Turning emerging AI concepts into clear, usable, and high-conviction product experiences with OpenAI
Since February 2026 until now, I'm working as the lead designer across multiple 0→1 product explorations for OpenAI. Venice Inc., a design studio based in LA brought me in the project to work on projects with OpenAI directly.
These projects ranged from AI-native business creation, to transparent feed personalization, to agent-based communication systems, to a lightweight iMessage sticker generation experience.
Across all of them, my role was the same: Take abstract, fast-moving ideas and shape them into product experiences that feel understandable, useful, and ready to test in the market to validate.
Company
OpenAI
Role
Lead Designer
Date
Feb 2026 - Present
Platform
Desktop & Mobile
Designing at the frontier
Before diving into the OpenAI products I designed, it’s important to understand the environment those products were were shaped in.
As we know, OpenAI is one of the frontiers in the AI tech space. This means that they operate in a space where innovation is the baseline. Their work isn't just about improving existing products. it’s about exploring entirely new ways AI can show and influence's in people’s lives.
That meant all of the projects I worked on here, often started before there was a clear category, flow, or interaction model to follow. It started as an abstract idea or concept.
Where this work started
Across all projects, the starting point looked similar:
A rough concept
A product direction
Sometimes just a question: “what if we could do this?”
But what was missing was everything in between:
What does the flow look like?
What does the user actually do first?
What should the system explain vs hide?
How does the AI behave?
When does it help vs stay out of the way?
Many more, and most importantly, the core design problem i was solving across every project was
Operating without a Playbook
All of the projects we worked on here weren’t traditional design problems. There were a few constant realities:
Everything was 0→1
Most ideas were still forming while we designed
As OpenAI is the frontier in AI technologies, we had to move fast enough to validate before the market shifts
There were no exact references for what “good” looks like as OpenAI is always will be the innovation frontier
This meant design wasn’t just about execution. It was about:
Defining the structure
Shaping the experience
And often deciding what the product actually is
What I Owned Across All Projects
I worked as the only designer across all these projects, and we're only collaborating purely with the engineers on the team with direct intiatives from the executives like Sam Altman (CEO) , Fidji Simo (CEO of Apps), Vijaye Raji (CTO of Apps), and many more.
There was no Product Managers at all, meaning I'm owning the product and shaping the direction together with the engineers.
Turning early concepts into structured product flows
defining interaction models for new AI behaviors
designing the UI and overall experience
aligning closely with engineers to make ideas buildable
shaping product direction through design decisions
PROJECT 1
From Idea to a Real Business
One of the unique product I worked on was AI-native business builder designed around a simple but powerful idea:
Instead of using one generic assistant, the product introduced specialized AI agents that could act more like:
a CTO helping shape the product
a CMO helping with go-to-market
a COO helping with operations
and other role-based support across the journey
Unlike most AI builders that stop at generation, this concept aimed to support the full journey of turning an idea into something real.
It may sound straightforward as an idea, but the complexity in terms of the designing the experience to make it also feel very simple and easy to users was the biggest challenge as you can imagine, combining the process of building a real company from scratch. The goal was to help user go from:
Setting up your business flow
The gap in the current market
Most AI builder tools already solve one part of the problem well: they can generate something quickly
But after that moment, users are often left with the harder part:
Shaping the product direction
Setting up the business side
Figuring out how to monetize
and understanding what to do after the first output is created
That meant the real gap wasn’t generation. But it was
How I approached the experience
A big part of my role here was shaping the product so it didn’t feel like a traditional business setup tool. We don't want the experience to feel like users are filling up a long form, admin work, legal setup, generic dashboard, etc. Because if that is the case, then the product would lose it's magic almost immediately.
So I designed the experience around a different feeling: The setup needed to feel like the user was already building something real — not filling out requirements before being allowed to begin.
Reframing setup as movement
One of the most important decisions was making the early flow feel more like a guided collaboration with AI than a rigid onboarding funnel.
Since we're utilizing AI agents here
Instead of dropping users into long forms, we shaped the experience so the system could:
Guide users through a few smart questions
Syntheize the direction
and quickly move the user into building
That shift mattered because the strongest part of the product wasn’t just what it could generate — it was how quickly it could make users feel like they were already in motion.
