Ahnaf Prio

Ahnaf Prio is a Senior Engineering Manager at Best Buy, where he leads the agentic commerce team. His work focuses on how merchants can serve customers using AI assistants while supplying accurate product information, maintaining cart and checkout state, and defining what an agent is authorized to purchase.

Background and earlier work

Prio came to the United States from Dhaka, Bangladesh, for higher education. He graduated from the University of Minnesota Morris in 2019 after studying computer science and later earned a master’s degree in management from the University of Illinois Urbana-Champaign.

In “Recent Graduate: Startup or Consulting?”, Prio described interning at a healthcare startup during his final college semester and then working as a software consultant with a team in the finance and insurance industry. He contrasted the startup’s small team, rapid feature delivery, and limited emphasis on testing and refactoring with the consulting environment’s structured onboarding, clearer ownership, and dedicated support channels. He presented both as useful but substantially different engineering environments.

Prio later served as CTO of a restaurant-technology company recognized as Minnesota’s Emerging Startup of the Year in 2021. The retained sources do not establish that company’s name or that he co-founded it. He is also the former CTO of Invive.io, a biotech startup that explored innovation in life insurance; the retained evidence does not establish that he co-founded Invive.io.

Building the merchant side of agentic commerce

In The Agentic Commerce Stack, Prio presents shopping as a sequence of operations with distinct requirements: discovery, product selection, loyalty, pricing, payment, fulfillment, and post-purchase service. He argues that merchants should expose structured product data and checkout APIs rather than depend on agents navigating screenshots, page structures, and forms. Browser automation can be slow and brittle and may trigger merchant security systems before a transaction reaches payment.

• Protocol responsibilities: Prio assigns each protocol a specific role. The Model Context Protocol exposes tools and capabilities; Agent2Agent supports communication between agents; ACP and UCP represent commerce concepts and transactions; and AP2 addresses scoped payment authority. This separation explains why tool access alone does not create a dependable shopping experience.
• Catalog synchronization: Product feeds let shopping services index merchant data before a customer makes a request, avoiding large numbers of live catalog queries. The tradeoff is that merchants must continually synchronize inventory and changing product attributes while adapting their data to differing ACP, UCP, and Meta feed schemas.

A bakery demonstration of checkout and permission

Prio demonstrated these ideas with a fictional bakery inspired by Ginny, his orange tabby. A customer agent requests products, adds shortbread to a cart, and proceeds through checkout. The example makes the underlying tool calls and agent-to-agent exchanges visible and models checkout in three states: not ready for payment, ready for payment, and completed. An item being in a cart does not itself make the checkout payable; a payment choice and authorization are still required.

The demonstration uses AP2 experimentally to represent bounded purchasing authority. Prio describes authorization that includes a maximum amount and currency, supports revocation, and permits a single use. He uses it to frame questions that more autonomous shopping must answer: who authorized the agent, what it may buy, and how much it may spend. He distinguishes this experiment from payment integrations available at the time of the presentation and identifies AP2 adoption, identity and consent standards, ACP–UCP convergence, and multi-agent checkout delegation as areas still developing.

Evaluating commerce agents

Behavioral evaluations are central to Prio’s approach. In the bakery demonstration, the agent refuses a request to reveal a discount code. He uses that exchange to show why testing must cover what an assistant discloses, whether it stays within its retail purpose, and how it responds to attempts to obtain sensitive information about other customers.

He also recommends protocol-compliance checks to verify that feeds and transaction data conform to receiving services’ requirements, latency benchmarks because delays can interrupt or divert a purchase, and response-quality evaluations developed with product colleagues. Where useful, a language model can act as a quality judge. The starter presented with the talk includes three services, evaluation templates, a catalog-synchronization process that converts product data into ACP, UCP, or Meta formats, and agent skills for customer agents, merchant agents, and catalog synchronization.

Prio also owns the open-source repository skills-evals, which is described as a tool for detecting when agent skills, instructions, or rules stop triggering or working. The retained GitHub profile does not support a more detailed account of its internal evaluation design.

Writing, open source, and community service

In “Number of years of experience !== Competent Engineer”, Prio argues that employers should identify the knowledge and responsibilities a position actually requires instead of treating years of experience as a dependable proxy for competence.

His GitHub profile also identifies him as the owner of Mama Lang, which the repository describes as an esoteric programming language. More detailed claims about its language design or cultural background are not established by the retained evidence.

Prio serves on the Board of Directors of the University of Minnesota Morris Alumni Association and on the Neighborhood Sales Tax Revitalization Board for the City of Saint Paul.