Qingni Wang

People long believed in human-like computer use. But software (even terminals) is built for humans, not AI agents. Why not let agents build their own? Meet 𝗔𝗴𝗲𝗻𝘁 𝗦𝟰, an agent scientist evolving its own interfaces, letting an open 27B model beats GPT-5.5 & Opus 4.8. With its evolved interfaces, the same model improves 𝟯.𝟴× and costs <𝟱% as much per task as either frontier model. Agents today read screenshots and act through click, scroll, type, maybe a shell. So an agent fills a spreadsheet cell by cell when one line of code would do, or prints two thousand lines of a file to find one function. Every new app or model release means a human reads trajectories, diagnoses failures, and hand-writes new primitives. Skill-distillation methods can reorder what the interface allows, but they can’t extend it. Agents compensate with more test-time compute instead of a better interface. 𝗠𝗼𝗱𝗲𝗹𝘀 𝗶𝗺𝗽𝗿𝗼𝘃𝗲; 𝗶𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝗲𝘀 𝘀𝘁𝗮𝘆 𝗳𝗿𝗼𝘇𝗲𝗻. “We want AI agents that can discover like we can, not which contain what we have discovered.” — Rich Sutton, The Bitter Lesson Hand-designed interfaces are what we have discovered. Agents can read their own trajectories, spot what’s missing, and write and test code. 𝗜𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝗲 𝗱𝗲𝘀𝗶𝗴𝗻 𝘀𝗵𝗼𝘂𝗹𝗱 𝗯𝗲 𝗮𝗻 𝗮𝗴𝗲𝗻𝘁 𝗰𝗮𝗽𝗮𝗯𝗶𝗹𝗶𝘁𝘆. That’s what Agent S4 does. Agent S4 makes a 𝗿𝗲𝗽𝗼𝘀𝗶𝘁𝗼𝗿𝘆 𝗼𝗳 𝗶𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝗲𝘀 𝗮𝗻𝗱 𝘂𝘀𝗮𝗴𝗲 𝗸𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 the unit of evolution. A task agent attempts synthetic tasks with no success labels or verifiers. A meta-agent studies the trajectories, forms hypotheses about which actions, observations, or knowledge would help, and builds descendants through mutation or crossover. Validation tasks the meta-agent never sees decide which repositories survive. The model and harness stay fixed, and the meta-agent 𝗻𝗲𝘃𝗲𝗿 𝘀𝗲𝗲𝘀 𝗮 𝗯𝗲𝗻𝗰𝗵𝗺𝗮𝗿𝗸 𝘁𝗮𝘀𝗸. 𝗢𝗻𝗹𝘆 𝘁𝗵𝗲 𝗶𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝗲 𝗲𝘃𝗼𝗹𝘃𝗲𝘀. With the evolved repository, Qwen3.8-27B goes from 𝟱.𝟲% to 𝟮𝟭.𝟯% on OSWorld V2, 𝟯.𝟴× the same model with fixed interfaces and a new state of the art for open-source computer-use agents. It beats GPT-5.5 (13.0%, $25.50/task) and edges past Opus 4.8 (20.6%, $72.40/task) at $𝟭.𝟭𝟬 per task. On ScienceBoard, the same model goes from 41.4% to 58.6%. The analysis is the more interesting part. Evolution mostly deepens interfaces the agent already has: 4 new interfaces and 127 extensions. Give the meta-agent a scratch VM to test its code, and it builds 12 new interfaces and 253 extensions, reaching music scores, audio projects, and 3D scenes. The best repository came from crossing two earlier branches. In one case study, the seed agent hand-builds a browser client mid-task and debugs through failure after failure; the evolved agent calls page_text and reaches the same message at tool call 𝟭𝟭 instead of 𝟯𝟳. Human-like computer use was a starting point, not the destination. The interface an agent acts through is a second axis of improvement, beyond better models and harnesses, and agents can now design it themselves. 𝗘𝘃𝗼𝗹𝘃𝗶𝗻𝗴 𝗮𝗴𝗲𝗻𝘁 𝗶𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝗲𝘀 𝗶𝘀 𝗮 𝗻𝗲𝘄 𝗱𝗶𝗺𝗲𝗻𝘀𝗶𝗼𝗻 𝗼𝗳 𝗮𝗴𝗲𝗻𝘁 𝘀𝗲𝗹𝗳-𝗶𝗺𝗽𝗿𝗼𝘃𝗲𝗺𝗲𝗻𝘁.

