Quick summary:
- Most enterprise AI tools require Python developers, locking out the 99% of employees who could benefit from automation
- One-third of CEOs don’t trust AI, but hallucination-free platforms using neuro-symbolic architecture are changing that
- Boring back-office processes like procure-to-pay offer the safest place to build AI skills before tackling revenue-generating workflows
JBI Interiors reads construction plans, fabricates custom wood and metal finishes, and coordinates with a dozen other trades on every job. It’s exactly the kind of old-line manufacturing business you’d never expect to run on AI agents.
Yet the company now processes purchase orders, negotiates with vendors, and manages supply chains using autonomous workflows that make decisions without hallucinations or human intervention. The transformation started with one simple question: What if AI could read standard operating procedures in plain English and turn them into code?
Trevor Seim, head of sales at Kognitos, builds hallucination-free AI platforms for enterprises. On this episode of The Disruption Is Now, he tells host Greg Matusky why most corporate AI projects fail, how to make automation trustworthy for skeptical CEOs, and why the boring back-office processes are the best place to start your AI journey.
Watch now:
Key takeaways:
Eliminating AI hallucinations means abandoning goal-oriented models
Most AI systems fail enterprises because they’re built to achieve goals by any means necessary. Give a language model a task and it will find creative ways to complete it, even if those ways vary wildly from one attempt to the next.
Kognitos solves this with neuro-symbolic architecture. The platform uses generative AI to understand workflows from documentation like SOPs or process descriptions. Then it translates those workflows into what Seim calls “English as code,” creating deterministic decision trees that produce identical outputs every single time.
“Any time that you use a goal-oriented outcome, your answers may vary,” Seim explains. “What we’ve come up with is we are using the best of Gen AI, the ability to understand the workflows that you’re trying to achieve. We are using Gen AI to create English as code to codify the workflow and the processes. And we’re going to create that into a very deterministic decision tree.”
This matters for finance teams processing invoices, procurement teams negotiating contracts, and supply chain operations where inconsistent outputs create compounding errors.
The 99% problem that limits AI potential
Seim estimates only 1% of employees can actually build and deploy enterprise AI workflows with current enterprise platforms. The other 99% includes line-of-business experts who understand processes intimately but lack coding skills.
English as code changes this dynamic. It allows subject matter experts to read every step of an automated workflow in plain language. They can see how data moves, how decisions get made, and where exceptions might occur. When the AI encounters something it can’t handle, it sends a message to the appropriate person. This accessibility accelerates adoption.
“Our belief is much like we talked about at the very beginning for end user productivity tools, we need to extend that experience and we need to extend that experience into all of enterprise operations,” Seim says. “We can now have all of the line of business experts interface with AI.”
CEOs trust AI when they can read what it’s actually doing
One-third of CEOs don’t trust AI. That statistic explains why so many enterprise projects stall despite executive mandates to “use AI” and promises of 10X returns.
Trust requires transparency. When AI operates as a black box, executives have no way to verify outcomes or understand errors. Kognitos addresses this by giving enterprise clients what Seim calls “a steering wheel for AI.”
Every automated workflow exists as readable, auditable English as code. Non-technical executives can review decision logic. Compliance teams can verify processes meet regulatory requirements. When something breaks, line-of-business users can diagnose the problem without decoding Python scripts.
The transparency extends to exceptions and human-in-the-loop interventions. Instead of mysterious failures requiring developer troubleshooting, the system routes exceptions to appropriate users with context. A missing price on a purchase order doesn’t crash the workflow or generate cryptic error messages. It simply asks a procurement specialist to fill in the gap, then learns from that input for future orders.
“What we simply do is redirect to the line of business user,” Seim explains. “And that line of business user would have an exception that would simply state, price not found on quote or on order, please help. And it would give them the information they need to have so they can go through and help us and reinforce the model so that we can now learn and have reinforcement learning.”
Boring back-office processes teach enterprises how to use AI
Most organizations want to deploy AI on their most visible, revenue-generating activities. Seim recommends the opposite approach: Start with manual, repetitive back-office work.
