Sai launches fleet-based computer-use agents
Simular’s Sai is a “robosecretary” that coordinates autonomous computers in the cloud or on a user’s device to complete recurring desktop work across apps, terminals, and legacy software without APIs. It can turn discovered workflows into replayable code, with Simular claiming 90% lower token use on repeated tasks. [Official announcement](https://www.simular.ai/articles/sai-your-first-robosecretary)
Sai’s compelling idea is not just giving an agent mouse control; it treats computer use as a learn-once, replay-many operations layer. That could make GUI automation economically viable, but reliability, permissions, and recovery from UI changes will determine whether the fleet is genuinely useful.
- –Interface-level automation reaches legacy desktop apps, internal portals, and authenticated software that API-first agents cannot touch.
- –LLM discovery followed by deterministic code replay targets the biggest weakness of computer-use agents: paying for fresh reasoning on every repetition.
- –Cloud VMs, bring-your-own-device support, live visibility, and approval gates make Sai more operationally credible than a chat-only assistant. [Product details](https://www.sai.work/)
- –The 73% OSWorld claim is a useful capability signal, but production value will depend on recovery, secrets handling, and real-world workflow success rates. [Product Hunt launch](https://www.producthunt.com/products/simular?launch=sai-5)
DISCOVERED
1h ago
2026-09-21
PUBLISHED
7h ago
2026-09-21
RELEVANCE
AUTHOR
[REDACTED]