Data silos may still exist 10 years from now, but Boomi’s Michael Bachman believes they will no longer keep organisations from finding and accessing their data. As Head of Research and Emerging Tech Strategy within Boomi’s AI Centre of Excellence, he and his team are tackling data silos and fragmentation through semantic comparisons and AI.
Bachman holds two patents, one for the automated generation of data objects and another for automated data mapping, both using semantic comparisons and AI. He believes these methods could make many of today’s automation and integration challenges a thing of the past. During Boomi World 2026 in Chicago, he discussed these ideas and more in an exclusive interview with Frontier Enterprise.
What major technological and market changes do you expect over the next five to 10 years?
I see many things being different from today. This is my opinion, and a great deal can change over the next 10 years, so I don’t know exactly what will happen. But if I were to identify some of the major areas in which markets will change, we’ll probably have the first commercially available quantum computer sometime after 2030. We’ll also have to think about quantum applications that need to integrate with conventional computing applications.
We’ll probably be closer to whatever artificial general intelligence (AGI) means, once we have a solid definition of general or superintelligence and what it takes for a system to be recognised as such.
We’ll need conventional computing systems and applications that are ready for a post-quantum world. In my opinion, as soon as commercially available quantum computing enters the market, it will put any conventional system without post-quantum encryption at risk. We, as a species, already have this technology. That doesn’t mean it has necessarily been implemented by enterprises, including our own, but we’re certainly examining it.
I don’t believe large language models (LLMs) will get us to AGI, whatever that turns out to be. We’ll need different types of computational and logic-based systems to get there. LLMs will be part of it, but world models and symbolic models will also be important.
Ultimately, energy will be critically important and something we’ll have to consider in the years ahead. There are many other things I’m thinking about, but the major areas for the next five to 10 years are quantum computing, chips, energy, and how we’ll integrate those future environments with today’s world.
Will data silos remain a challenge for organisations five to 10 years from now?
Probably, if they aren’t accessed through a good semantic network. If they are, perhaps not. A good semantic layer means that however you and I see the world, and however we understand our organisation’s context, the metadata and semantics shaping those perspectives must refer to the same underlying objects or data.
With that solid semantic foundation, we no longer need a common user interface. You need yours, and I need mine. Everything may remain siloed, but it can be made virtually addressable and available through a semantic layer over those silos.
That way, it doesn’t matter where the data resides. What matters is how it is accessed and what the user has permission to access. How it is accessed depends on the semantic layer knowing both where all the data resides and how the user thinks about it.
Many organisations don’t know where their data is or, worse, what data they have. How can they address this?
These are big questions, and I don’t have all the answers, but I have some ideas about how organisations might approach them. One thing Boomi does is provide access to data wherever it resides. Establishing connections to that data, regardless of the protocol, and making it accessible is an important first step. Once the data can be accessed, it can probably be searched and traversed.
As semantic tools become more sophisticated, machines will do more of the tagging currently performed by people. More importantly, the semantic layer will need to distinguish between users such as you and me.
For example, when you write an article, you know your style. I know mine, too. I love using semicolons, but I don’t like using em or en dashes, although many people do. Now that generative AI uses them so frequently, people are avoiding them because they don’t want their writing to be mistaken for AI-generated content. In the same way, I know the context I prefer, what my tells are, and what I want when I ask for something. My semantic interface will need to know where to retrieve it.
The first step is connecting to all the systems. The second is having a good knowledge graph and semantic layer that know where to retrieve the data. The third is having that semantic system understand you and your persona, so it becomes a matchmaker between what you’re thinking and the data you need.
When it comes to agentic AI, what do organisations still misunderstand?
There are many possible answers, but one is the assumption that agentic systems are simply agents. Traditional software applications use deterministic logic, including if-then statements, switch statements, cases, and similar constructs. Their rules are defined by humans, so the same input produces the same output each time.
Agents, by contrast, use stochastic or probabilistic logic at their core. In other words, there is an element of randomness. Although an agent can produce consistent answers, its logic is typically coupled with an LLM or another type of generative language model built on neural networks. Consequently, the same input won’t always produce the same output, and the expected outcome can vary.
An agentic system combines the two: deterministic logic from conventional software or human-defined rules and probabilistic or stochastic logic from agents, applied where each makes sense. I like to say that agents are used to discover novelty, while determinism provides consistency. Boomi applies that combination in its platform.
What are some of the security blind spots associated with agentic AI?
Model Context Protocol is still evolving, and newer protocols can introduce security blind spots. Blind spots can also arise when organisations lack the appropriate systems to audit and track activity.
In my view, however, most blind spots still come from the same source: people. Humans are the biggest blind spot in any security breach. If a system can use your credentials to access data, your face as an image, or your voice as audio, each can become an attack vector.
A more common day-to-day blind spot in enterprises is using AI, agents, or agentic systems without understanding what they’re doing or why they’re doing it. When we talk about return on investment, that also means protecting the investment. It is therefore critical to understand human weaknesses both when a person is involved in the process and when no person is involved.
Can AI help address human-related security blind spots?
I think we’re trying to use it that way. As a species, we’re already exploring how AI can address human-related security risks. At Boomi, we govern how tools and agents are created within our platform and apply guardrails, although those measures can protect users only to a point.
More broadly, phishing provides a simple example. AI can help train people to identify phishing emails and recognise where blind spots exist. We’re trying to become more self-aware and determine how AI can help people better understand the security risks associated with building or deploying agentic systems.
How do you expect quantum computing and AI to intersect over the next five to 10 years?
Their intersection is inevitable, but it will be challenging at first because quantum computing behaves more like an analogue system than a digital one. AI today is designed for computation on digital systems, not analogue ones. Moving AI closer to the edge could help it understand how the world works because the world itself is analogue.
From a quantum-computing perspective, error correction is the first challenge to address and remains the field’s biggest problem. At present, computation requires many qubits and is expensive because errors, including those arising from poorly entangled systems, must be accounted for. Such errors remain widespread.
Once error correction is addressed, perhaps in the early 2030s, we can begin determining how AI will interface with quantum systems. Before AI and quantum computing merge directly, I expect an arrangement similar to an agentic system: Quantum computing will solve a particular set of problems, while AI will incorporate those results into a broader ecosystem. That will require an integrative approach that brings different computing systems together.
Eventually, I believe AI and quantum computing will merge, perhaps in the early to mid-2030s, although much could change before then, and we may not be as close as some industry observers suggest. The combination will probably begin at the system level, followed by AI being embedded in quantum systems.
