Hi, I'm Nishtha.

I lead design and research at Workato, where I'm building the platform that gives a flock of unpredictable AI agents their rules and shape.

Strategic leadership

We almost split the product in two

Leading design through a product split, a strategic reversal, and a compressed launch; then using research findings to reshape how enterprise agents are built, tested, and governed.

Product incubation

MCP tools enter their UI era

Moving quickly outside the roadmap, I led design to take MCP servers beyond text. Drawing on lessons from Workbot for Slack, we replaced our no-code direction with a bet on an AI-assisted, code-first approach.

Organization design

Only 2 chose to leave in a decade

Scaling Product Design and Research at Workato from one person to a global team of 22, with the ownership, career paths, and leadership capacity for quality to thrive beyond my direct involvement.

Design direction

Stop babysitting your AI agents

Designing a peek-a-boo bar that keeps a fleet of agents within reach across Workato, so people can supervise, steer, and queue parallel work without leaving their current task.

Hands-on design

A 2019 redesign that still works

Redesigning Workato's Recipe Editor from principles to launch, then examining why its core model held for seven years and what the agentic era demands next.

Co-inventor

AI needs conversational etiquette

Replacing a jittery, ambiguous response stream with a patented conversational cue that shows when AI is still responding and when it is the user's turn.

Some tinkering. A few musings.

5 levels of AI maturity in design

Developing a bird murmuration

Everyone prototypes now

Voice, trust, and machines

Nishtha presenting on stage at a Workato conference

I've been leading Product Design and Research at Workato for the last 10 years. In that time, design has scaled from one person to a 22-person global team. I'm quietly proud that it's a team almost nobody leaves.

Along the way, I've designed the same platform through three eras: integration workflows, copilots, and now agents that act on their own.

I'm formally trained in physical product design, visual communication, psychology and human factors. I've been researching how people learn to trust machines since before that question had an industry attached to it.

I'm looking to build with companies, in software or hardware, where the hard part is making powerful systems people can trust.

LinkedIn comment from Vijay Tella, Co-founder and CEO of Workato: “Nishtha Mehrotra, I'm incredibly grateful. We're so lucky to have you leading the intuitive product experience we're known for. I've spent 30 years in integration middleware, and this tech is complex and whack under the hood. Refactoring six complex platform categories into one while making it seamless for both pro-code integration experts and business teams? No joke. What you and your team have built is one of the great feats of enterprise software design. So much so, you have created a marketing challenge as Workato confounds people. How can something that looks so easy be so powerful? What blows me away is your first-principles thinking and relentless drive to keep attacking the complexity. We don't have real competition for our product experience, productivity, and speed; our biggest challenger is ourselves. You and your team are awesome role models of our core value to keep raising our game. Thank you.”

From Vijay Tella, Workato’s CEO

Stop babysitting your AI agents

Contribution
Design direction
Team
Serene Chen
Timeframe
2026
Outcome
Design under development; patent filing planned
The peek-a-boo agent bar expanded over a Workato page with three active agents

The agent fleet stays within reach while the user continues working elsewhere in the product.

How do you supervise agents without becoming their full-time supervisor?

A single AI assistant can fit neatly inside a chat panel. Managing a fleet of agents is more complex and remains largely unsolved.

People need to assign work, leave agents running, and continue with their own tasks. At the same time, they need to know when an agent finishes, gets stuck, or requires a decision.

Imagine an automation lead starting three agents:

  • One debugging a failed invoice sync in Finance
  • One building a lead-routing workflow in Sales
  • One investigating errors in an employee-onboarding process

Rather than waiting, the automation lead begins building an approval workflow manually. The agents continue working in the background while the lead remains free to check their progress, redirect them, or queue additional work. The automation lead only needs to return to the full agent session when more context is required.

Why is this particularly difficult in Workato?

A Workato user can belong to several workspaces, perhaps one for Sales, another for Legal, and another for Finance. Each workspace may also contain development, test, and production environments.

An agent could therefore be working in a different workspace, environment, or product area from the page the user is currently viewing. If agent supervision lived inside a page or asset-level chat, people would have to remember where each agent started and navigate back to that location to check its progress.