Breaking the step-by-step mental model
Another important product decision was moving away from a strict linear flow. Using the power of AI agents, users can now work on parallel work togethers.
Rather than forcing users to complete everything one stage at a time, I proposed the experience around a more dynamic structure where:
The AI could guide progress
The product could evolve in real time
and the business setup could happen in parallel
That made the experience feel more alive and more AI-native. More importantly, it changed the feeling from:
Designing the post-launch state differently
A lot of products fall into the same trap after launch: They become dashboards full of charts, infrastructure, and generic analytics. But that’s rarely what users actually need first.
So we pushed the post-launch direction toward something simpler and more useful:
That led to a more focused “business command center” direction, centered around:
business health
key signals
actionability
and next steps
Instead of trying to show everything, the design prioritized what actually helps a user stay in motion.
A powerful tool for indie-builders or early entrepreneurs
What made this project interesting wasn’t just the interface. It was the challenge of designing the bridge between AI capability and real-world action.
The hard part for users wasn’t getting AI to generate something. As that is already a common mental model or idea that exist in the current market in terms of AI creating tools.
It was making the experience feel clear enough, motivating enough, and structured enough that a user could realistically move from idea to building something that they could actually launch, continue managing, expanding and improving their business without huge overhead.
PROJECT 2
A social feed users can actually control
Another major app that I designed was reimagining social discovery around context-aware personalization, algorithm control, identity trust, and AI transparency.
A social feed where user doesn't start from fresh zero discovery, starting from the system already knows about you. Because users sign in with their ChatGPT account, unlike most social products that need weeks of signals to slowly figure out a user’s interests, this app gets that advantage as he product can already understand a lot about what they care about from day one.
These includes signals from chat history, memories saved by ChatGPT, behavioral patterns, interest, and many more data under the user's ChatGPT account.
That created a much more powerful starting point for personalization, one where the feed could feel immediately relevant instead of generic.
But that was only one part of the idea. The bigger shift was also:
And in a world where AI-generated identities and synthetic content are becoming more common, As we know from other social apps like Twitter, Threads, Instagram, etc AI-generated accounts, agents, and contents becomes more common another equally important question emerged:
That meant rethinking the concept of this social app from the ground up, not just as a place to endlessly scroll, but as a social-product where user could feel personally relevant, give more user control, and more trustworthy from the start.
The problem with modern feeds in social apps
Most social feeds today are designed to feel personalized, but they still start with a weak understanding of the user. The system has to infer everything slowly through: clicks, likes, follows, watch time, and repeated engagement loops
And even once it becomes more “accurate,” users still have very little visibility into:
Why something appears
What the system thinks they care about
and how to meaningfully influence the algorithm
That creates a familiar problem where the feeds feels personalized, but not truly user-shaped.
At the same time, another trust problem is becoming more relevant. As users are increasingly exposesd to content and accounts without always knowing whether they're interacting with a real person, an AI-generated account or a synthetic identity designed to behave like one.
That creates two major gaps at once:
Starting with context instead of cold
One of the strongest ideas behind this social app was that the product didn’t need to start from zero.
Because users could sign in with their ChatGPT account, the system could already begin with a richer understanding of:
what topics they naturally care about
what kinds of questions they ask
what themes repeatedly show up in their behavior
and what long-term interests are already visible through memory and usage
That made it possible to create a much more compelling first experience: One where the app could immediately feel like it already “gets” the user. Instead of showing generic onboarding interests or broad categories, the product could surface a more personalized set of initial recommendations that already reflected who the user is.
That became one of the most interesting parts of the concept:
Turning personalization into something user can shape
Even though the system could start with a strong understanding of the user, the goal was never to make the feed feel fully automatic. A major part of the product philosophy was that personalization should feel editable, not invisible. That meant users could:
Refine what the system thinks they care about
Tune alogrithm topics up or down
Remove or reject signals that didn’t feel right
and actively shape what they want the feed to become over time
This created a much healthier relationship between user and algorithm, beacuse:
Making the algorithm legible
If users are meant to control the algorithm, they first need to understand what it’s doing.
A major part of the design challenge was figuring out how to expose the logic behind the feed without making it feel technical or overwhelming. That meant exploring lightweight ways to show:
Why a post appeared
What topic or signal it maps to
and how users could tune what they want more or less of
The goal wasn’t full transparency. It was enough clarity to make the alogirthm feel understandable — and therefore more trustworthy.
