Emergent, the India-based "vibe-coding" startup backed by SoftBank, launched Wingman, an AI agent that automates routine tasks through chat interfaces on WhatsApp and Telegram. Unlike fully autonomous agents, Wingman deliberately asks for user confirmation before executing significant actions while learning user preferences over time to streamline future interactions.
This measured approach puts Emergent at odds with the "full autonomy" trend dominating the agent space. While competitors race toward OpenClaw-style agents that operate independently — as I covered when OpenClaw hit 100K stars last week — Emergent is betting that users want guardrails, not just speed. It's a smart positioning against the backdrop of Google's Jules coding agent exiting beta and Adobe's new Firefly assistant promising autonomous tool usage within creative suites.
The timing aligns with broader industry momentum around "agentic AI" — systems that manage workflows rather than just respond to prompts. Investment activity supports this shift, with CG Power's arms investing in Japanese AI chipmaker EdgeCortix specifically for "Generative and Agentic AI applications," and Anthropic reportedly drawing VC interest at up to $800 billion valuation for its Claude models powering many of these agents.
For developers building agent workflows, Emergent's confirmation-first approach offers a middle ground between fully manual tools and black-box automation. The WhatsApp/Telegram integration is particularly clever — meeting users where they already communicate rather than forcing adoption of yet another interface. Whether this cautious approach can compete with faster, fully autonomous alternatives remains the key question as the agent space matures.