BOSTON – In this Q&A with Matthias Patzak, AWS executive in residence, from the vendor’s media event on Wednesday, Patzak discussed some of the findings and insights from the AWS Reimagine: Turning AI into Value report released in September.
The conversation touches on AI governance, AI agents and deriving ROI from agentic systems.
Patzak is a veteran German technologist whose career has focused on agile and organizational redesign and AI technology.
What have been some of the most prominent concerns that AWS customers have brought up with you about AI?
Matthias Patzak: I’d say the most common concerns are about governance and using AI in a responsible way. But it differs. In some organizations, cost is a greater concern. Others are more concerned about adoption. And a large part, maybe my number two topic after governance, is that they don't see any value. All these topics are very interesting conversations.
Mostly, I have been focusing on innovation and driving value because if you solve the value issue, then you can turn the conversation from a cost conversation to a return on AI. ROI conversations are getting easier for the customers.
As for governance, do you speak to organizations in different countries? How do they try to mitigate that?
Patzak: Yes, so I do work internationally. I mostly engage with customers who face the same challenges I solved as a customer, because the value our team provides is based on experience, not book knowledge. So, I work with automotive companies. I work with banks. It’s always in regard to using AI in software development and creating agents and business processes. Organizations with a vague understanding of governance and vague rules, mostly defined in vague policies, are also not very good at establishing good governance for their agents.
Surprisingly, the organizations that are most advanced in adopting and scaling AI are in the financial services industry, because they have a lot of governance written down, so they can provide these guidance documents as context to an agent, which is not really a best practice. But the agents need to know which rules they have to follow.
So, it’s important to create governance as an infrastructure, as a container, where you can deploy the agent. Then put governance in the platform and set boundaries for the agents. So, manufacturing, automotive, and financial services companies are the most advanced in creating proper governance for their customers. These are also the organizations, especially in Germany, with their automotive companies, that have high cost pressure right now, with a lot of competition from China. They are curious and ambitious about deploying AI for redesigning business processes. They do not just use AI to provide chatbots for employees; they are really starting to think about AI governance as an important business process and how they can improve it.
I wanted to talk to you about that idea of redesigned decisions. I think that was part of the Reimagine: Turning AI into Value report. What does that look like? What does it mean to redesign decisions, or to let AI navigate the decisions that you're redesigning or working on?
Patzak: I see several different ways. Subject matter experts use AI tooling to digitize poorly documented processes and applications, mostly eliminating Excel sheets. This is the first part, but this is very local optimization. But I've seen with some customers that have a subject-matter expert in a leasing department or an automotive factory plant who creates an agent, which is then used by 200 colleagues. It’s truly bottom-up digitization, which impressed me because it wasn't possible before. And then, when I asked these two people in two different organizations in the research why they hadn’t done this before, the answer was that there was no budget. Asking for budget is super complicated. It takes a lot of time.
Another is organizations taking a business process and not trying to optimize it locally but really optimizing the current business process with agentic AI. And then I see another group. For example, there was a CIO of a large German bank. He took 50 of his most capable people out of the office for a week, put them in a hotel, and said, ‘Hey, what is the most complicated business process we have?’ They said it's mortgages. If a private person asks for a mortgage, it's a complicated process. Usually, you want an answer instantly. Do you get a mortgage? What is the interest rate? But sometimes it can take days or even weeks, because it's a complicated process. And he said, ‘Reimagine this business process from scratch.’
When you say AI native, how do you define that? Because that term gets thrown around a lot.
Patzak: A lot. Interesting question. So, it starts with as few humans as possible involved in this business process, maximizing automation.
Regarding the data, it seems you're saying we have different agents. The data is attached to that agent. But is there a time when they work together, or how do you make sure that the orchestration of those agents runs smoothly?
Patzak: The current best practice, and it’s an evolving space, is that you have an orchestration agent. What really doesn't work is assuming you create an agent and that the agent won't make any errors. So, you need a second agent and a third agent in a different role. Agents in specific roles need deep knowledge, but their context windows are limited. They need an orchestration agent to oversee the overall process, but it is currently implemented as individual software components that organizations and developers must set up. We probably will see products soon.
Esther Shittu is a TechTarget news writer and podcast producer and host covering AI technology.
Shaun Sutner, lead news editor of TechTarget AI, edited this interview for clarity and conciseness.
