Presenting Sai World Where frontier computer-use technology meets the very human desire for breezier work (And to save tokens - our favorite feature of Sai)

Our head of research @xwang_lk joined Barry Barish, Terence Tao and other scientific visionaries at the @SAIRfoundation Science x AI Summit. Great conversations on open research, AI safety, agent market signals. And we're cooking something 🧑‍🍳

We are throwing a party during #SFTechWeek to nerd out on all things CUA 🖥️ ChatGPT Astra just pushed computer use agents back into the spotlight. But what needs to happen before CUA become ubiquitous? Simular is gathering folks who built agents at OpenAI and DeepMind, alongside a select group of researchers, investors and founders to debate the biggest questions in CUA today: reasoning, reliability, autonomy, economics. Could 2027 be the year computer-use agents reach AGI? Join us to find out at SF a16z @techweek_! partiful.com/e/YUKBbGuv0GuGM…

Task #28: Vaccine booking This involves getting an email about required immunizations, a scanned vaccination record on the desktop, and a booking site with price, distance and date constraints. The requirements also state a Hepatitis B dose-spacing rule: dose 3 must come at least 16 weeks after dose 1 and at least 8 weeks after dose 2. The record shows three doses given 70 and 40 days apart, so a fourth booking is due. @sai_borg scored 1.0, Opus 4.7 scored 0.4, and Sol scored 0.0.

We challenged @sai_borg vs Opus vs GPT on tasks that we (humans) actually do in the real world. Task #68: The agents play Chrome Dino, a repetitive real-time game, and clear a score target on a live page. Scoring is pass/fail on the final score, and all three agents scored 1.0, so what actually separates them is the cost of getting there. Sai finished in 23 model turns, Sol needed 46, and Opus 4.7 took 201.

Our computer-use agent @sai_borg just beat Opus 5 and GPT-5.6 Sol on OSWorld 2.0. Sai scored 73% on the CUA benchmark where each complex task takes a skilled human over an hour to execute. And Sai did it at about 2/3 the cost per task of Opus and GPT 💸 The unlock is our neurosymbolic framework that pairs the exploratory power of neural networks with the logic of symbolic code, resulting in reliability and cost saving. Read the full report: simular.ai/articles/sai-tops…

Since our amazing intern published her open source CUA-on-smart-glasses project 'Sai-Fi', we've received lots of questions. So Jamie wrote this post to answer all the things you want to know about putting computer-use agents onto wearables. Read the full low down here: simular.ai/articles/why-type…

Introducing Sai-Fi: an open source app that lets you talk to AI agent Sai through a pair of Meta Ray-Ban glasses. Sai is an autonomous computer-use agent. It can see the screen, move the mouse, type, and keep working even when you are away. We used Meta Ray-Bans for the first prototype but the glasses aren't really the point. Underneath is the Sai API and a voice concierge designed to work across different devices. The interface could live in an earpiece, a car, a small box on a workbench, or something nobody has built yet. Give it a microphone, a speaker, and a connection, and Sai has another way in. The bigger idea is an autonomous computer you can reach from anywhere. Bring @sai_borg to your own @Meta @ray_ban glasses today!

We are one more step closer to freeing you from the computer. Guess what we're brewing?

Something is cooking 🍳 Last night: dinner, small room, 90 minutes, everyone left having handed their most manual lead gen + outreach workflow to @sai_borg -- the infinitely patient autonomous computer that... reasons, sees, clicks, and move on the screen for you. Always on. No yearning. Pictured: our team sharing with the group their secret Sai workflows 👀 Thanks everyone for coming out!

The most interesting AI conversation doesn't just happen on stage. So we wanted to make something fun. < AI over Kaya Toast > Grateful to @airwallex, Openspace Capital , and the Claude Singapore Community for co-hosting and to everyone who joined us. Follow for future event updates 🇸🇬 #AGIPlayground #Singapore #simular

What does it actually take to cover the last mile to AGI? Our CEO @angli_ai joined @bernardleong on @analyseasia — sharing his 9 years of research into agent reliability, and what autonomous computers can really handle today. Full episode coming soon.

