Human-Agent-Experience
© 2025 Outshift. All Rights Reserved.
© 2025 Outshift. All Rights Reserved.
HAX Research Laboratory
HAX Research Laboratory
We focus on creating agents that participate within human thinking processes instead of functioning as independent tools. These agents act as cognitive partners that help articulate, organize, and extend ideas, supporting users as they reason, explore, and make sense of complex problems. The intent is not to automate human cognition, but to amplify it—enabling richer forms of understanding and collaboration.
Agents must have built-in tools for users to stay in control. Whether it’s logging, monitoring, or intervention mechanisms, we prioritize designs that reinforce user autonomy, cut cognitive overload, and maintain trust.
Agents must have built-in tools for users to stay in control. Whether it’s logging, monitoring, or intervention mechanisms, we prioritize designs that reinforce user autonomy, cut cognitive overload, and maintain trust.
Mistakes are inevitable—especially in AI. Whether through version history, explicit undo mechanisms, or checkpoints, we provide users with ways to heal, rewind, and move forward with agility, without fear of losing their work.
Agents should act like teammates: We emphasize fluid, transparent collaboration where the agent collaborates like a teammate, the human adds their own intuition and judgment, creating a symbiotic relationship where both parties contribute their unique strengths.
From reports and version control, to the ability for the user to track the ‘what’ and ‘why’ behind every decision, our systems support complete transparency whether through action references, continuous input/output logging, or data verification flows.
A comprehensive 6-step process that helped us transform raw research into validated, production-ready design solutions.
Research Plan, Scooping Guide.
Evidence Map, Scooping Guide.
Patterns, Opportunities.
Designing Tomorrow's Worlds.
Prototypes, Consequence Maps.
Scenario Matrix, Action Plan.
Define the phenomenon that you want to explore, not the feature or product.
"How might X change the way people Y in a world where Z is true?"
"How do operators build trust in agent decisions during incident response?"
Deliverable
Build a grounded understanding of what's already happening and what's starting to happen.
Deliverable
Cluster what you observed into repeated patterns:
Deliverable
Turn opportunities into designable questions:
“How might we surface agent reasoning in a way that feels accountable, not overwhelming based on what’s coming?”
Deliverable
Before polished UI, simulate the interaction loop / workflow /sensation:
Key question: does this change the user's behavior or feeling in the way we predicted in step 4?
Deliverable
Translate speculative insights into present-day opportunities, constraints, and design levers.
Deliverable
Real world applications of our design principles across different domains and contexts, demonstrating measurable impact and user-centered outcomes.
Building transparent AI systems that assess infrastructure changes, verify modifications, conduct automated testing, and manage approval workflows - all while maintaining clear visibility into agent decision-making and human oversight.
Problem
Infrastructure changes carry high risk, but manual impact assessment, testing, and approval processes create bottlenecks. Organizations struggle to balance automation speed with safety and accountability, often lacking visibility into what AI agents are actually evaluating and why certain changes get flagged.
Design principles applied:
Creating intuitive interfaces for AI-assisted code composition that understand semantic structure, suggest contextual improvements, and integrate seamlessly into developer workflows without disrupting creative flow or imposing unwanted suggestions.
Problem
Developers need intelligent code assistance that understands semantic context and project architecture, but existing tools either interrupt flow with irrelevant suggestions or fail to capture the nuanced intent behind code structures. Balancing proactive AI help with developer autonomy remains a core challenge.
Design principles applied:
Designing command centers where multiple AI agents monitor threats, coordinate responses, and present unified security intelligence—ensuring human operators understand agent actions, can override decisions, and maintain situational awareness during critical incidents.
Problem
Security operations require rapid threat detection and response, but multiple specialized AI agents create information overload and coordination challenges. Operators struggle to understand why agents flag certain events, how they coordinate actions, and when to trust automated responses versus manual intervention during high-stakes incidents.
Design principles applied:
