Societal Implications: The Bigger Picture

This project sits at an inflection point for AI in work. The decisions we make now-about how AI is integrated into jobs, how much agency humans retain, how trust is built-will shape how workers experience AI for years.

What This Means for Workers

• The Risk: "AI-assisted" can become a euphemism for "AI-monitored." If agents feel the copilot is tracking their mistakes, second-guessing their decisions, or quietly replacing their expertise with automation, the experience shifts from "support" to "surveillance."

• The Design Answer: Transparency about what the AI is doing (context retrieval, suggestion ranking, confidence scoring) makes it a tool, not a judge. The agent sees the reasoning, not just the output. They maintain control, not just autonomy theater.

• Real Impact: Agents in our research reported feeling more confident, not less. The AI handled drudgework (reading 40-message threads) so they could focus on skill work (empathy, judgment, tone). That's not job displacement-it's job elevation. And workers know the difference.

​

What This Means for Trust in AI

We're in a trust deficit. People have seen AI fail, disappoint, and surprise them. Support agents, especially, have been burned by opaque AI systems that made bad suggestions but gave no explanation.

• The Design Answer: Build trust through honesty about limitations. When confidence is low, say so. When the AI doesn't know, defer visibly. Show reasoning, not just output. The copilot isn't trying to convince agents it's always right-it's trying to show it knows when it might be wrong.

• This is counter to a lot of AI product thinking, which optimizes for "just make it work." But in human-critical domains (healthcare, customer service, hiring), honesty about uncertainty is more valuable than false confidence.

• Real Impact: Agents who see the AI admit uncertainty trust it more on the wins. They adopt suggestions faster. They feel in control. That's not a bug in the design-it's the whole point.

​

What This Means for the Future of Human Work

This project is fundamentally about this question: What if we designed AI systems that made human judgment more valuable, not less?

In most automation narratives, the trajectory is:

Human does work → AI learns the work → AI does the work → Human is redundant.

But human-centered AI can flip that:

• AI handles the mechanical, repetitive parts (reading long conversations, ranking similar cases)

• Humans handle the judgment, empathy, nuance, accountability parts (deciding how to respond, when to escalate, how to address underlying customer needs)

• The combination is more powerful than either alone, and the human becomes more critical, not less

This doesn't prevent all job displacement. But it's a fundamentally different model than "automate the human out."

The Bigger Stake: If we can prove this model works-that AI + human judgment beats AI alone or human alone-it becomes a template for other knowledge work. That matters for millions of people whose jobs are at the intersection of information work and human judgment.

Critical Design Stance

This copilot succeeds only if it makes agents feel more capable, not more monitored. That requires building trust through:

• Radical transparency about what the AI is doing

• Honest uncertainty about what it doesn't know

• Real control (not control theater) over every decision

• Respect for the agent's expertise as the final arbiter

• ​

If any of those are missing, the tool becomes a tool for optimization that happens to workers, not with them. And that's when AI in work becomes something to resist, not embrace.