Simular turns three. Three years ago the bet was simple and slightly out of step with the industry: while everyone built chatbots or waited for software vendors to expose an API, we believed general intelligence had to operate a real computer — see the screen, click, type, navigate any GUI. We ran it as a research-first lab and here are some milestones looking back: 1️⃣ Oct 2024 — Agent S, the first open-source CUA. 20.58% on OSWorld. 2️⃣ Mar 2025 — Agent S2. 34.5% on OSWorld, 50% on AndroidWorld.  3️⃣ Dec 2025 — Agent S3. 72.6% on OSWorld, past the 72.36% human baseline. Then the greater part: turning it into something people can actually afford to run. @sai_borg launched in March, and our neurosymbolic work has brought token usage down to 1/100 of the industry average. Today we're 30 people across Palo Alto and Singapore, shipping @sai_borg weekly. Thank you to everyone who's been in the room for the last 36 months. 🎂

Our CEO @angli_ai at AGI Playground Singapore 🇸🇬 @GardensbytheBay He brought a fun metaphor on stage: imagine sending your agent to the gym. You don't go to the gym to learn anything new. You do the same boring movement again and again until you can do it without thinking about it. That's the part missing from how we build agents.  What's missing in agents isn't intelligence but reliability. So we send our @sai_borg to do reps 🏋️. It practices a task in a cloud sandbox until it finds a route that works, and we save that route as code. From then on the code does the job, and the model only gets called if the code breaks. That's muscle memory.  Thank you @GeekParkHQ for the stage.

Our open source Agent S3 has been accepted to the Transactions on Machine Learning Research @TmlrOrg 🎉 Computer-use agents (CUAs) hold promise for automating everyday digital tasks, but their performance on long-horizon, complex problems remains unreliable. In this paper, we introduce a way to compress agent iterations into behavioral summaries, and created a judge framework that utilizes summaries to correctly choose the successful one over many repeated workflows. Agent S3 was the first to surpass human-level performance on OSWorld, achieving a 72.6% success rate, in December 2025. Congrats again to our research team! Read the paper here: openreview.net/challenge?red…

AGI Playground 2026 Fringe Event: Kopi, Kaya & Computer Come taste your way through a Singaporean breakfast with @SimularAI — while we talk frontier AI. 🗓 Aug 4 · 9:30–11:00AM 📍 CBD, Singapore ☕ Kopi + kaya toast tasting With our friends @airwallex, Openspace Capital, and the Claude SG community. Come for the toast, stay for the agents. RSVP 👇 luma.com/simular-g3u8

What happens when you put ambitious GTM builders in one room and give them a few hours to reinvent how companies grow? This Wednesday, we partnered with  @ColtonKaplan and @PilotHQ to host a GTM Hackathon - and the creativity in the room was incredible. We were genuinely surprised by how teams used @sai_borg to turn ambitious ideas into working GTM products. They built systems that: → Ranked event attendees against a company’s ICP and generated personalized conversation openers → Detected buying signals across G2, Reddit, LinkedIn, Luma, and job boards → Turned QR-code scans into interactive experiences that captured leads after delivering value → Generated, launched, measured, and improved paid acquisition campaigns → Measured a company’s visibility across AI answer engines and identified opportunities to improve it → Converted a company website into a full week of on-brand content, including posts, carousels, newsletters, and videos → Automated prospect research, CRM workflows, outreach, follow-ups, and next-step recommendations The strongest projects shared one important characteristic: closed the loop between identifying a signal, deciding what to do, and taking action.  That is where we believe computer-use agents—and our flagship product, Sai — can fundamentally transform GTM, business development, and sales by navigating authenticated websites, gathering context across applications, operating existing tools, and turning ideas into live, usable workflows. Congratulations to Team SignalScout and PrismCRM for winning the prizes! 🎉 Thanks to everyone else who participated. Until next time!

Who knew Sai could be an installation at an art gallery? Last Monday, we tried something different. Simular partnered with @MakePortals and @withpersona   to build an interactive art gallery for AI agents, where every exhibit was a real, preconfigured software environment. Most AI demos are designed to show the perfect run. Ours let visitors take control and run it live. Our exhibit was a lead follow-up workflow: Sai reviewed Slack messages, pulled out relevant details on potential leads, drafted and sent personalized emails, then turned the whole thing into a daily routine. But the best part wasn't the workflow we prepared. It was watching people go off-script. They opened different conversations, changed the context, tested unexpected actions, and pushed Sai through edge cases on their own. That is where a computer-use agent proves itself. Sai isn't built to run one scripted demo. It understands context, adapts when the situation changes, and gets work done across the tools people already use. Try Sai at sai.work.

What does a normal knowledge-worker day look like in 2030? Clips from the AIIS × Simular fireside chat in Seoul with Prof. Jonghyun Choi @SeoulNatlUni, @angli_ai (@SimularAI) and Dong Keun Jo @youtubejocoding If you’re building or using AI agents, this one’s worth a watch. #AI #AIAgents #FutureOfWork #Simular #AIIS #Seoul

"Intelligence needs a body." Our CEO @angli_ai spoken with Michelle Ye Eun J. from @Koreanherald in Seoul during the #ICML week to talk about why Korea might be one of the most important markets for the next phase of AI — not chatbots, but agents that actually operate a computer on your behalf. The core takeaway : the industry has spent years optimizing GPUs for intelligence. But computer-use agents expose a different bottleneck: CPU, memory, and the hardware that lets an agent act. That's exactly where Korea's strength in semiconductors, devices, and manufacturing comes in. Smartphones, PCs, appliances, cars are not just endpoints anymore. They're where the agent gets to work.

