The opportunity
What happens when the way people interact with software fundamentally changes?
For years, product design has centered on interfaces. We design screens, workflows, navigation, and interactions that help people accomplish their goals.
But AI is changing that relationship.
Users are increasingly able to express their intent directly, delegate tasks to agents, and interact with products without navigating every step themselves. At the same time, the tools used to build software are evolving, making it possible for designers to move beyond prototypes and contribute directly to production experiences.
At Criteo, a global commerce technology company, these shifts created an opportunity to rethink both the products we were building and how we built them.
My work focused on a central question:
How do we evolve an established, UI-driven advertising platform into an ecosystem designed for humans, AI agents, and the people building the next generation of commerce experiences?
Rather than treating AI as another feature, I approached it as a transformation of the product experience, the underlying platform, and our design practice.
The challenge
Several shifts, all at once
Criteo's Retail Media platform supports complex advertising workflows across brands, retailers, and technology partners.
Historically, much of the experience centered on people navigating a UI to configure campaigns, manage creatives, analyze performance, and make decisions.
AI introduced several challenges simultaneously.
- The interaction model was changing. Users no longer needed to navigate every workflow manually. AI assistants and agents could increasingly interpret requests, retrieve information, and execute tasks on their behalf.
- Commerce was becoming conversational. Retailers were exploring AI shopping assistants that could influence how consumers discover, compare, and purchase products. This introduced new monetization opportunities, but also new questions about trust, relevance, and control.
- Our platform needed to support more than human users. APIs, documentation, MCP servers, and machine-readable capabilities were becoming critical customer touchpoints, not simply technical infrastructure.
- The way we delivered design needed to evolve. AI-assisted development created opportunities for designers to participate more directly in implementation, reducing handoffs and closing the distance between identifying a usability problem and shipping its solution.
These challenges were interconnected. Solving only one would not meaningfully prepare the organization for the broader shift.
My approach
Four connected areas of transformation
I helped advance AI transformation across four areas:
1. Intelligent product experiences
Designing how users interact with AI inside the platform.
2. Agentic commerce
Exploring how advertising and product discovery work within conversational shopping.
3. Agent-ready infrastructure
Improving how developers and AI agents discover, understand, and use Criteo's capabilities.
4. AI-native design delivery
Enabling designers to move from identifying experience problems to contributing production-ready changes.
Together, these initiatives connected customer experience, platform strategy, new business opportunities, and the way our team operated.
01 · Making AI useful inside the product
From navigating interfaces to expressing intent
Traditional advertising platforms ask users to understand the system before they can accomplish their goals.
To optimize a campaign, a user might need to navigate several pages, interpret performance metrics, identify an issue, and decide what action to take.
AI creates an opportunity to reverse that relationship.
Instead of asking users to find the right tools, the platform can help them understand what's happening and guide them toward an action.
Designing the in-platform AI experience
I worked on the experience strategy and interaction patterns for AI capabilities within Criteo's Retail Media platform, including the in-platform assistant, recommendations, contextual insights, and conversational support.
The work explored how AI could help users:
- Understand campaign performance without navigating multiple reporting surfaces.
- Identify optimization opportunities and understand their potential impact.
- Receive contextual alerts and recommendations.
- Move from an insight to the relevant workflow or action.
- Maintain continuity while navigating different parts of the platform.
A central design consideration was that AI should not become another disconnected destination.
The value of an assistant comes from its ability to understand the user's context and connect intelligence to meaningful action.
We explored patterns that brought together conversation, alerts, recommendations, and existing platform capabilities, while preserving the user's ability to review and control important decisions.
Creating clarity across multiple AI assistants
As Criteo introduced different AI technologies, we faced an emerging experience problem: multiple assistants with different responsibilities.
One assistant focused on product intelligence, reporting, and recommendations. Another, powered by Intercom Fin, supported documentation, troubleshooting, and escalation.
From a technical perspective, their responsibilities were distinct. From a customer's perspective, both could appear to be simply another chat.
I proposed an experience strategy that separated these responsibilities through clear product positioning and predictable entry points.
The Criteo assistant would focus on understanding and improving advertising performance, while Help & Support would remain the natural destination when users were stuck, needed documentation, or wanted to escalate an issue.
This approach reduced the burden on users to understand the underlying AI architecture, while giving the organization a scalable model for introducing additional capabilities.
Key principle Users should understand what an AI experience can help them accomplish without needing to understand which model, agent, or vendor powers it.
02 · Designing monetization for agentic commerce
What happens to advertising when AI becomes the shopping interface?
Retail media has traditionally monetized familiar shopping experiences: search results, product listings, and product detail pages.
But conversational commerce changes where product discovery happens.
Instead of searching for products through a series of filters, a shopper might ask:
“I'm looking for running shoes for a half marathon. What should I consider?”
An AI assistant can clarify the shopper's needs, explain trade-offs, compare alternatives, and recommend products.
This creates an entirely new challenge for advertising.
Where should sponsored experiences appear when the interface is no longer a predictable page, but an evolving conversation?
And how can retailers monetize those interactions without compromising the trust that makes an AI assistant valuable?
