Should Agent Instructions Be Visible? [Champions Slack Insider]

[#1]

A discussion started when @Rob Hean shared a screenshot showing a message many users hadn't noticed before:

That simple notice sparked a surprisingly interesting conversation:

Should AI agent instructions be visible or hidden?

The answer depends on who built the agent and why.

Short Answer

Neither approach is universally right.

• For internally developed agents, transparency often builds trust and supports governance.
• For commercial or proprietary agents, hidden instructions may be necessary to protect intellectual property and prevent misuse.

The challenge is balancing transparency, security, and ownership.

Why Some Agent Instructions Are Hidden

As @Ciara Twomey Nielsen pointed out, Marketplace vendors and software developers often invest significant time creating and refining agent instructions. In many cases, those instructions are part of the product itself. Making them publicly visible could expose:

• Proprietary workflows

• Business logic

• Specialized prompting techniques

• Competitive advantages

From that perspective, hiding instructions is no different than protecting application source code. The instructions become part of the intellectual property.

Why Some Customers Want Visibility

The opposing viewpoint raised an equally valid concern. Organizations increasingly use AI agents to:

• Access business information

• Guide decisions

• Perform actions

• Support employees and customers

When instructions are hidden, users may reasonably ask:

What is this agent actually being told to do?

This becomes particularly important in environments with:

• Security requirements

• Compliance obligations

• Internal governance policies

• Industry regulations

Trusting an agent is easier when you can understand how it was designed.

The Governance Question

This discussion ultimately becomes less about technology and more about governance. Most organizations already accept that:

• Application source code is often hidden

• Vendor algorithms are rarely exposed

• Commercial software contains proprietary logic

However, AI introduces a new wrinkle. Instructions directly influence behavior. That means organizations may increasingly want visibility into:

• What actions an agent can take

• What data it can access

• What rules guide its responses

• How it handles sensitive situations

Those questions don't necessarily require exposing every prompt, but they do require some level of transparency.

What About Internal Agents?

The conversation becomes even more interesting for internally developed agents. Unlike Marketplace applications, these agents are often created by employees for employees. In those situations, organizations may choose to make instructions visible because doing so:

• Builds trust

• Improves collaboration

• Supports auditing

• Makes troubleshooting easier

• Helps teams learn from one another

The right answer may differ from one organization to another.

A Naming Clarification

The discussion also surfaced some terminology confusion. Forge is the development platform. Developers can use Forge to build many different types of applications, including AI-powered agents.

Rovo Agents can be created through Studio or built programmatically using Forge. When a Rovo Agent is built through Forge, its instructions can be hidden in ways that are not available for agents created directly through the Studio interface.

Champion Takeaway

The most interesting part of this discussion wasn't whether hidden instructions are good or bad. It was realizing that AI instructions are becoming a new form of organizational knowledge and intellectual property.

As AI adoption grows, organizations will increasingly need to decide:

When should agent behavior be transparent, and when should it be protected?

The answer will likely depend on whether the goal is governance, trust, security, collaboration, or competitive advantage. In reality, most organizations will need some combination of all five.