Safely manage your Zendesk from the AI assistant you already use, via the Deltastring MCP. Beacon configuration platform
← Back to news
ai

Ultrahuman’s former hardware VP raises $5.5M for devices that control AI agents, not just record you

Aina, a hardware startup founded by Apoorv Shankar (former VP of Hardware at Ultrahuman), has raised $5.5 million to build physical devices that control AI agents rather than passively record user activity. The company's first product, Dune, is a three-key macro keyboard designed to trigger shortcuts and scripts based on context—controlling meeting audio/video, running automations, and invoking AI workflows. Aina also developed Radiance (a tabletop video call remote) and Shift (a single-tap agentic button), but user testing revealed Dune resonated most strongly, leading the company to consolidate features and ship that device first. The funding round included backing from Redstart Labs and 360 ONE, alongside notable individual investors including WhatsApp's Kunal Shah and Razorpay's co-founders, signalling confidence in the hardware-as-agent-control thesis.

The distinction Aina is making—between passive capture devices and active control interfaces—matters significantly for CX teams already investing in agentic AI. As PwC and OpenAI bet on agentic customer service, the question becomes whether your team's agent infrastructure will require dedicated hardware to operate effectively, or whether browser-based and software-only triggers will suffice. Shankar's emphasis on "action-oriented" devices that leverage existing context from phones and laptops suggests the next wave of agent tooling won't be about new sensors or always-listening hardware, but rather about reducing friction between human intent and agent execution. For support teams running Agentforce or similar platforms, this raises a practical concern: will agent adoption plateau without better physical interfaces, or are software-based invocation methods sufficient for most CX workflows?

The broader implication is that the agent control layer is becoming a distinct product category, separate from both the AI models themselves and the business applications they power. With OpenAI releasing custom keypads for Codex, Qualcomm experimenting with 40+ device prototypes, and multiple startups pursuing different form factors, the market is signalling that how you invoke agents matters as much as what agents do. For CX professionals, this means vendor lock-in risks may shift: rather than being tied to Zendesk or Salesforce's agent capabilities, teams could find themselves dependent on third-party hardware ecosystems to operate those agents efficiently. The real test will be whether these devices become essential infrastructure or remain niche productivity tools—and whether smaller CX platforms can afford to build native agent control experiences before hardware makers commoditise the interface layer entirely.