We’ve all been there: A brand-new gadget arrives at our door, we pull out the product, and immediately wish an expert could just tell us what to do first. The instruction manual is dense, the diagrams are confusing, and ten minutes later we’re watching a random YouTube video hoping it’s the right model.
SharkNinja decided that moment deserved better. They built an Agentforce agent specifically for unboxing — a QR-code-triggered, mobile-first experience that meets customers the moment they open the box and walks them through setup in a real conversation. No PDFs. No phone trees. Just a scan, a greeting, and step-by-step guidance tailored to the exact product in front of them.
SharkNinja launches roughly 25 new products a year. Their Agentforce deployment spans multiple use cases and countries. The unboxing agent is the latest addition — and it’s already changing how customers experience their first interaction with a new product.
Here’s how they built it, what they learned, and what your team can borrow.
How the unboxing agent works
Since the unboxing agent launched on a pair of premium espresso machines, imagine one has just arrived on your doorstep. Your experience starts with a QR code printed on the product packaging. You scan it on your phone, land in a mobile agent experience, and get a customized welcome message for your specific machine. From there, the agent delivers step-by-step setup instructions in a highly visual, interactive format — guiding you through everything from assembly to brewing your first drink. It answers follow-up questions in context and surfaces relevant videos when a visual would help.
Say you ask an off-script question mid-setup: “What’s the difference between espresso strength options?” The agent pulls from the ingested product documentation to answer in real time, then picks up right where you left off in the guided flow.
Carolin Duerkop, Technology Transformation Partner at SharkNinja, describes the experience. “The QR code on the box is the new instruction manual,” Duerkop said. “Scan it, and you’re in a conversation with someone who knows exactly which product you have and what you’re trying to do. We think every physical product company should be doing this.”
Making product manuals AI-ready
Getting product documentation into a format the agent can actually use was one of the hardest problems the team solved. SharkNinja’s manuals are dense, image-heavy PDFs — never designed for AI consumption.
The team’s solution: Let AI do the first pass. For Stan Konopka, VP of Digital Technology at SharkNinja, the approach was pragmatic. “In some cases we use AI to optimize it, funny enough,” he said.
AI reads the product PDF and generates draft setup steps; a content author reviews, refines the instructions, and adds images before publishing. They prioritized premium SKUs first and used that process as a repeatable template for onboarding new products.
The team is also evolving beyond PDFs as the source of truth, building toward a workflow where structured content is authored directly for the agent from the start.
What powers it: Knowledge articles
Konopka summed up the foundational lesson in SharkNinja’s Behind the Build interview: “The technology’s there, but it’s only as good as the data and the knowledge and the product catalogue.”
The unboxing agent’s architecture is deliberately simple. Both of its data sources live within Data 360: knowledge articles and the Agentforce Data Library containing the original Quick Start Guides and Owner’s Guides as PDFs. An Apex retriever queries them and feeds the results to the agent. Structured content and source documents go in, guided conversation goes out. That simplicity is what allowed the team to build the agent quickly and scale it to new products without waiting on complex data integrations. Each new SKU just needs its knowledge articles created, and the agent can serve it.
The unboxing agent’s logic is defined in Agent Script — a code-based representation of subagents, actions, and routing logic. SharkNinja’s team used AI-assisted tooling to translate their existing agent configuration into Script, making it accessible to team members who aren’t traditional developers. That means non-technical users can shape agent behavior in a builder that feels more like editing a document than writing software.
The team also uses Intelligent Context to ingest unstructured product documentation — Quick Start Guides, Owner’s Guides, FAQs — and make it available to the agent as chunked, searchable content.
A built-in “Compare with Source” view lets the team see exactly which chunks the agent referenced when answering a question, alongside the original PDF. For a brand where product accuracy matters, that traceability is critical. If the agent gives an unexpected answer, the team can trace it back to the specific passage it read and fix the source content directly.
The Digital Concierge takes Data 360 further, pulling in customer context, order history, and product catalog data alongside B2C Commerce. The team is exploring how to connect these capabilities across agents through orchestration. For now, the unboxing agent stands on its own, and ships faster because of it.
