As AI continues to transform processes across industries, many believe that the future of enterprise AI lies in domain-grounded systems rather than generic language models. In this interview, David Sack, CEO and founder of AKA Foods, offers his thoughts on how breakthrough platforms that structure data, enable workflow integration and leverage human expertise can transform fragmented R&D knowledge into reusable, institutional intelligence
Q1. Can you introduce yourself and what you’re building?
I’m David Sack, founder and CEO of AKA Foods. We build AI systems designed to accelerate complex product development processes, starting in food and beverage R&D. Our core platform, AKA Studio, is a secure AI-powered system that focuses on structuring and activating domain-specific knowledge. In food development, that means formulation data, ingredient functionality, process parameters and sensory feedback. The broader ambition is not simply to apply AI to workflows, but to redesign how tacit expertise is captured and reused.
Q2. AI has become a crowded field. What distinguishes the latest generation of AI systems from earlier tools?
The key shift is from generic language models layered on top of workflows, to domain-grounded systems that integrate structured data, decision logic and real-world feedback loops. Early enterprise AI experiments often relied on chat interfaces sitting above unstructured data. That can be useful for summarisation, but it does not fundamentally change how complex development work is done.
What we are seeing is that the more advanced systems now emerging are vertically integrated; they combine proprietary datasets, structured ontologies, domain constraints and workflow logic. For me, that is where real productivity gains begin.
Q3. Why is domain grounding so important?
In most industries, critical knowledge is fragmented and partially tacit. It sits in spreadsheets, emails, lab notebooks and people’s heads. Generic AI can help search and summarise, but it does not understand functional relationships, regulatory constraints or sensory cause and effect. Without that grounding, outputs may sound plausible but lack technical validity.
Domain-grounded systems encode relationships between variables, so they understand not just language, but function. That is essential in regulated and high-precision environments such as food, pharmaceuticals, advanced materials and biotech.
Q4. How does this apply specifically to product development?
Product development is iterative and knowledge-intensive. Teams run trials, gather feedback, adjust formulations and repeat. Over time, valuable learning accumulates, but it is rarely structured in a way that makes it reusable at scale.
Advanced AI platforms can change that dynamic. By capturing ingredients, formulations, sensory evaluations and process parameters in a structured system, organisations can reduce repeated work, shorten development cycles and build cumulative intelligence over time.
I believe that the real value is not in generating ideas, but by institutionalising learning.
Q5. What role does human expertise play in these systems?
Human expertise remains central. AI should not replace domain experts, but instead should structure and amplify their knowledge. In our view, the most effective systems combine machine intelligence with validated human sensory or performance data.
For example, in food development, taste, texture and aroma are subjective yet measurable. Integrating real human evaluation data into an AI framework creates a feedback loop that pure language models cannot replicate.
Ultimately, I think it’s worth remembering that the goal is augmentation, not automation for its own sake.
Q6. Are these principles limited to food, or do they extend to other sectors?
No, they’re definitely not limited to food, and are easily transferable cross-sector. Any industry where development depends on complex variable interactions, tacit knowledge and iterative testing can benefit from domain-grounded AI systems. That includes pharmaceuticals, cosmetics, materials science and even advanced manufacturing. The common thread is the need to transform fragmented expertise into structured, reusable intelligence.
Q7. What do you see as the biggest misconception about AI in enterprise settings?
For me, the biggest misconception is that access to a powerful language model is enough. In reality, value comes from system architecture. That means proprietary data, structured taxonomies, workflow integration, validation layers and security controls. I firmly believe that without that infrastructure, AI remains a tool. With it, AI becomes an operating layer.
Q8. You’ve mentioned security and data governance a couple of times. How significant an issue is this in practice for enterprise AI adoption?
It’s one of the most significant barriers we encounter, and I think it’s often underestimated in broader conversations about AI.
The paradox is this: the more valuable your data, the more powerful your AI system becomes, but the more reluctant you are to expose that data to a system you don’t fully control. For large food and beverage companies, their formulation libraries, process parameters and supplier relationships represent decades of accumulated IP. Putting that into a generic cloud-based AI tool is not a trivial decision.
So, for example, what we’ve built is a system where the client’s data remains entirely within their control. It is not used to train shared models, it is not accessible across the platform to other users, and there is full auditability of how it is used. That’s not a feature, it’s the foundation. Without it, serious enterprises will always hold back, and they’ll never get the full value from the technology.
As AI moves deeper into core R&D processes, I expect data sovereignty to become a board-level conversation in most large organisations.
Q9. What does the next phase of AI in industry look like?
I think that we will see a move away from horizontal AI tools toward specialised, vertically integrated systems embedded directly into domain workflows. Security and compliance will also become decisive. As companies begin to integrate proprietary formulations, processes and strategic IP into AI platforms, auditability and data governance are not optional.
The winners will be platforms that combine deep domain grounding, robust architecture and measurable workflow impact.
