OVERVIEW
Making AI work visible, reviewable, and safe.
Making AI work visible, reviewable, and safe.
Making AI work visible, reviewable, and safe.
A supervised AI-operations concept that shows the agent’s plan, permissions, uncertainty, approvals, and audit trail in one accountable workflow.
A supervised AI-operations concept that shows the agent’s plan, permissions, uncertainty, approvals, and audit trail in one accountable workflow.
ROLE
Product Designer
COMPANY
Self-initiated
FOCUS
AI Operations UX
YEAR
2026
01 / OVERVIEW
A concept for supervised AI work
A concept for supervised AI work
A concept for supervised AI work
Blueberry explores a common high-stakes moment in customer operations: a service outage demands fast action, yet a support manager remains accountable for what an AI agent changes or tells customers. The concept connects task creation, plan review, scoped access, live activity, human approval, and an audit trail. It is a self-initiated design project and clickable prototype—not a shipped product or a claim of tested operational performance.
Blueberry explores a common high-stakes moment in customer operations: a service outage demands fast action, yet a support manager remains accountable for what an AI agent changes or tells customers. The concept connects task creation, plan review, scoped access, live activity, human approval, and an audit trail. It is a self-initiated design project and clickable prototype—not a shipped product or a claim of tested operational performance.
Blueberry explores a common high-stakes moment in customer operations: a service outage demands fast action, yet a support manager remains accountable for what an AI agent changes or tells customers. The concept connects task creation, plan review, scoped access, live activity, human approval, and an audit trail. It is a self-initiated design project and clickable prototype—not a shipped product or a claim of tested operational performance.
02 / CONTEXT
Delegation is hard when intent is opaque
Delegation is hard when intent is opaque
Delegation is hard when intent is opaque
AI can accelerate response work, but broad permissions and invisible reasoning create risk. A manager needs to know what the agent intends to do, what systems it can touch, when it has encountered uncertainty, and which actions will require human judgment. The problem is not simply how to display more system status; it is how to make accountability understandable without forcing someone to supervise every trivial step.
AI can accelerate response work, but broad permissions and invisible reasoning create risk. A manager needs to know what the agent intends to do, what systems it can touch, when it has encountered uncertainty, and which actions will require human judgment. The problem is not simply how to display more system status; it is how to make accountability understandable without forcing someone to supervise every trivial step.
AI can accelerate response work, but broad permissions and invisible reasoning create risk. A manager needs to know what the agent intends to do, what systems it can touch, when it has encountered uncertainty, and which actions will require human judgment. The problem is not simply how to display more system status; it is how to make accountability understandable without forcing someone to supervise every trivial step.
03 / DESIGN PRIORITIES
Define the trust contract before the UI
Define the trust contract before the UI
Define the trust contract before the UI
The design priorities follow the decisions that carry consequence. First, preview intent before execution. Second, match access to the specific task rather than grant blanket permission. Third, pause at moments of uncertainty or external impact. These are principles for the concept, not measured outcomes from a live service.
The design priorities follow the decisions that carry consequence. First, preview intent before execution. Second, match access to the specific task rather than grant blanket permission. Third, pause at moments of uncertainty or external impact. These are principles for the concept, not measured outcomes from a live service.
The design priorities follow the decisions that carry consequence. First, preview intent before execution. Second, match access to the specific task rather than grant blanket permission. Third, pause at moments of uncertainty or external impact. These are principles for the concept, not measured outcomes from a live service.
01
Preview the plan
Show sequence, evidence, and stopping points before the agent begins.
02
Scope access
Explain the exact resources and permissions requested for this task.
03
Pause for judgment
Escalate consequential choices to a reviewer with enough context to decide.
04 / APPROACH
Map risk to the moments that matter
Map risk to the moments that matter
Map risk to the moments that matter
I began with an outage-response scenario and mapped what a manager would need to predict, inspect, approve, or reverse. That led to a workflow from command center to task creation, plan preview, permissions review, live activity, uncertainty handling, grouped approvals, completion, and audit. The interface avoids a generic AI dashboard: each stage answers a concrete operator question, such as what will happen next, what the agent can touch, whether it is still on track, and who approved an action.
I began with an outage-response scenario and mapped what a manager would need to predict, inspect, approve, or reverse. That led to a workflow from command center to task creation, plan preview, permissions review, live activity, uncertainty handling, grouped approvals, completion, and audit. The interface avoids a generic AI dashboard: each stage answers a concrete operator question, such as what will happen next, what the agent can touch, whether it is still on track, and who approved an action.
I began with an outage-response scenario and mapped what a manager would need to predict, inspect, approve, or reverse. That led to a workflow from command center to task creation, plan preview, permissions review, live activity, uncertainty handling, grouped approvals, completion, and audit. The interface avoids a generic AI dashboard: each stage answers a concrete operator question, such as what will happen next, what the agent can touch, whether it is still on track, and who approved an action.