Mapping the conversational shopping journey
I explored conversational shopping as a non-linear decision-making process rather than a traditional purchasing funnel.
The framework identified distinct moments, including initial exploration, constraint setting, feature research, comparison, validation, and purchase handoff.
- Initial exploration
- Constraint setting
- Feature research
- Comparison
- Validation
- Purchase handoff
Each moment had different user needs and different levels of tolerance for commercial influence.
For example, a shopper exploring unfamiliar products might benefit from discovering relevant sponsored products. But a shopper asking whether a particular product has safety limitations needs objective, trustworthy information.
The same advertising treatment should not apply to both.
Defining principles for responsible monetization
I developed experience frameworks and concepts focused on two advertising formats:
- Sponsored products: Clearly identified paid product placements relevant to the shopper's expressed intent.
- Sponsored prompts: Clearly identified, commercially supported suggestions that help shoppers explore relevant products or categories.
The work established principles around relevance, disclosure, user control, and separation between sponsored placement and the assistant's independent reasoning.
A central principle emerged:
Sponsorship can influence what is presented, but it should not purchase the assistant's judgment.
This meant exploring restrictions on when sponsored placements could appear, how they should be disclosed, and when the experience should prioritize unbiased assistance over monetization.
Translating principles into product concepts
I developed concepts and frameworks for sponsored prompts, conversational product discovery, and retailer-controlled monetization experiences.
The work considered not only the shopper-facing interface, but also the implications for retailers adopting these formats: their control over placements, eligibility, policies, and the broader shopping experience.
This helped move the discussion beyond the question of where to put an ad.
Instead, the design work addressed how monetization could operate responsibly within an AI-driven decision-making system.
The outcome A set of experience concepts, interaction models, and guardrails to inform product discovery and future experimentation.
03 · Preparing the platform for AI agents
When the interface isn't the primary customer touchpoint
As AI agents become more capable, interacting with a technology platform no longer necessarily begins with opening its website.
A developer might use an AI coding assistant to discover an API. A campaign-management agent might interact with Criteo programmatically. A third-party application might orchestrate several advertising capabilities without using the traditional UI.
This changes what it means to design a product experience.
Documentation, API errors, authentication, terminology, and machine-readable interfaces become part of the customer journey.
Establishing an API-first design direction
I authored and championed an API-first product design charter for Retail Media, advocating for a broader definition of the product experience.
The goal wasn't to replace the UI.
It was to recognize that different customers need different ways to interact with the same platform.
Some will continue to rely on graphical interfaces, particularly for discovery, review, and workflows that benefit from visual context. Others will increasingly work through APIs, developer tools, and AI agents.
I explored how design could bring consistency and usability across all of these surfaces.
Researching developer needs
To ground this strategy, I conducted external interviews with teams at Stripe, Salesforce, Cloudflare, Contentful, and BrowserStack.
- Stripe
- Salesforce
- Cloudflare
- Contentful
- BrowserStack
The research explored how technical teams approach developer workflows, integrate APIs, evaluate documentation, and use tooling.
It reinforced the importance of predictable behavior, clear documentation, effective developer tooling, and maintaining useful UI experiences alongside programmatic access.
The research helped inform where product design could contribute beyond traditional screens.
Improving the developer experience
I led and contributed to several connected efforts:
- API error experience: Audited error patterns and developer support journeys, then proposed a standardized approach using RFC 9457 to make errors more actionable and consistent.
- API versioning: Explored how clearer change communication, migration guidance, and deprecation practices could reduce disruption and help customers adopt new capabilities.
- Developer portal: Assessed the experience of discovering, understanding, and integrating with Criteo's APIs, including readiness for AI-assisted development workflows.
- MCP enablement: Explored how developers could connect Criteo capabilities to tools such as Claude, Cursor, and VS Code, as well as opportunities to measure usage and identify gaps.
- Cross-surface terminology: Identified inconsistencies in how concepts were described across the UI, API, and MCP interfaces, and proposed starting with a small creative-management workflow to establish shared language.
These initiatives reflected a broader product design principle:
An API response, a documentation page, and a tool description are all user experiences.
As software becomes increasingly agent-mediated, their clarity and consistency directly affect whether humans and agents can successfully use a platform.
04 · Transforming how designers ship
What if designers could fix the problems they find?
AI was changing not only what we designed, but what designers were capable of delivering.
Traditionally, even small UI improvements could pass through several stages of handoff, prioritization, implementation, and release.
A copy inconsistency might be easy to identify and straightforward to fix, but still require engineering capacity and coordination.
I saw an opportunity to challenge that operating model.
Introducing AI-assisted delivery
As Retail Media UX Lead, I helped establish a pilot in which designers could make targeted front-end changes using AI-assisted coding tools and contribute those changes through Criteo's engineering workflow.
This required more than access to an AI tool.
We needed to understand repository access, development environments, Gerrit submissions, review and approval permissions, translations, testing, and deployment dependencies.
The team began with deliberately small changes to learn the process without taking unnecessary production risks.
