Agent studio
Agent Studio allows developers to accelerate the creation of AI agents. Built on Algolia’s AI-first infrastructure, it gives users the power to combine large language models (LLMs) with real-time search, behavioral data, and configurable logic.
I redesigned the agent creation experience to address low adoption driven by uncertainty of what to build, lack of control, and high perceived risk. By shifting from a fragmented, code-first workflow to a preview-first builder, I made agent behaviour visible and testable from the outset; enabling teams to understand what they were building, iterate quickly, and confidently move towards implementation. This established a more accessible and scalable foundation for adopting AI within the Algolia platform.
Hypothesis: Making agent experiences visible, controllable, and fast to iterate will reduce perceived risk and drive adoption across the platform.
Year: 2026
Role: Staff Product Designer (UX/UI) covering the full end-to-end process
Team: Working across 4 squads each with a Product Manager and 4 engineers
Company: Algolia
Why adoption was low
Agent Studio’s capabilities were strong; adoption was not. I worked closely with research to synthesise customer insight and identify four structural barriers. Customers were unsure what to build, which led to decision paralysis. They lacked behavioural control, which created fear of unsafe or unpredictable outputs. Setup and iteration were slow, making experimentation feel costly. And there was perceived risk to brand, cost, and reputation, which prevented teams from confidently shipping to production.
This made the problem clear. Adoption was not limited by features; it was limited by clarity, control, and confidence. This diagnosis shaped the direction of the work that followed.
Starting with the end experience,
not the builder
To address uncertainty, I shifted the focus away from the dashboard and towards the customer’s end user. Instead of iterating on the builder, I reframed the problem around what customers were actually trying to create.
I developed a set of experience concepts, including conversational assistants, contextual prompt suggestions, and intelligent search interactions. These were grounded in real product surfaces and shared with customers to validate desirability.
This repositioned Agent Studio from a configuration tool into an experience-led product. Once we defined what good looked like on the frontend, the builder had a clear direction. This gave the team a clear north star; the builder was no longer driving the experience, it was enabling it.
Defining the product model: agents and experiences
The experience vision exposed a deeper issue. Internally, we lacked a shared understanding of what we were building; “agent” and “experience” were used interchangeably, creating both UX confusion and architectural risk.
I led working sessions with engineering to define the model. We clarified what an agent represented technically, how multiple agents could support a single experience, and which configuration layers could be shared, including LLMs, guardrails, and data connections.
We established a modular structure where experiences are outcome-led and powered by reusable agents beneath the surface. This created alignment across teams and gave us a clear foundation for designing a scalable builder.
Evolving the model through prototyping
To move from definition to execution, I used prototyping to explore how the model would translate into the dashboard.
I created interactive prototypes using a mix of Cursor, Claude Code and Figma MCP to test how agents and experiences would be structured, using them to ground discussions with engineering and surface trade-offs early.
👉 View interactive prototyping hub
As requirements evolved through engineering RFCs, including shifts from multi-agent to single-agent implementations for specific experiences, prototyping allowed us to adapt quickly while maintaining clarity in the user experience.
To support this, I also created a shared prototyping repository, giving engineers direct access to working concepts and enabling faster iteration between design and implementation.
Making the value visible
With the model defined and tested, the next challenge was making the value clear to customers. I restructured the template gallery around outcome-driven experiences, framing templates as real use cases rather than individual features.
Working with engineering, we explored generating live scenarios using real datasets, but hit limitations; LLM output quality was too dependent on the data, making responses unreliable for demonstration.
To address this, I created playable previews that allowed customers to step through key scenarios. I did this using Cursor and the Figma MCP inside my prototyping hub. By simulating responses, I could clearly demonstrate intended behaviours without relying on inconsistent data.
This made the experience tangible, reducing uncertainty and giving customers the confidence to move forward.
Shaping the agent builder to support real experiences
Once customers understood what to build, the next challenge was ensuring the builder could support these experiences in practice. I explored how the builder needed to adapt for more complex patterns, particularly agentic search, rethinking how configuration is structured, how entry points are defined, and how behaviours are tested.
In parallel, I introduced a quick start experience to reduce time to value, prioritising essential decisions while progressively introducing more advanced configuration; shared settings remained reusable, while experience-specific controls were contextual.
Through prototyping, I tested how these concepts could work end-to-end within the builder, ensuring the system remained flexible without becoming overwhelming. This extended the builder into a more adaptable system, capable of supporting a wider range of agent experiences.
Impact & reflection
This work repositioned Agent Studio from a technical configuration tool into an experience-led builder, improving clarity, control, and speed to value across the product.
Following these changes, production adoption increased from 13 to 24 customers (+85%), indicating a clear shift from experimentation to live deployment. Usage also scaled significantly, with completions and searches increasing by over 10x during the same period (January to March), resulting in ARR of ~$2.07 million.
While these results were influenced by multiple factors, they suggest that reducing uncertainty and increasing confidence played a key role in driving adoption.
This work reinforced the importance of designing beyond the interface. Adoption was not solved through UI improvements alone, but by aligning vision, architecture, and onboarding into a coherent system. It shaped how I approach complex products; starting with the experience, defining the model early, and using prototyping to drive alignment.
More projects
Gousto
Homepage redesign
Redesign of the Gousto homepage to best showcase our choice and variety. Result was a 4.8% increase in overall sign-up conversion rate, generating £4.2 million profit over a year from sign-up.
