What it is

Forward deployed design is an embedded practice. The designer works inside the operation they are designing for: on the factory floor, in the trading room, on the hospital ward, in the back office. They learn the context by living in it.

A product designer shapes one product for many users. A forward deployed designer shapes many surfaces for one organization: its workflows, services, authority structures, and decision rights. The embed itself is the work.

Technical deployment speeds up individual tasks and leaves the business much as it was. This practice redesigns the whole flow, so value moves faster from end to end.

The core argument

Embedding engineers with a customer works well for traditional software. Generative AI asks for more. Traditional software starts from a user's pain in an existing workflow and builds an interface that gets the same job done faster. An accounts payable clerk who spends three hours matching invoices to purchase orders across two systems gets a cleaner, integrated version of that task.

AI changes what the old workflow is for. Used the familiar way, it makes individual employees faster at the tasks they already do. A paralegal summarizes a brief in seconds. A developer writes boilerplate instantly. A marketer drafts campaign copy from a prompt. Performance on those rote tasks improves by roughly 40%.

Across the whole company, those local gains rarely reach the bottom line. Optimize a few steps in a long chain and the delay moves to the steps you left alone, which then set the pace.

Take a commercial loan approval: data gathering, risk assessment, compliance review, pricing, managerial sign-off. Speed up risk assessment by 40% and the assessors finish faster, then pile up at the door of a compliance review that runs at its old pace. The customer's decision lands a few minutes sooner. The business model has not moved. The work now waits faster.

The unit of analysis is system throughput. A company earns more when the whole value stream moves faster, with fewer defects, less rework, and less expediting. Local AI tends to raise work in progress: more half-finished cases, more pending approvals, more items in queues. The organization feels busier while the customer waits the same.

AI-native workflow redesign

To use what AI can now do, redesign the workflow around its capabilities and retire the steps built for human limits.

Corporate workflows were shaped around human constraints. People process one demanding thing at a time, hold little in working memory, need rest, and work fixed hours. Approval hierarchies exist because one person can review only so many documents an hour. Siloed departments exist because human expertise is narrow. Status dashboards exist because a manager cannot hold ten thousand live operations in mind.

Agentic systems carry none of these constraints. A capable model reads a 500-page filing in seconds, checks it against ten years of company history and current policy, and returns a probabilistic risk assessment. It runs thousands of these in parallel, around the clock, across time zones.

Using AI only to help a person read faster leaves most of its value on the table. Redesign moves the person out of processing and into orchestrating, deciding, and handling exceptions.

Once AI clears the cognitive bottleneck, authority becomes the constraint. The question shifts from what the system can work out to what it is allowed to do. AI-native design redraws decision rights, escalation rules, audit trails, and accountability. Leave those alone and the organization pairs fast machine recommendations with the same slow permission structure.