Behind the scenes 🎬 something's coming. This week, @lihautan, Member of Technical Staff at @SimularAI sat down with Chong Rui from @NUSingapore NUS Production Club.  The conversation went deeper than we expected, especially the part about the last mile to AGI. full episode dropping soon 🔜 #Simular #Sai #ComputerUseAgents #AIAgents #BehindTheScenes #AI #ComingSoon

Last week, @SimularAI flew in seoul for @icmlconf and showed up for events that we're really proud of: 🎤 Computer-Use Agents Researcher Roundtable — we gathered a room of researchers at the frontier of CUAs to talk about trajectory-level benchmarks, reward/judge models for RL, uncertainty, and a compiler-layer architecture showing 10–15x efficiency gains. 🎓 @SeoulNatlUni AIIS × Simular Research Forum — we shared about our Agent S3, trust-by-design, and taking agents from research to product.  Thank you to everyone who joined the conversation and pushed our thinking further. More to come! Our AIIS Fireside highlight is on the way. Stay tuned. #ICML2026 #Seoul #AI #ComputerUseAgents #Simular #AIagents

Wrapped our @SeoulNatlUni AIIS x @SimularAI Computer-Use Agents Research Forum in Seoul this morning 🇰🇷 #ICML2026 And a fireside that didn't hold back 🔥 thank you @ritacyliao moderated this session with Prof. Jonghyun Choi (SNU), @angli_ai, and @youtubejocoding on what AI actually does to human work. Thank you SNU AIIS for having us 🙏  #ComputerUseAgents #AI

Great memories made in #ICML2026 @SimularAI hosted a Computer-Use Agents Researcher Roundtable, gathered a room of researchers publishing at the frontier of CUAs.  A warm evening of real research: trajectory-level benchmarks, reward and judge models for RL, uncertainty calibration that teaches an agent when to defer, and a compiler-layer architecture already showing 10–15x efficiency gains. IF YOU FOMO, we still have one more event coming this Friday 😌 #ICML

From research paper to real product, how do computer-use agents actually get built, evaluated, and put to work? @SimularAI will be co-hosting a research forum with @SeoulNatlUni 's AI Institute this FRIDAY, giving talks on Agent S3, trust-by-design, and taking agents from research into real workflows. #ICML2026 JOIN US! Register below to save your seat. 👇

Guess how we shipped our #ICML2026 researcher roundtable event so perfectly 🫡 credit goes to our @sai_borg

LAST CALL: a couple of speaker slots just opened on tonight's Computer-Use Agents Researcher Roundtable, near COEX. If you're deep in this space, apply through the Luma link. 🇰🇷 #ICML2026

Moved to Singapore knowing no one. couldn't remember which uni friends even lived here. @sai_borg surfaced all of them + wrote all my coffee chat invites dormant connections reunion tour, activated. Thank you Sai~ ✨ (we're dropping Sai codes over at @sai_borg this week if you want your own reconnecting era 🎟️)

The Q&A after our CEO @angli_ai's talk at AI Engineer World's Fair in SF ran long. If you've been wondering how computer-use agents actually handle reliability, guardrails, and legacy software, this post has you covered. 1. On stateless vs. stateful programming The neurosymbolic approach is "stateful programming" - every line of code checks the environment/state as it runs. This is a new, more stable programming paradigm for the future. No single modality (screenshots, accessibility tree, DOM) is fully reliable on its own, so they combine multiple signals, like a self-driving car using camera, radar, and LiDAR together. 2. On agent accuracy 100% accuracy is the goal but not guaranteed by any single model. Our solution is a self-verification loop: after the agent completes an action (e.g., filling a healthcare form), a separate check asks the model to "reverse" and verify whether the output makes sense. If something's wrong, the task is retried. 3. On guardrails and security On the product side, irreversible actions trigger a stop-and-approve step from a separate monitoring system. For OS-level restrictions (e.g., blocking specific apps), that requires control at the VM/operating-system layer, which is why we favor cloud-managed VMs, where whitelists and stricter controls can be built into the machine image itself. Same neurosymbolic principle: code-based rules handle it first; if rules don't cover a case, it escalates to the agent, and if still unresolved, to a human for verification. 4. On code vs. vision dependency Token/vision usage should trend toward zero over repeated runs, as the code accumulates more branching logic to handle known scenarios, similar to how humans need less conscious effort as they gain experience with a task. 5. On legacy software prompting Currently prompts do need to be fairly specific for unfamiliar software. Long-term direction is "exploration": let the agent freely explore an environment (when exploration is low-risk/low-cost) to build familiarity, reducing the need for precise prompts over time. The goal is just finding one successful trajectory - once found, it can be replayed reliably. Using a "swarm" of agents to continuously re-explore and repair the map/trajectory when the underlying software changes is the right direction. But this only works best when exploration is cheap (low-risk environments); in costly/sensitive contexts (e.g., banking) free exploration isn't feasible.

