A KPMG engineer's billable work moved from writing code to assembling context

Read "months" at its lowest plausible value, two, and the preparation on that engagement ran at least twice as long as the build itself ^[2]. That changes what a statement of work has to price. It also changes who counts as senior: the person who can sit with a client's business owners for a quarter and come away with usable requirements.

"The bottleneck used to be software development and writing code; now it is gathering context and refining intent before development begins," Johnsen said ^[11]. He put the destination of that work in the data and integration layers, aggregating large volumes of data, getting AI to reason over it, and connecting systems that do not naturally talk to each other ^[14]. Much of his day is collecting the information AI needs, then pushing those outputs into client deliverables and applications ^[7].

After the fast delivery, the client asked how the team had moved so quickly, then asked Johnsen to help its own people adopt the AI and prompting practices behind it ^[10]. He says a forward-deployed engineer offers more than the traditional handoff, in which an integrator gathers requirements, builds the software and hands it back ^[17]. "I have seen plenty of mangled Salesforce implementations," he said ^[16]. The client gets the application plus the best practices, documentation and artifacts needed to adopt it ^[18].

Teaching a client's team to do the prompting is capability transfer, and it competes with the firm's next implementation fee. Here the transfer was what the client bought next ^[10].

The record is one practitioner's account, edited and condensed by Business Insider ^[20]. It identifies Johnsen as KPMG's lead technical architect ^[1]. It does not say whether forward-deployed engineer is a formal grade at the firm or how many people hold it ^[21]. It does show the calendar: five of his six years at KPMG predate the role ^[1], and he started as a software engineer when all the code was written by hand ^[4].

Johnsen said he can maintain context across multiple clients and projects at once ^[12]. He also prefers to be on site, because conversations happen organically and body language is easier to read in person than over Teams ^[15].

The signals used to measure how risky a customer or partner might be were brought together, AI reasoned across them and generated a score, and independent reviewers could then use that score in their assessments ^[13].