The design challenge was to make agent supervision independent of any asset, page, workspace, or environment. It needed to belong to the user and remain accessible throughout the product.

Sales, Legal, and Finance workspaces with agents operating across development, test, and production environments under one product-wide agent layer

Agents may work in different workspaces and environments, but their supervisory layer stays with the user.

The first concept confused location with attention

An early concept combined agent status with the workspace and environment indicators.

I pushed back on this direction during critique. The environment switcher answers, “Where am I working?” The agent indicator needs to answer, “Does an agent need my attention?” Combining the two suggested that agent activity belonged only to the current environment.

My feedback in Figma
Can AIRO occupy its own layer across the product? If I have agents working in workspace A and workspace B, I shouldn't have to switch workspaces just to see what is happening.
The environment switcher already communicated location. Agent supervision needed a separate, product-wide layer.
An environment switcher with an agent alert, showing that location and attention are two different questions

Two different questions - where the user is working, and whether an agent needs their attention - needed two different indicators.

Making the principle tangible

I used our internal product scaffolding and Claude Code to build a rough prototype. It gave Serene and me something concrete to evaluate and made three interaction principles visible:

  • Running agents should remain accessible throughout the product.
  • Their status shouldn't depend on the current workspace or environment.
  • Anything requiring a decision should be reachable in one action.
Three design principles for the agent layer: persistent access, global status, and one-action entry

The prototype made three principles concrete: persistent access, global status, and one-action entry.

Finding the right edge of the screen

Serene and I explored three placements.

A centered notch

Our first direction used a centered notch at the top of the screen. Inspired by the MacBook notch, it could expand to provide a quick view of active agents.

We liked that it kept agent activity nearby without introducing a permanent panel. However, it collided with an existing control in the recipe editor, where people switch between Edit and Test modes.

A full-width top bar

We then tried placing a bar across the top of the product.

This created more room for active agents, but expanding the bar pushed the page content down and competed with the product's existing navigation. The top of the screen was already carrying too much responsibility.

The bottom of the viewport

Serene moved the interaction to the bottom of the screen. This resolved several problems at once:

  • It avoided the existing navigation and mode controls.
  • It could appear on any page without restructuring that page.
  • It remained in peripheral view without competing with the current task.
  • It created a natural transition into a detailed agent view.
  • The action for starting an agent could sit close to the conversation that followed.
The bottom placement worked consistently across the product while allowing the interface to grow when the user needed more context.
Three agent interface placements compared: centered notch, full-width top bar, and bottom bar

The notch collided with Edit and Test controls. The top bar pushed content down. The bottom bar fit every page without restructuring it.

Every agent, always within reach

The final agent bar stays at the bottom of the viewport throughout Workato. It doesn't belong to the current page, workspace, or environment. It belongs to the user.

From anywhere in the product, people can:

  • See which agents are active
  • Check progress without changing context
  • Respond when an agent needs input
  • Redirect an agent
  • Return to any agent session
  • Queue or start additional work
The bar remains unobtrusive while agents are working, but keeps their activity visible at a glance. When the user asks for more information, it expands into a richer view or opens the complete agent experience. The user controls when agent supervision occupies more of the screen.
I think of this as a “peek-a-boo” pattern: the interface stays at the edge of attention until there is something worth seeing, then recedes when the user returns to their work.
The interaction can move from a quiet status signal to full supervision, depending on how much context the user needs.
The same persistent agent bar on a recipe editor, connector builder, and operations dashboard

The same supervisory pattern follows the user across substantially different Workato pages.

Six states of the agent bar: collapsed, active, needs input, completed, expanded, and full screen

The bar moves from a quiet signal to full supervision only when the user needs more context.

My contribution

I reframed the problem as a product-wide information architecture challenge and redirected an early concept that tied agent activity to the current environment. I established the core interaction principles and built the first prototype to communicate the direction. I also critiqued subsequent concepts and helped guide the design as it developed.

Serene led the detailed interaction design and identified the bottom placement that made the pattern work across the product.