Win a Sai code 🎟️ For the next 10 days, we're dropping Sai access codes through @sai_borg. Every day we pick 20 people with the best answer to that day's theme. Tell us what you actually do alongside your answer to increase your odds. Winners get a code by DM (single-use, 72h) - follow + turn on notifications 🔔 #saicoded #trysai

🌱 @sai_borg runs on Windows 365 for Agents — bringing computer-use AI to legacy enterprise environments with no APIs needed. As an early-adopter pilot partner with @msPartner, we've been building toward this moment. See it in action ↓ Read more on simular.ai/articles/simular-…

As a solo hacker at Cal Hacks, @vivianbuilds found a team -- Sai's autonomous computers. She built STING, an anti-scam browser guard in 24 hours while still hitting all the workshops. How @sai_borg helped: ✅ Visual QA with roleplaying - tested STING as an absent-minded elderly person, a brain-rotted Gen-Z user, and everything in between ✅ Demo video production - created a walkthrough with the correct UI and product features (no hallucination). Shows how STING gives plain-English scam explanations instead of vague red warning screens. 🏆 Simular Pick at @CalHacks

We’re proud to support BUILD2026 Hackathon, alongside @Microsoft, @ExaAILabs, @IMDAsg, and friends. 12 hours. Saturday 27 June, 9am–9pm. Come solo or with a team of up to 4. What’s on the table: — $5,000+ in cash and credit prizes — $300+ builder credits per person — Technical and design workshops from people who’ve shipped — Food, and early-bird swag worth grabbing One thesis: build something people want. Sign up: luma.com/zda76sc1

When an 85-year-old says "play some jazz," should a nurse stop critical care to find the Spotify button? Tendly says no. And Simular makes it possible. Patient says: "play my favorite song" Claude triages urgency Safe, non-medical tasks → Simulang automation scripts execute autonomously The Magic: Simulang's computer automation = zero caregiver intervention for simple tasks This matters because: Caregivers focus on actual medical care Patients get instant responses to simple needs 🏆 Simular Pick @CalHacks

At @CalHacks 2026, we are thrilled to celebrate two exceptional projects that showcase the power of computer use agents (CUAs) and their creative, real-world applications of Simular's tech. 🥇 STING - A platform that detects scams early on, explains them in plain language, and provides evidence—going beyond simple flagging. STING showcases a novel use of Sai by simulating diverse user personas (elderly, Gen-Z, etc.) and leveraging @sai_borg's perception and computer control capabilities for product QA like website navigation, and optimization. The (one-person) team also used Sai to create their demo video. Congrats @vivianbuilds! 🥇 Tendly – An app that listens and reacts to elderly patients' needs, addressing a critical societal challenge: seniors often struggle to clearly communicate their needs while caregiver was resources remain scarce. Tendly generates TypeScript Simulang scripts and executes them directly on the desktop via Simular — no caregiver required. Congrats Caden Lewis, Henry Bonomolo, Seth Rojas, and Deven Mittal! Stay tuned for the demos. Thanks to everyone else who participated. Until next time!

"AI can't handle my LinkedIn networking.' That's what a sales professional told us during NYC Tech Week. "I'd love AI to arrange my connection requests and networking opportunities. But I doubt AI will be able to do that." Turns out @sai_borg can. LinkedIn hates bots, but Sai isn't one. Sai is a computer-use agent that works exactly like a human would (reads profiles, personalizes messages, finds the right people). Nothing happens without user approval and it knows when to ask for human control. AI doesn't replace the relationship building. It just handles the grunt work of getting us there.