Blueberry concept screens · use the arrows or swipe to explore
Blueberry concept screens · use the arrows or swipe to explore
05 / SYSTEM
Make control visible without constant interruption
Make control visible without constant interruption
Make control visible without constant interruption
Plan previews expose intent before execution. Task-scoped permissions name the resource, access level, purpose, and potential consequence in plain language. A live timeline keeps evidence and uncertainty in view while work runs. Approvals are grouped around consequential decisions instead of scattered as arbitrary pop-ups. The audit trail records what changed and who approved it; an undo path makes a completed action less opaque.
Plan previews expose intent before execution. Task-scoped permissions name the resource, access level, purpose, and potential consequence in plain language. A live timeline keeps evidence and uncertainty in view while work runs. Approvals are grouped around consequential decisions instead of scattered as arbitrary pop-ups. The audit trail records what changed and who approved it; an undo path makes a completed action less opaque.
Plan previews expose intent before execution. Task-scoped permissions name the resource, access level, purpose, and potential consequence in plain language. A live timeline keeps evidence and uncertainty in view while work runs. Approvals are grouped around consequential decisions instead of scattered as arbitrary pop-ups. The audit trail records what changed and who approved it; an undo path makes a completed action less opaque.
06 / MY ROLE
A self-directed product design exploration
A self-directed product design exploration
A self-directed product design exploration
I defined the scenario, information architecture, interaction model, visual system, and clickable prototype. The project allowed the interface to be designed as one connected workflow rather than isolated screens. Because it was self-initiated, the case study does not attribute the work to a client team, product launch, or user-research result that the source material does not document.
I defined the scenario, information architecture, interaction model, visual system, and clickable prototype. The project allowed the interface to be designed as one connected workflow rather than isolated screens. Because it was self-initiated, the case study does not attribute the work to a client team, product launch, or user-research result that the source material does not document.
I defined the scenario, information architecture, interaction model, visual system, and clickable prototype. The project allowed the interface to be designed as one connected workflow rather than isolated screens. Because it was self-initiated, the case study does not attribute the work to a client team, product launch, or user-research result that the source material does not document.
07 / OUTCOMES
A testable design direction—not a launch claim
A testable design direction—not a launch claim
A testable design direction—not a launch claim
The concept demonstrates how supervised AI operations might remain understandable and controllable when a manager delegates a time-sensitive task. It has not been shipped or user-tested. The meaningful deliverable is a coherent interaction model and a prototype that could be evaluated against real operator behavior before any claim of effectiveness is made.
The concept demonstrates how supervised AI operations might remain understandable and controllable when a manager delegates a time-sensitive task. It has not been shipped or user-tested. The meaningful deliverable is a coherent interaction model and a prototype that could be evaluated against real operator behavior before any claim of effectiveness is made.
The concept demonstrates how supervised AI operations might remain understandable and controllable when a manager delegates a time-sensitive task. It has not been shipped or user-tested. The meaningful deliverable is a coherent interaction model and a prototype that could be evaluated against real operator behavior before any claim of effectiveness is made.
REFLECTION
Friction should follow consequence.
Friction should follow consequence.
Friction should follow consequence.
The next step would be scenario-based evaluation: can an operator predict the agent’s next action, notice uncertainty, intervene effectively, identify who approved a send, and reverse it if necessary? Plan, permission, and approval reviews should be more deliberate for high-consequence work, not imposed identically on every task.
The next step would be scenario-based evaluation: can an operator predict the agent’s next action, notice uncertainty, intervene effectively, identify who approved a send, and reverse it if necessary? Plan, permission, and approval reviews should be more deliberate for high-consequence work, not imposed identically on every task.
The next step would be scenario-based evaluation: can an operator predict the agent’s next action, notice uncertainty, intervene effectively, identify who approved a send, and reverse it if necessary? Plan, permission, and approval reviews should be more deliberate for high-consequence work, not imposed identically on every task.
THAT’S THE STORY
Trust comes from visible control.
Trust comes from visible control.
Trust comes from visible control.
KEEP EXPLORING
More work to explore.
More work to explore.
01 /
Bugasura
Product design
A clearer way for teams to track and resolve bugs.
↗
02 /
My ESG Planet
Enterprise UX
Making complex ESG reporting easier to navigate.
↗
03 / YOU’RE HERE
Blueberry
AI product design
Making AI work visible, reviewable, and safe.
↗
04 /
Delicacy
Mobile experience
Making balanced meal planning feel simpler.
↗
Let’s make something worth remembering.
Let’s make something worth remembering.
Amal brings the ideas, the craft, and the obsessive attention to detail. I bring moral support and occasional interruptions. Between us, you’re in good hands.
Amal brings the ideas, the craft, and the obsessive attention to detail. I bring moral support and occasional interruptions. Between us, you’re in good hands.
© 2026 Amal K Saji · Designed with care. Supervised by Luca.