![Flow diagram showing how SharkNinja's unboxing agent retrieves product knowledge from two sources (Knowledge Articles and Agentforce Data Library) via an Apex retriever, then delivers a guided setup experience through a rich visual interface. Left-to-right flow diagram with three sections. Left: Content Sources (Knowledge Articles and Agentforce Data Library containing Quick Start Guides and Owner's Guides). Center: Unboxing Agent (deterministic logic and targeted LLM reasoning), connected by an arrow labeled Apex Retriever. Right: Customer Experience (rich visual interface, step-by-step setup, follow-up Q&A, and video tutorials), connected by an arrow labeled Delivers.]](https://www.salesforce.com/blog/wp-content/uploads/sites/2/2026/05/SN-Unboxing-Simple-Diagram.jpeg?strip=all&quality=95)
Under the hood, the unboxing agent is organized into specialized subagents: Product Greeting, Product Assistance, Warranty Registration, Recipe Search, and more. Each subagent is responsible for a distinct part of the customer conversation. An Agent Router determines which subagent should handle each turn, with dedicated subagents for escalation, off-topic requests, and ambiguous questions. This modular design means the team can update one capability (say, adding a new product greeting) without touching the rest.
From FAQ bot to agentic actions: The bigger picture
The unboxing agent didn’t come first. SharkNinja built their Digital Concierge first — a single agent on sharkninja.com that handles both service topics (troubleshooting, FAQ, order tracking) and shopper topics (product recommendations, cart actions) in one conversation. That agent proved the model, scaled to 250K+ conversations, and gave the team confidence to build something more ambitious.
The progression tells a story other teams can learn from: Start with a high-value use case (FAQ deflection and guided shopping), prove it works, build the iteration muscle, then extend into net-new experiences, like guided unboxing that didn’t previously exist.
Here’s how the unboxing agent fits into SharkNinja’s broader Agentforce architecture. The unboxing agent and the Digital Concierge (Sales & Service agent) share a common data foundation in Salesforce Data 360, but each serves a distinct purpose and set of subagents.
Guardrails and “attack the bot”
Building the first version of an agent is the easy part. Making it good enough to represent your brand at scale is where the real work begins.
Konopka describes it this way: “AI in some cases is eager to please. You don’t want that. To make sure the brand is reflected accurately in the agent, there’s a lot that goes into guardrails. The tone, the data, the guardrails: all those things come together. You could have it 95% correct, but that last 5% can make it feel clunky.”
SharkNinja invented what they call the “attack the bot” program to close that gap. Call center reps spend a few minutes each day trying to break the agent: asking the toughest questions, attempting to trip it up, testing edge cases no QA script would catch.
“Get subject matter experts who know their stuff to test the AI,” said Konopka. “That’s where the magic happens.”
The team also invested in “conversation engineering” — weekly meetings dedicated to refining how agents communicate, conversation workshops with CX colleagues, and structured lessons-learned sessions after each major iteration.
“The ‘attack the bot’ sessions with call center reps surfaced more edge cases in two weeks than our QA scripts caught in two months,” said Duerkop. “If you take one thing from SharkNinja’s playbook, it’s this: your frontline team is your best testing tool. Use them early and often.”
5 tips from SharkNinja
Ready to build an agent like SharkNinja’s unboxing experience? Konopka shares this advice for getting started with Agentforce.
1. Start with one high-value use case and prove it out. “Build on the success, instrument it properly from day one. Don’t try to do everything.”
2. Invest in data foundations early. “Understand how AI digests information. Without structured, reliable knowledge underneath, the agent experience will plateau.”
3. Build the iteration muscle from day one. “Create the feedback loops, do the conversational workshops if you can.”
4. Bring your frontline teams in early. “They know the customer better than any engineer does. Your customer experience team can become your most valuable testers and co-creators.”
5. Embrace the journey. “The teams that will win with agents aren’t the ones that get it right the first time. They’re the ones that collaborate and learn the fastest.”
SharkNinja built their unboxing agent by shipping live, learning as they went, and iterating fast. The QR code on the box is now a portal to a guided conversation, and any team with a physical product can follow the same playbook.
Ready to learn more about how SharkNinja is using Agentforce? Check out their Demo Day story here.
