All case studies
Case Study · Operative
From Embedded Designers to an AI-First Center of Excellence
A 20% workforce reduction left fewer designers covering the same five-product portfolio. I rebuilt how Design delivered coverage — AI for the repeatable work, people for the judgment calls.
Role
Vice President, Product Design
Initiative
AI-First Design Center of Excellence Operating Model
Scope
5 enterprise product lines
Organization
~20 across Design, Product + Front-End Engineering
Geography
North America + EMEA
The 15-second version
The story at a glance
20%
Smaller team after a workforce reduction — with the same roadmap
5
Enterprise product lines kept covered across North America + EMEA
6
AI-assisted workflows, each with a defined human checkpoint
01 The situation
A 20% workforce reduction left a smaller design team supporting the same portfolio workload. The embedded-designer model no longer scaled without spreading people too thin or reducing design coverage.
02 My move
I shifted Design toward an AI-first Center of Excellence: centralize standards and governance, distribute self-service capability, and use AI for repeatable, lower-risk work while protecting designers for higher-risk decisions.
03 The outcome
The model helped preserve delivery momentum with a smaller team while keeping work consistent, on-brand, and compliant with shared product standards, with human review for accessibility and quality. Its AI agents for design-system, accessibility and microcopy checks were adopted by 100+ engineers and 10 product managers. Over the same period, Operative’s broader modernization delivered 18% higher release velocity and 20–30% faster design/engineering cycles.
The pivotal decision
Rather than preserve the embedded model by spreading the remaining designers thinner, I moved us toward an AI-first Design CoE. We centralized standards, systems, patterns, and governance, then distributed capability through guided AI workflows. AI handled repeatable, lower-risk tasks; designers stayed in the loop for exceptions, new patterns, and high-risk work.
The full story
For those who want the detail
Chapter 01
The challenge
Operative was supporting a complex five-product enterprise portfolio when a 20% workforce reduction changed the capacity equation. The roadmap did not shrink with the team. We still needed to support the same volume of work, but the embedded model depended on designers being available inside individual product teams for intake, guidance, review, and validation.
Spreading fewer designers across more work would have created slower response, uneven design involvement, and more one-off decisions. AI could add leverage, but only if it was connected to a stronger operating model. Otherwise we would simply create inconsistency faster.
Chapter 02
My mandate
My mandate was to redesign how Design created coverage: move from fragmented embedded support toward a Center of Excellence, use AI to absorb repeatable low-risk work, and keep designers accountable for the decisions where judgment, trust, accessibility, security, safety, and product risk mattered most.
The team model
A Design CoE supported Product and Engineering with reusable standards, AI-assisted intake and sizing, design-system guidance, content and accessibility checks, persona validation, and defined human checkpoints.
Chapter 03
How I approached it
01
Reframe the model around risk and leverage
I separated work that required design judgment from work that was repeatable enough to augment. Discovery, new interaction models, trust-sensitive flows, accessibility edge cases, and strategic experience decisions stayed designer-led. Documentation, first-pass validation, and other lower-risk tasks became candidates for AI support.
02
Automate intake and make sizing more consistent
A Jira ticket-creation agent turned meeting notes and working documents into structured ticket drafts, reducing manual translation between conversations and delivery systems. A t-shirt-sizing agent provided a consistent first-pass sizing framework. People still reviewed scope, dependencies, priority, and final estimates before work moved forward.
03
Put the design system inside AI-assisted coding
As vibe coding and AI-assisted development increased, we added design-system guidance directly into the workflow. A plug-in gave vibe coders and AI dev coders approved components, theme, brand, style, and usage context so generated interfaces were less likely to drift into one-off patterns.
04
Add microcopy and accessibility checks early
We introduced focused AI-supported audits for microcopy and accessibility. The microcopy agent checked tone, terminology, and approved product language. The accessibility agent flagged common issues against our standards during prototyping and implementation. These were early-warning systems, not certification; ambiguous or high-severity issues went to people.
05
Use persona agents to pressure-test solutions
Persona and research-validation agents helped pressure-test flows, assumptions, and likely user concerns before scarce design time was spent on deeper review. They gave designers, vibe coders, and AI-assisted developers another validation layer, but did not replace direct customer evidence, research, or human product judgment.
06
Build human checkpoints into the system
Guardrails existed before and after generation. Teams started with approved patterns, content rules, accessibility expectations, and design-system context. AI checks then looked for drift and surfaced issues by confidence and severity. Routine low-risk issues could move quickly; security, safety, quality, brand-sensitive, or high-risk experience decisions stayed human-led.
Chapter 04
Results
This was not theoretical. At Operative Media, we put the model into practice after the workforce reduction, extending design coverage while keeping human accountability.
Design coverage
Reduced capacity
Same workload covered
Intake & sizing
Manual / variable
AI Jira drafts + consistent sizing
Brand / UI
Drift risk
On-brand AI coding guidance
Content & access
Manual reviews
Earlier AI audit flags + escalation
Validation
Bottleneck risk
Persona checks + human ownership
Delivery speed
Baseline
18% higher release velocity; 20–30% faster cycles (broader modernization)
Chapter 05
Why it worked
Operating model before automation
We did not layer AI onto the embedded model and hope for capacity. We changed where standards lived, how support was delivered, and which work needed a designer.
Guardrails at the point of work
Intake, sizing, coding, content, accessibility, and persona validation were connected to shared guidance so teams did not have to find standards after the fact.
Risk-based human involvement
Low-risk, repeatable work received more automation; security, safety, accessibility edge cases, brand-sensitive decisions, and new experience patterns received more human judgment.
Enablement instead of gatekeeping
Product and Engineering gained self-service support while the CoE maintained a common system for consistency, quality, and escalation.
Chapter 06
What I learned
AI can increase design coverage, but it also amplifies the operating model beneath it. Without shared standards and governance, it can create inconsistency faster.
Human-in-the-loop does not mean a designer manually reviews every output. The scalable model is to define risk thresholds, automate routine checks, and make escalation explicit.
The best use of saved capacity is not production volume for its own sake. It is more designer time for research, cross-product architecture, new interaction models, and the high-risk decisions where judgment creates the most value.
Chapter 07
The system, visualized
Leadership takeaway
AI worked because it was part of the operating model, not a side tool. The CoE centralized standards, agents expanded low-risk support, and human checkpoints protected security, safety, accessibility, quality, and high-risk decisions. That let a smaller team support the same work without giving up consistency, brand alignment, or design accountability.