Designing trust into identity
Another major part of the app involved designing for a social environment where humans and AI could coexist without creating confusion.
If that line isn’t clear, the product quickly loses trust.
So part of the work focused on how to create clearer signals around:
Who is a real user
Who is an AI agent
What kind of account users are interacting with
How identity should be represented in a way that feels obvious, not ambiguous.
To answer this, we explored an identity and verification patterns that could help seperate verified humans AI-created agents public entities from the start that stayed legible inside the product. Making trust and authenticity feel built into the experience itself.
A social concept that feels smart and trustworthy
What made this app especially interesting to designed on was that it challenged three assumptions most social products still rely on
That personalization should start cold
That the algorithm should stay invisible
That identity inside the system will always be easy to trust.
Since none of the solution exists yet, we had to shape the direction, one where the app could start with utilizing the rich context from the user’s existing ChatGPT behavior, while still giving users the ability to shape what they see and trust who they’re interacting with in the current AI-native world.
PROJECT 3
Turning ChatGPT Into a Discovery Layer for People and Businesses
Another product I worked on focused on expanding what ChatGPT could be — not just a tool users ask questions to, but a platform where people and businesses can be discovered directly through conversation. The core idea was simple but powerful:
Instead of only generating answers, ChatGPT could also surface real profiles that are relevant to the user’s intent.
This included:
personal profiles, where individuals could make themselves discoverable for relevant needs
business profiles, where companies could register themselves and be surfaced contextually inside ChatGPT responses
For businesses, the product also introduced the ability to create a custom AI customer service agent, allowing users to chat directly with that business in a more personalized and contextual way.
The design challenge here was making profiles feel relevant, trustworthy, and naturally integrated into the conversational experience without feeling like it breaks privacy. This project pushed ChatGPT to help users to get real-world opportunities and increase discovery for businesses.
PROJECT 4
Having an agent by your side
This next product I worked on focused on a different kind of problem, with the power of current AI agents (e.g: OpenClaw, ClawdBot, etc), we tried shaping how multiple AI agents can work together inside one user experience.
Not just how one AI should behave, but how if everyone has their own agents, and they can work together in a way users actually understand on your usual apps.
Imagine having an assistant that automatically shares context to everyone when you're talking to, and can do all those small tedious work without you having to lift a finger, and you'll can expect for the job to be done everytime you opens the app again.
As AI systems become more capable, a single assistant isn’t always the best model. Some tasks are better handled through multiple specialized agents working together behind the scenes or in visible collaboration.
That introduced a new design challenge:
How should different agents be represented?
How do users understand what each one is doing?
When should collaboration feel visible vs invisible?
How do you make a multi-agent system feel useful instead of chaotic?
My work focused on shaping interaction patterns for:
agent coordination
role clarity
contextual task support
and making multi-agent behavior feel understandable in-product
PROJECT 5
iMessage Sticker Extension
After all those complex product, there's also a fun social extension that I designed that user can use on iMessage. It lets users generate stickers with ChatGPT and send them directly inside a conversation.
Journey Reflection
Where design helped define the product
Being able to get the opportunity to work at the frontier of AI technology fundamentally changes the role and the way I think as a designer.
What this body of work has reinforced for me is that design becomes most valuable when there’s no clear playbook yet. When the product itself is still being defined, design isn’t just about refining interfaces, it becomes a way to shape how new technology turns into something people can actually understand and use.
That was the reality across all of this work. The pace, the ambiguity, and the level of innovation meant there was no clear playbook to follow. It pushed me to think more deeply about:
How products should behave
How users form trust
And what makes something feel genuinely usable, influence people's live, and not just technically impressive.
Working Model
The working model also played a big role. Without traditional PM structures, the direction of the product was shaped through close collaboration between design and engineering, where both sides actively contributed to defining how the product should work, not just how it should be built.
And what makes this experience even more meaningful is that it’s still ongoing. These products are still evolving, still being shaped, and still being pushed forward.
This changes the future dynamic companies can work with the help of AI.
What's next?
As I'm currently working on projects with OpenAI as you read this story, I will continue sharing my learnings and outcomes of my work and experience on how it's like, to help inspire everyone reading.