𝐁𝐞𝐭𝐭𝐞𝐫 𝐫𝐞𝐬𝐮𝐥𝐭𝐬 & 𝐥𝐨𝐰𝐞𝐫 𝐜𝐨𝐬𝐭, 𝐚𝐬 𝐬𝐨𝐨𝐧 𝐚𝐬 𝐲𝐨𝐮 𝐩𝐮𝐭 𝐟𝐫𝐨𝐧𝐭𝐢𝐞𝐫 𝐚𝐠𝐞𝐧𝐭𝐬 𝐨𝐧 𝐋𝐄𝐀𝐒𝐇! 🦮 Why use a frontier model at every step when a lightweight agent can handle lots of the work? Introducing 𝐋𝐄𝐀𝐒𝐇: 𝐔𝐧𝐜𝐞𝐫𝐭𝐚𝐢𝐧𝐭𝐲-𝐀𝐰𝐚𝐫𝐞 𝐖𝐞𝐚𝐤-𝐭𝐨-𝐒𝐭𝐫𝐨𝐧𝐠 𝐄𝐬𝐜𝐚𝐥𝐚𝐭𝐢𝐨𝐧 𝐚𝐧𝐝 𝐎𝐯𝐞𝐫𝐬𝐢𝐠𝐡𝐭. LEASH unleashes the strong agent only when the weak agent's latent uncertainty signals the need, and keeps it on a leash afterward. Two key challenges: 1️⃣ 𝐖𝐡𝐞𝐧 𝐭𝐨 𝐞𝐬𝐜𝐚𝐥𝐚𝐭𝐞? Use the weak agent's latent uncertainty as an inexpensive signal that can be monitored at every step. 2️⃣ 𝐇𝐨𝐰 𝐭𝐨 𝐦𝐚𝐢𝐧𝐭𝐚𝐢𝐧 𝐨𝐯𝐞𝐫𝐬𝐢𝐠𝐡𝐭 𝐚𝐟𝐭𝐞𝐫 𝐞𝐬𝐜𝐚𝐥𝐚𝐭𝐢𝐨𝐧? Even frontier agents can go astray. And when served through black-box APIs, their internal states are inaccessible. LEASH enables continued oversight without accessing the strong agent's internals. 𝐓𝐡𝐞 𝐤𝐞𝐲 𝐢𝐧𝐬𝐢𝐠𝐡𝐭: Knowing when to escalate—and when to intervene—is just as important as knowing how to act. The future of agentic AI isn't about using the strongest model everywhere. 𝐈𝐭'𝐬 𝐚𝐛𝐨𝐮𝐭 𝐝𝐞𝐩𝐥𝐨𝐲𝐢𝐧𝐠 𝐢𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞 𝐰𝐡𝐞𝐫𝐞 𝐢𝐭 𝐦𝐚𝐭𝐭𝐞𝐫𝐬 𝐦𝐨𝐬𝐭.

𝐌𝐞𝐞𝐭 𝐭𝐡𝐞 𝐰𝐨𝐫𝐥𝐝'𝐬 𝐛𝐞𝐬𝐭 𝐜𝐨𝐦𝐩𝐮𝐭𝐞𝐫-𝐮𝐬𝐞 𝐚𝐠𝐞𝐧𝐭, 𝐒𝐚𝐢. 💻 Sai now 𝐭𝐨𝐩𝐬 𝐎𝐒𝐖𝐨𝐫𝐥𝐝 2.0 𝐰𝐢𝐭𝐡 𝐚 73% 𝐬𝐮𝐜𝐜𝐞𝐬𝐬 𝐫𝐚𝐭𝐞, outperforming GPT-5.6 Sol and Claude Opus 5 𝐚𝐭 𝐬𝐢𝐠𝐧𝐢𝐟𝐢𝐜𝐚𝐧𝐭𝐥𝐲 𝐥𝐨𝐰𝐞𝐫 𝐜𝐨𝐬𝐭. What excites me most is that OSWorld 2.0 moves beyond short benchmark tasks toward longer, more realistic computer workflows, where reliability, planning, memory, and efficiency really matter. This is exactly the direction we have been pursuing at Simular: 𝐁𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐚𝐮𝐭𝐨𝐧𝐨𝐦𝐨𝐮𝐬 𝐜𝐨𝐦𝐩𝐮𝐭𝐞𝐫𝐬 𝐭𝐡𝐚𝐭 𝐜𝐚𝐧 𝐫𝐞𝐥𝐢𝐚𝐛𝐥𝐲 𝐡𝐚𝐧𝐝𝐥𝐞 𝐫𝐞𝐚𝐥 𝐰𝐨𝐫𝐤, 𝐧𝐨𝐭 𝐣𝐮𝐬𝐭 𝐝𝐞𝐦𝐨𝐬. Proud of the team! There is still a long way to go toward reliable, efficient, superhuman computer use, but it is closer and closer!