How many hours a day do you spend moving your cursor? @lihautan walked the room through what changes when an AI agent uses your computer the way you do: clicking, typing, navigating real software - not just calling APIs. Great crowd, great questions. Thanks for having us, #AIContext 🙏

Hackers of @CalHacks, get a chance to win $500/teammate! We are excited to be part of Cal Hacks at UC Berkeley June 20–21. Build with our computer-use agent (CUA) @sai_borg or our scripting language Simulang - every hacker gets UNLIMITED credits for 2 days. 🏆 Rules for participation: sai.work/prize 🎯 Join our workshop "Vibe Hacking" for tips & hacks 📍 Find us at the booth. Founded in 2023 by former Google DeepMind researchers, Simular is backed by investors including Felicis, NVentures (Nvidia), South Park Commons, Basic Set, Lenny Rachitsky and more. Go Bears! #CalHacks #Berkeley #AI

You are wasting your tokens — and you don't have to. Your agent pays full price every time. No learning. Same task, same amount of token spent. @sai_borg learns. Run it once → cheaper every run after. More reliable + more efficient + more affordable. 🌱 Watch our CEO @angli_ai explains ⬇️

If you missed @angli_ai on panel — here's what mattered: Many agents went viral. But no enterprise can actually deploy them. No guardrails. No audit logs. No security review path. That's the opening. 👀

My @sai_borg edited the slides. I didn't open my laptop. Here's what makes this different: 👉 AI slide tools have always had one flaw — you can't really edit the output. You're stuck prompting back and forth forever. ✅ Sai generates the deck and uploads it directly to your Drive. Every element is yours to edit.

It's 10am on Friday and @ZeningChen42844 has checked out already... watching @sai_borg work for him. Plus, Sai is getting a parallel view of agents in action and Pixel art workspace 👾

Two days at @superai_conf, 10,000+ in the room, one Main Stage moment we won't forget Thank you Singapore🇸🇬

Agent Hour 🇸🇬 — done. A room of founders, engineers, and operators packed into @WeWork for the realest fireside of Singapore AI Week: Agents, Honestly — what's working, what's hyped, what's next. @angli_ai · @lionelsimai · Jay Teo — moderated by Freddy Lim. The gap between demo works and customer pays got real. So did the room. Huge thanks to our co-hosts Claude SG · The AI Capitol · First Prompt · @buildclub_ · @AgoraIO — and to @sos_266, @AnnieLiao_2000 & team for pulling it off.

We're attending @superai_conf this week! 🇸🇬 Find us at Kiosk Kl6, see @sai_borg do real computer work, and have a go at our gachapon. Also, catch @angli_ai on stage: 🎤 Jun 10 · 5:00 PM · WEKA Stage: One Society, or Two? The Age of Autonomous Computers 🎤 Jun 11 · 3:10 PM · Plaud Main Stage: The 100x Company: Rebuilding the Business Around AI with @WilliamBryk (@exa), @hyunjinvkim (@INSEAD), Haydn Sallmann (@googlecloud), moderated by @bernardleong (Dorje AI). Autonomous. Always on. Actually done.

Attention founders - here are the things people want Claude to do but can't yet. At NYC Tech Week, we asked for people's dream workflows. Plot twist: @sai_borg can already do some of them (like LinkedIn tasks 👀). Find out what people said and give Sai a try ➡️ sai.work Huge thanks to Can Atasav (Brightake.ai), Nadav David (Kidoz), Abhinav Singla (Canvas Entertainment), Amore Philip (startup advisor), Sai Kanaparthi & Vineeth M. (Bhalam) for sharing your AI journeys!

Partying at NYC Tech Week and got no time for news and social feeds? @sai_borg got ya. 🤖⚡

Today @MiniMax_AI ships M3 — the first frontier model purpose-built for computer-use agents. Natively multimodal. One model, almost every real-world task. And today, @sai_borg runs it. Our users are the first to use M3 the way it was designed to be used — on a real desktop, doing real work. This is what model-and-agent alignment looks like. → Try it: sai.work

Who's in Singapore during AI Week × @superai_conf ? We are co-hosting Agent Hour with Claude SG, Claw SG, and First Prompt. Light bites and drinks are provided, and seats are limited! Come hear @angli_ai what's actually working in AI agents — along with other fireside speakers. 📅 Date: Tuesday, June 9 🕓 Time: 4:30 PM - 6:30 PM 📍 Location: WeWork, 5 Temasek Blvd, Level 17, Singapore 📩 RSVP: [luma.com/simular-mc8r]

Last week at @aiDotEngineer Singapore we ran a "Sticky Wall" at our booth. Deal: post the task you hate doing most. Winning sticky gets a $500 Sai Pro plan. Winner: "buy my groceries from RedMart every week." One sticky 👉️ Built on Simulang. P.S. the winner forgot to leave their handle 🫠 if you wrote that sticky at our booth, DM us — your Pro plan is waiting. Next task to automate? Tell us ↓