thisNot going to pretend I'm not happy about this one. OpenRouter benchmarks search providers independently - hard, multi-step "find the thing that isn't on page one" queries. Perplexity ranked #1 on quality in all four evals, including 89% on BrowseComp. Proud of the team behind this. We've spent the last year rebuilding the search serving stack, and results like this are why. https://lnkd.in/dZM-fika
-
Johnny Ho liked thisJohnny Ho liked thisI am thrilled to share that I joined Perplexity! I'm excited to be joining at a moment when AI is moving beyond answering questions to helping people accomplish more. That's what drew me to Perplexity, the belief that curiosity is an edge and that AI should help turn ideas into action. Perplexity does this by bringing together the best AI models, so every breakthrough in AI makes the product better for users. Excited for what’s ahead and to build alongside such a talented team.
-
Johnny Ho liked thisJohnny Ho liked this📈 We're expanding Growth at Perplexity. 📈 I'm hiring across design, fullstack, and mobile engineering to own Perplexity's user journey — from first install through upgrade to experiencing Computer. 🖥️ The frontier of AI is moving from models to orchestration, and Computer is the sharpest product bet on that thesis. Millions of people use Perplexity to turn knowledge into action. Growth engineering builds the loops that get them there: onboarding, lifecycle, upgrade experiences, in-task upsells, notifications, and the applied AI that makes each of those smarter. Every experiment ties back to whether users are activating, upgrading, and getting real token value out of their subscriptions and Computer credits. We ship many experiments in parallel, own outcomes end-to-end, and create durable surface areas and flywheels inside the core product — not one-off tests. Come work with and learn from an incredibly self-sufficient and talented team across SF and NYC. ➡️ Software Engineer, Computer Growth — full-stack growth across activation, lifecycle, monetization, and applied AI that powers them. https://lnkd.in/gSiQrKfz ➡️ iOS Engineer, Computer Growth — own the iOS funnel end-to-end. https://lnkd.in/gJCTa8Pd ➡️ Android Engineer, Computer Growth — same end-to-end ownership, Android side. https://lnkd.in/gSiQrKfz ➡️ Design Engineer, Growth & Marketing — turn perplexity dot com into a real growth engine. Product launches, brand system, A/B tests. Reach out to our lead design engineer, Zeyao L., with your portfolio. https://lnkd.in/gttiZN9Q #hiring #perplexity #aiorchestration #AIJobs #growthengineering
-
Johnny Ho liked thisJohnny Ho liked thisA few weeks ago we hosted another Greylock Partners casino night cruise, this time in New York, bringing together interns from across AI and quant finance. This one was especially meaningful to me. Every summer during college, I interned in New York in quantitative finance. I remember how formative those internships were, not just because of the work, but because of the people I met. It's fun to now help create those kinds of moments for the next generation of students. In attendance were interns from companies including Cognition, Modal, Ramp, Jane Street, Hudson River Trading, Citadel, and Two Sigma. Seth Rosenberg and I want to thank everyone who came!
Experience
Education
Honors & Awards
-
ACM ICPC (International Collegiate Programming Contest)
-
Gold medal (3rd place)
-
International Olympiad in Informatics
-
1st place 2012 (perfect score)
Gold medal 2011, 2012
Explore more posts
-
Will Stewart
Northflank is partnering again with Cerebral Valley, PyTorch, CoreWeave (and more) for the OpenEnv Hackathon in San Francisco, March 7–8. Participants will be building RL environments and post-training base models across 5 themes, solving the most pressing problems in RL and agentic orchestration. We'll be supporting teams in deploying their models through the Northflank platform with compute from our wonderful partners, CoreWeave (Matthew Lu, Jacob Feldman). $100K+ prize pool and and a stacked lineup of judges and mentors from Meta, Hugging Face, University of California, Berkeley, Cursor, Scale AI, and more. You can submit an application at the link in comments.
3 Comments
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content