Can Block turn an agentic voice layer into a merchant growth tool?

Block is currently navigating a pivotal crossroads in its product evolution, attempting to bridge the gap between high-fidelity conversational AI and the pragmatic, high-stakes environment of restaurant operations. By positioning its Square Kiosk—a $1,999 self-service hardware solution—as a testing ground for advanced voice interfaces, the company is signaling a shift from traditional point-of-sale systems to proactive, autonomous commerce agents. At the heart of this strategy lies Dessa, a proprietary, research-stage speech-to-speech model that represents a significant departure from standard industry approaches to natural language processing.
The Technological Divergence: Dessa vs. Industry Standards
The current landscape of voice AI in retail is dominated by multi-stage pipelines: Automatic Speech Recognition (ASR) converts audio to text, a Large Language Model (LLM) processes that text to generate a response, and a Text-to-Speech (TTS) engine converts the output back into audio. This conventional architecture is often plagued by latency, loss of emotional nuance, and the inability to process non-verbal cues.
Block’s Dessa model seeks to bypass these bottlenecks by utilizing a direct speech-to-speech architecture. Developed by Block’s research team and trained from scratch on NVIDIA hardware, Dessa processes raw audio signals without relying on pre-existing text-based LLMs or intermediary transcriptions. By training exclusively on human conversational data, Block aims to achieve a level of latency and naturalism that makes the machine sound less like a robotic interface and more like a human interlocutor.
However, from an engineering perspective, there is a critical distinction between "natural conversation" and "transactional utility." While Dessa excels at the fluid mechanics of human-like speech, it is not currently configured to interface with external APIs or backend databases—the essential components of an agentic system. In a restaurant setting, an agent must be able to verify real-time inventory, cross-reference dietary restrictions, apply promotional codes, and authorize payments. While Dessa provides the "voice," the "agentic layer"—the ability to execute these commands reliably—remains a separate development challenge.

Square Kiosk: The Commercial Sandbox
The Square Kiosk, launched in October, serves as the physical vessel for this research. Priced at $1,999, the hardware is designed to address the persistent labor shortages and operational inefficiencies in the quick-service restaurant (QSR) sector. By integrating voice-ordering capabilities, Block is attempting to convert the kiosk from a static digital menu into a dynamic, interactive concierge.
The rationale for this move is supported by broader industry trends. According to recent QSR industry reports, self-service kiosks have been shown to increase average ticket sizes by 15% to 20% due to consistent upselling prompts and a reduction in order entry errors. If Block can successfully marry the naturalism of Dessa with the transactional reliability of the Square ecosystem, it could potentially offer merchants a tool that not only replaces a human cashier but actively drives higher margins through personalized, conversational upselling.
A Chronology of Block’s AI Ambitions
Block’s pivot toward AI-driven commerce did not happen in a vacuum. It is the culmination of a multi-year strategy to integrate artificial intelligence across its entire product suite:
- 2021-2022: Block intensified its investment in machine learning, focusing primarily on fraud detection and risk management for Cash App and Square Capital.
- Late 2023: Internal research teams began the development of Dessa, shifting focus toward generative audio models that could differentiate Block’s offerings from competitors reliant on off-the-shelf APIs from OpenAI or Google.
- October 2024: The formal launch of Square Kiosk, providing a dedicated hardware platform for testing voice-first interfaces in live, high-traffic retail environments.
- Q4 2024/2025: Current phase, characterized by the testing of the voice interface among a select group of researchers and pilot merchant locations, with an emphasis on fine-tuning accuracy for diverse speech patterns and noisy environments.
Implications for the Fintech Landscape
The competitive implications of Block’s strategy are substantial. Major incumbents like Toast and NCR Voyix are also experimenting with AI-driven ordering, but many are relying on partnerships with third-party model providers. Block’s choice to build Dessa from the ground up on NVIDIA hardware suggests a long-term goal of vertical integration. By owning the model, the hardware, and the payment processing layer, Block theoretically gains the ability to optimize for lower latency and higher security—two metrics that are non-negotiable in a high-volume restaurant environment.
However, the "agentic" gap remains the most significant hurdle. For a restaurant merchant, an AI agent that sounds human but fails to accurately record an order is a liability, not an asset. The history of voice ordering is littered with failed implementations that struggled with the complexities of human speech—accents, background noise, and ambiguous requests like "can I get that but without the onions." To prove the business case, Block must demonstrate that its model can move beyond "natural conversation" to "high-reliability execution."

Analyzing the Business Case
For Block, the transition to agentic AI is as much about merchant retention as it is about innovation. As the payment processing market commoditizes, the company’s ability to provide "value-added services" (VAS) becomes the primary driver of growth. If Square Kiosk can demonstrate a measurable increase in merchant revenue through its voice interface, it becomes an indispensable tool that locks merchants into the Square ecosystem.
From a financial standpoint, the cost of running high-compute models like Dessa at scale is a significant variable. Training models from scratch on NVIDIA hardware is capital-intensive. Block will need to prove that the operational efficiency gained by the merchant justifies the price of the hardware and the potential subscription fees associated with advanced AI features. If the ROI for the merchant is not immediately apparent, the adoption of voice-integrated kiosks may remain limited to high-volume, luxury, or tech-forward QSR chains.
The Road Ahead: Challenges and Benchmarks
As Block continues to develop its voice-ordering technology, several benchmarks will determine its long-term viability:
- Accuracy under Pressure: The model must maintain a high success rate in environments with significant ambient noise, such as crowded dining halls.
- API Integration: The transition from a voice interface to an agentic system requires seamless connectivity with existing POS software, including real-time stock monitoring.
- Scalability: Can the system handle thousands of concurrent transactions across different geographic regions without significant latency?
- Security and Trust: With voice-based ordering, ensuring that the AI can handle sensitive payment data without compromising the merchant’s or the customer’s security protocols is paramount.
The industry will be watching closely as Block transitions Dessa from a research-focused project to a feature set within its production software. While the promise of agentic AI is transformative, the transition from "natural conversation" to "functional commerce" is a difficult one. Block’s ability to execute this transition will likely determine whether the Square Kiosk becomes the new standard for modern retail or simply another piece of high-priced hardware in an increasingly crowded marketplace.
Ultimately, Block is betting that its deep integration across the merchant lifecycle—from the register to the kitchen display system—gives it a distinct advantage in training AI that actually understands the business of food service. If they succeed, they will have created more than just a chatbot; they will have created a digital employee that can scale alongside their merchants, marking a fundamental shift in how small and medium-sized businesses manage their operations in the digital age.







