This is a conversation with Justin Johnsen, the lead technical architect at KPMG. This conversation has been edited and condensed for clarity.
I joined KPMG six years ago, and I’ve been a forward-deployed engineer for about a year, working across different clients.
When I started with KPMG as a software engineer, all the code was written by hand.
I studied computer science. I built my career in architecture and later got into consulting and client work.
A forward-deployed engineer works directly with a client and stays with the problem from understanding the business need through delivering a working solution. We work directly with client teams, often on-site.
A lot of what I do now is AI. AI is incredibly capable, but it’s really only as good as its context. Much of my work is gathering the information needed for AI to generate valuable assets and artifacts, then spreading those outputs across client deliverables and software applications.
I also coach client teams on how to use AI effectively — helping them combine technical depth, business understanding, and clear communication to improve and amplify their work.
For one client, I was initially hired to integrate several applications into a complex software system. After months of gathering requirements and refining the context, we used AI to deliver the application in about a month.
The client wanted to understand how we had moved so quickly — and why the process had been so transparent. The client then asked me to help its teams adopt the AI and prompting practices that had made the work possible.
AI has made it easier for me to multiply my output and, in a sense, act as my own enterprise. I can maintain context across multiple clients and projects simultaneously. The bottleneck used to be software development and writing code; now it is gathering context and refining intent before development begins.
One recent application I delivered involved risk scoring: using AI to assess how risky a customer or partner might be for a client to do business with. We brought together the signals used to measure that risk, then used AI to reason across them and generate a score. Independent reviewers could then use that AI-generated score in their assessments.
A pattern I’ve noticed is that more software-development work is moving toward the data and integration layers. Many enterprise challenges come down to helping AI make sense of data, bringing information together for monitoring and visibility, and deriving value from data spread across legacy and cloud applications. Much of my work involves aggregating large volumes of data, using AI to reason over it, and connecting systems that do not naturally communicate with one another.
Being on-site has real advantages. Conversations happen organically, and it is easier to read body language and emotions in person than over Teams. Some clients are remote, but when teams are in the office, I am usually there too.
A forward-deployed engineer offers more than a traditional software handoff. In a conventional implementation, a systems integrator may gather requirements, build the software, and hand it back to the client. That can create a “throw it over the wall” dynamic. I have seen plenty of mangled Salesforce implementations.
By working directly with client teams while developing, training, and coaching, I can transfer knowledge in real time. The client receives the application, along with the best practices, documentation, and artifacts needed to adopt it.
AI gives me more space to define problems and make critical decisions instead of spending that time on repeatable, monotonous tasks. I can focus more on the things that require human judgment.
