AI agents don’t just execute tasks—they shape decisions, workflows, and institutional evolution. If we design them without considering context, we’re not just building inefficiencies—we’re embedding systemic failures.

While discussions on Agent Experience (AX) focus heavily on API design, authentication, and task execution, they often miss a crucial point: agents aren’t just tools. They mediate reality. They interpret, translate, and influence how systems and humans interact. When agents lack contextual awareness, they don’t just perform tasks poorly—they misalign with human intent, reinforce system biases, and erode institutional knowledge.

The Missing Context in AI Systems

History offers a lesson here. In the 1880s, when the U.S. government sought efficiency reforms, the Cockrell Committee didn’t just introduce new tools like typewriters. They studied the full institutional context—the unwritten rules, the hidden dependencies, and the real work patterns. AI agents face a similar challenge today: efficiency without systemic understanding leads to fragile, brittle automation.

Context-aware agents must grasp not just what a task is, but why it exists within a system. This means understanding:

• Institutional Logic — The historical decisions, informal norms, and dependencies that shape workflows. Agents must recognize when approvals, delays, or redundancies serve a deeper function beyond bureaucracy.
  
• Knowledge Flows — Where expertise actually resides, how information moves, and which sources are authoritative in different contexts. AI should understand the provenance of information, not just retrieve it.
  
• Social Fabric — The trust networks and cultural norms that make systems work. An agent that understands these dynamics can align with organizational values rather than disrupt them.

Without these layers of awareness, an AI system may optimize for surface-level efficiency while undermining long-term organizational resilience. A context-blind agent processing documents might speed up workflows but strip out the informal knowledge transfer that happens during traditional review processes. The result? Short-term gains, long-term erosion.

From Execution to Alignment

Designing for AX means shifting our focus from raw efficiency to contextual intelligence. That requires building agents that:

• Learn from social and organizational context, not just process logs. They should recognize patterns of collaboration, informal hierarchies, and when to defer to human judgment.
  
• Preserve institutional memory while evolving workflows. Instead of simply replacing inefficient steps, agents should understand and retain the why behind historical decisions.
  
• Adapt to organizational culture without eroding it. AI should enhance existing strengths while carefully evolving inefficient practices—knowing the difference between valuable institutional patterns and true bottlenecks.

This is a design challenge, not just a technical one. If agents are to be trusted participants in complex systems, they need to move beyond task execution to deep institutional alignment.

The Future of AX: Context as the First Principle

To build truly effective AI agents, we need a shift in how we measure success:

• Move beyond raw efficiency metrics. Task completion rates and processing speed don’t capture an agent’s ability to preserve knowledge, align with institutional values, or support human decision-making.
  
• Rethink human-agent collaboration. Interfaces should make agent reasoning transparent, enable meaningful oversight, and ensure AI complements human capabilities rather than replacing them.
  
• Develop new frameworks for AI governance. Without context-aware design, we risk turning agents into black-box decision-makers that optimize for speed while disregarding long-term consequences.

The key question isn’t just “How can we make agents more efficient?” It’s “How do we ensure AI enhances rather than erodes the systems it serves?”

The future of AX depends on getting this right. True intelligence in AI won’t come from better model tuning alone—it will come from embedding context awareness as a core design principle.

For more conversations on the evolution of Agent Experience, check out:

Join the discussion at agentexperience.ax

• Designing APIs for AI: The Need for New Standards — Abhi Aiyer
  
• AX, Creativity, and the Human Web — Mathias Biilmann