Top 5 tips from Lucian Puca of Mixam on launching Agentic AI transformation

Lucian Puca, Digital Product Manager and Automation and Workflow Lead of Mixam, has been at the forefront of the digital transformation of Mixam’s printing services. Enabling Mixam to lead the way in packaging printing with innovative solutions and a commitment to quality, setting new standards in the industry.

Mixam is a leading online printing platform that creates customizable print products for audiences all over the world, with operations across the United States and Canada, crossing overseas to service the UK, Ireland and Germany, and spanning continents to Australia. Their mission statement is to make print easy, accessible and affordable.

Operating in the traditionally conservative industry of print, Mixam chose to take a revolutionary step in embarking upon the Agentic AI transformation. And it began its AI transformation by finding an agentic AI development partner which had the deep expertise and practical knowledge required to turn their ambitions into a reality, revolutionizing printing quotes with an instant, automated price calculator. Mixam, with Lucian leading the way, turned to Vstorm.

The core challenge which needed to be addressed was in communicating specific needs with customers who lacked knowledge of printing or graphic design jargon (e.g., paper types, CMYK). These customers were possessed of a clear vision of what they wanted with no means of communicating how it should be achieved. This presented an opportunity for AI to simplify the process and guide customers to the end product they truly wanted, translating the customer’s needs into a final printed product.

Translating vision into reality

Lucian views AI as an “engine upon which they can build a car,” implying numerous options for enhancing the user experience and integration with parts of a complex system. But while AI in itself holds a lot of promise, Lucian also recognizes that if it were a ski jumper, it would jump impressively high but fail the landing each and every time. That is why it needs to be helped along by painstakingly engineered processes, such as a robust and practical AI agent capable of sticking the landing.

But Lucian’s core philosophy is based on ethical applications of Agentic AI. The ideal to be kept is that AI assists, not replaces, making tasks easier and saving time for both customers and internal teams.

From the start, Vstorm and Mixam aligned on many fronts, from practical expertise in applied solutions to ethical approach. In a personal interview with Lucian following our cooperation and delivery of an AI agent for Mixam which was up to the challenge, he shared his hands on advice for the successful implementation of AI agents in complex workflows.

We have assembled his top five tips for the successful integration of agentic AI below, but please feel free to dive into the interview and hear Lucian’s words for yourself.

Partnering with Vstorm #

Lucian, on behalf of Mixam, recognized that a complex AI agent capable of delivering their desired results required a specialized and tailored approach, not a simple low-code or no-code off-the-shelf platform, as their core business is selling print products, not developing software to order and direct large language models.

Vstorm was chosen as a development partner because we "spoke the same language" both in terms of expertise and ethics, providing clear explanations of precisely "how" we could achieve Mixam's vision, not just saying "yes, we can do it."

The turning point for Lucian was after the first big workshop, when the entire workflow (from training through validation) was laid out for examination and refinement, and later when the very first proof of concept successfully generated an accurate quote from intentionally vague input (e.g., "I want a book")

Lucian Puca’s Top 5 tips to launch the Agentic AI transformation #

Lucian has been at the forefront of Mixam’s AI transformation, forging Mixam’s reputation as a digital pioneer in the packaging and print industry. His wealth of experience in applying valuable AI integrations to his company workflows have resulted in the following pearls of wisdom:

1. Prepare thoroughly and understand project expectations. As we have all been caught in an information landslide — facts, opinions, and predictions about Agentic AI make it truly harder to manage expectations early on in the process. It may be counterintuitive, yet it remains true. That’s why managers like Lucian take the road less travelled by; instead of trying to soak up the information firehose on AI, Lucian and his team engaged in small steps to “try things out” for themselves. Little by little, they gained the knowledge they needed to know what is doable, what is not (yet), and what they can pursue. Understanding the scope of what can be achieved and how you would like to apply these solutions is the first step to effectively communicating with any solution provider.
2. Stay grounded and realistic about what is achievable. One of the most common traits of managers is to drift off keeping realistic after overdosing on hyperoptimistic projections or doom-mongering. The key element of Mixam’s success was in breaking that pattern. Instead, they remained realistic — thanks to their initial preparation and tinkering with the technologies being applied. Keep the practical applications of your AI agent at the forefront and do not let the hype train carry you away. Chasing the limits of an excited imagination can lead any project with potential off track.

1. Ensure the backend systems are prepared to handle Agentic AI workflows. From the perspective of managers onboarding an AI agent, creating workflows that involve the utilization of LLM’s is a mixture of R&D and classical project deployment. Due to this, Lucian calls these workflows “experimental.” No R&D project starts with clear results at the start, and so targeted results (such as the accuracy of an AI agent) are achieved during the project’s development without early guarantees. To ensure the best results achievable, your backend needs to be prepared for integration. This means getting your own systems in order so that the new AI agent can do its work

1. Implement AI in small increments. Mixam initially implemented AI to more easily navigate knowledge-based pages, finding relevant articles or specific information hidden within long texts. The next step was to launch a live chat agent capable of looking up customer orders and providing status updates. The "big thing" was the integration of a live agent that could provide quotes for products, even if the customer was not aware if Mixam sold it, or what specific options (like paper type or color) would work best. AI has many fields of application and can provide different benefits based on a variety of use cases. But not all use cases produce equal value. Implementing step by step solutions and augmentations can help you realize your company’s full potential through the agentic AI transformation process.

1. Don't be afraid to ask for help. Just as Mixam found the expertise they needed in Vstorm, it is crucial to connect with a trustworthy partner with the practical experience needed to fill knowledge gaps and provide the skillsets beyond those that exist in your team. In such a fresh space, such as Agentic AI, it is too early for your internal IT department to become experts. This status, however, can be claimed by engineering teams that have dozens of AI agent implementations already under their belt. Lucian knew where the limit of his team’s competencies were and was not afraid to ask for help. A realistic understanding of your own company’s limitations and the capabilities of potential AI development partners is key to achieving top results. Enhancing your own tech stack and level of expertise with hands-on experience and know-how can turn visions into practical applications.

Vstorm’s AI agent for Mixam #

Mixam is a self-publishing company that primarily provides printing and fulfillment services for independent authors, publishers, and creators.

70% of new users need help in choosing their way through a plethora of options that new customers might not know of when placing orders online 24/7. — This made the use case ideal for AI Agent deployment, and a relief for the CSS team.

The challenge of creating a true AI printing expert required overcoming the typical issues related to large language models, such as hallucinations. To make a bot not play the guessing game and instead offer what’s actually printable requires narrowing down its options to concrete product specifications. Vstorm achieved this by having the bot use Mixam product options, pulling them from their publishing systems with API calls.

The Vstorm team built the product advisor using the PydanticAI Python-centered framework, with FastAPI for inter-application processes and a powerful RAG vector store for matching requests with products based on Mixam’s always-up-to-date internal knowledge of their product features.

The AI agent uses a sophisticated orchestration pattern with specialized tools that handle different aspects of product specification creation and validation. Each tool is designed as a focused function the agent can call. As a result, the Agent can provide guidance, structure the product to the user’s liking, and generate an order link to pay for during check-out.

Talk to a knowledgeable expert, not a chatbot

At Vstorm we talk with people. Bart, PhD economist and our co-founder, is ready to leverage his experience to help you think through the AI transformation of our business.

Schedule a 20-minute introductory call

Vice President