Pilot 1: Improving the balances experience
Our first proof of concept focused on copy and formatting changes within the balances page.
A designer updated title casing to sentence casing, adjusted column and header text, and worked through translation requirements.
The change gave us our first practical experience moving a designer-created update through Gerrit.
It also exposed important workflow questions: What could designers safely change? What permissions were needed? What still required engineering support? How should changes be validated?
The value was not the size of the UI change.
It was demonstrating that a designer could participate directly in the implementation process.
Pilot 2: Unblocking a customer-facing feature
A second initiative demonstrated a more immediate product impact.
Criteo had introduced adaptive bid multipliers through its external API, but the corresponding UI still showed copy associated with standard placement bidding.
A customer using the API encountered inconsistent information when reviewing the same functionality in the interface.
A designer identified the experience gap and updated the UI copy to reflect adaptive multipliers when that capability was enabled.
The change coordinated with engineering's deployment process and feature-flag rollout, helping unblock the capability's general availability.
It was a practical demonstration of how an AI-assisted design workflow could address a real customer problem.
Rather than treating AI-assisted coding as a novelty, we connected it directly to usability, product quality, and business delivery.
Building a new design operating model
The broader ambition was to expand the role of product designers from defining experiences to taking greater ownership of their implementation.
The pilot created a foundation for scaling this approach across our six-person design team.
It also surfaced the operational changes needed to make the model sustainable: appropriate permissions, code review, testing, design-system alignment, collaboration with R&D, and clear boundaries for production changes.
The opportunity wasn't to eliminate engineering. It was to remove unnecessary distance between design decisions and their implementation.
Connecting the transformation
One direction, four problems
Although these initiatives addressed different problems, they shared a common direction.
AI was changing where users interacted with Criteo, how those interactions were monetized, what infrastructure supported them, and who could contribute to delivering them.
My role was to help connect those shifts to practical customer experiences and organizational capabilities.
The transformation required working across product, engineering, research, and design, moving between long-term strategy and immediate implementation.
In some cases, that meant exploring unfamiliar product models and defining principles before requirements existed.
In others, it meant identifying inconsistencies in existing workflows and helping the team implement small, measurable improvements.
It also meant challenging assumptions about where product design begins and ends.
Outcomes and impact
From strategic frameworks to implementation pilots
- 4
- connected areas of transformation
- 5
- companies interviewed about developer workflows
- 2
- designer-led, AI-assisted delivery pilots
- 6
- designers the approach was set up to scale to
Product experience
- Developed interaction strategies for in-platform AI capabilities, contextual recommendations, and support experiences.
- Proposed clearer roles and entry points for separate AI assistants.
- Helped establish patterns for connecting insights, recommendations, and user actions.
Agentic commerce
- Created conversational shopping frameworks identifying monetization opportunities and trust-sensitive moments.
- Developed sponsored product and prompt concepts with disclosure and relevance guardrails.
- Established experience considerations for retailer control and future experimentation.
Developer and agent experience
- Authored an API-first design charter and advanced developer experience as a product design responsibility.
- Conducted external developer-workflow research.
- Developed recommendations for API errors, versioning, documentation, MCP enablement, and cross-surface consistency.
Design delivery
- Introduced a designer-led, AI-assisted front-end delivery pilot.
- Supported the team's first Gerrit-based UI changes.
- Demonstrated an implementation workflow that addressed a customer-facing API/UI inconsistency and supported general availability.
- Established practical learnings for scaling the approach across a six-person design team.
These outcomes represent work at different stages of maturity. Some were delivered improvements, while others were strategic direction, prototypes, proposed standards, or exploratory initiatives.
Reflection
What I learned
AI transformation is bigger than adding AI to a product.
Some of the most important design problems emerge outside the conversational interface itself. They appear in permissions, documentation, error messages, integrations, organizational workflows, and the boundaries between human and machine decisions.
Trust has to be designed into the system.
As AI becomes more involved in recommendations and decision-making, trust cannot depend on a disclaimer alone. It requires clear boundaries, understandable behavior, appropriate controls, and systems that respect the user's underlying intent.
Design needs to expand beyond the screen.
When an AI agent becomes the intermediary, there may be no traditional UI at all. But there are still users, expectations, failures, decisions, and experiences to design. The design surface expands to include APIs, tool descriptions, machine-readable content, and system behavior.
The way we build is part of the transformation.
Introducing AI into the product is only one side of the equation. Giving designers the ability to explore, prototype, implement, and validate more directly can change the speed and quality of product delivery. But new capabilities only create lasting value when supported by the right processes, collaboration, and accountability.
Looking ahead
A screen, a conversation, an API, or an agent
I don't believe the future of product design is defined by whether people continue clicking through interfaces.
I believe it will be defined by our ability to help people accomplish meaningful things through increasingly intelligent technology.
Sometimes that experience will be a screen. Sometimes it will be a conversation, an API, or an autonomous agent acting on someone's behalf.
My work at Criteo gave me the opportunity to explore that future from several directions: designing intelligent experiences, establishing the boundaries of responsible monetization, preparing technical products for agents, and helping designers become more capable builders.