MeshMesh
Prompting

Steering Conversations

Guide the agent naturally through interruptions, corrections, and course changes

You're in control of the conversation. Interrupt, redirect, and course-correct at any time. The agent adapts to your steering.

Interrupt Anytime

You don't have to wait for the agent to finish. If you see it going in the wrong direction, stop it.

Wait — before you deploy, let me review the Flow logic
Hold on. I want to change the approach.
Stop. Let's think about this differently.

The agent pauses and waits for your guidance.

Add Context Mid-Conversation

Realize you forgot to mention something? Add it.

Actually, also segment this by region
Oh, I should mention — we have an existing automation
that handles new leads. Don't duplicate that logic.
One more thing: the VIP threshold is $10,000 annual spend

The agent incorporates the new context and adjusts its work.

Ask for Modifications

Not happy with what the agent created? Ask for changes.

That's not quite right. The VIP threshold should be $10,000,
not $1,000
Good start, but make the tone more casual and add emojis
This is too complex. Can you simplify it?
I like the structure but change the subject line to something
more urgent

The agent revises based on your feedback.

Respond to Questions

The agent asks questions when it needs clarification. Answer directly.

Agent: "Should I include historical data or just new records?"

Just new records going forward. We'll handle historical
data in a separate phase.

Agent: "Do you prefer approach A (simpler) or approach B (more flexible)?"

Let's go with A for now. We can always make it more
flexible later if needed.

If you're not sure how to answer, say so. The agent can explain the trade-offs in more detail.

Change Direction Entirely

Changed your mind about the approach? Just say so.

Actually, let's scrap this and try something different.
Instead of a Flow, let's use an Apex trigger.
I've been thinking about this wrong. Let's start over
with a focus on the mobile experience first.

The agent doesn't take it personally. It adapts and moves in the new direction.

Build on Previous Work

The agent remembers the full conversation. Reference earlier work.

Use the same structure as the email we created earlier,
but for the mobile version
Apply the same segmentation logic to this new data extension
Remember the requirements from phase 1? Add those to this plan too.

Ask for Explanations

If you don't understand something, ask.

Why did you choose this approach instead of using a formula field?
Can you explain what this Flow trigger does?
What would happen if we used the other option you mentioned?

The agent explains its reasoning and helps you make informed decisions.

When to Start Fresh

Sometimes it's better to start a new conversation:

  • The context has become cluttered with abandoned directions
  • You want to try a completely different approach from scratch
  • The conversation history is creating confusion

But most of the time, keep steering. The agent tracks context efficiently and can course-correct without starting over.

Next Steps