“Where are the manual and repetitive processes that exist? Where do customers have a high transaction volume? And how can we help alleviate the pain of processing these types of tasks?” Seim asks. “It’s a great place to start to build the skills and the competence in AI before you start to look into other facets of the business.”
Procure-to-pay workflows, order-to-cash cycles, and invoice processing don’t generate headlines. They also don’t require perfect AI performance on day one. These high-transaction-volume processes offer safe learning environments where teams can build skills, set expectations, and establish trust before tackling customer-facing applications.
The financial returns materialize quickly. Automated processes run 24/7 without holidays. They deliver consistent results. They free expensive human resources to focus on higher-value work that actually requires judgment and creativity.
You don’t need a perfect process to begin automation
Process optimization purists argue you should never automate a bad workflow. Fix the process first, then automate. Seim takes a more pragmatic view.
“I would make the claim that even automating an inefficient process would still have some gains because with automation, we are still running 24 by 7,” Seim says. “We will still ensure consistent delivery results. We don’t go on holiday. We’re working on your behalf all the time.”
Kognitos includes process refinement capabilities that let organizations optimize workflows after automation. As business requirements change, customer needs shift, or better approaches emerge, users can update automated processes without starting over.
Key moments:
- How neuro-symbolic AI eliminates hallucinations with deterministic decision trees (7:22)
- The importance of trust and transparency when one-third of CEOs don’t trust AI (11:08)
- English as code gives non-technical users control over AI workflows (13:41)
- Why boring back-office processes are the best place to start AI adoption (16:22)
- Why automating even inefficient processes delivers measurable value (18:25)
- What has surprised Seim most about AI (19:27)
- How far we are from fully autonomous revenue-generating enterprises (21:50)
Q&A with Trevor Seim, Head of Sales at Kognitos
Q: How does neuro-symbolic AI eliminate hallucinations?
A: We are using Gen AI to create English as code to codify the workflow and the processes. And we’re going to create that into a very deterministic decision tree. So every time we run through a process, this new process that has been creating is deterministic and will give you a very consistent output every time you put something through this type of workflow.
Q: What makes AI inaccessible to most enterprise employees?
A: Any of the other tools that are out in the market today require developers, and they require developers that carry deep skill sets in Python.
And this is meaning that AI is really only approachable to the 1% of the population that knows how to actually communicate with AI for said tasks that are enterprise workflow related. By using the Kognitos platform, we can now have all of the line of business experts interface with AI.
Q: How do you help CEOs trust AI deployments?
A: We like to think about it as giving our enterprise clients a steering wheel for AI so that they now have the ability to actually control the trajectory of AI and give them better command over how AI is being used and the confidence of how AI is being used by using English as code.
So now all of sudden what we’ve done is we’ve now codified in English so any non-technical user can understand every single step of the process that we are using with AI.
Q: What are the two main approaches enterprises take to AI adoption?
A: I think for me personally, what surprised me is the two varied approaches to trying to bring AI into organizations.
We see an approach, which is from the left from end user productivity, extending into code development, extending into workflow process automation, extending into other evaluated service for the enterprise.
And then we see organizations that are jumping into the deep end on the right, creating their own inference models, creating their own LLMs, and really building what they believe AI can be for their enterprise.
I don’t think there’s any wrong approach. I think that it’s a question of how fast you want to have a return on your investment and how predictable do you want your experience with AI to be.
And if you want a fast return, and if you want a very predictable experience, I think this is where looking at tools like Kognitos that are on this journey on the other side of the spectrum really make AI more approachable.
Q: How do you prevent automation from destroying the unique customer experience that differentiates your business?
A: It has always been and will always be the core IP of each organization that we’re working with that gives them that unique customer experience.
So if we rely on automation to just push out automation for the sake of automation, it may lose some of that tailored nature and that experience that customers are always looking for.
Figure out a way to become skilled in AI, figure out a way to take your valuable people, your valuable resources and focus them on the next business problem and help your organization become more efficient and more effective using AI as a core foundation.
