![Cover Image for [Hands-on Workshop] Let's Get Dangerous: Giving Customers "Root Access" to the System Prompt](https://images.lumacdn.com/cdn-cgi/image/format=auto,fit=cover,dpr=2,background=white,quality=75,width=400,height=400/event-covers/di/68e877e1-c6bc-4dad-821e-fba3f0be1248.png)
![Cover Image for [Hands-on Workshop] Let's Get Dangerous: Giving Customers "Root Access" to the System Prompt](https://images.lumacdn.com/cdn-cgi/image/format=auto,fit=cover,dpr=2,background=white,quality=75,width=400,height=400/event-covers/di/68e877e1-c6bc-4dad-821e-fba3f0be1248.png)
[Hands-on Workshop] Let's Get Dangerous: Giving Customers "Root Access" to the System Prompt
Every engineering instinct says: lock down the prompt. Sanitize inputs, hide the system prompt, never let customers touch it. But real systems break under that rule.
Customers need control over data shapes, constraints, and formatting. That means write access to prompts and real risk.
This workshop is about engineering that risk away: treating prompts as a compilation target, turning messy intent into validated schemas, executable constraints, and feedback loops that keep failures visible.
KEY TOPICS
The Prompt Security Paradox: Why locking down prompts kills flexibility and how to give customers control without chaos
Prompts as Compilation Targets: Treating prompts like code with validation, schemas, and structured constraints
Risk Engineering: Building guardrails that let customers customize while keeping the system safe
Practical Prompt Validation: Turning messy customer intent into executable, reliable agent behavior
Failure Visibility: Designing feedback loops that surface problems before they reach production
EXPECTED OUTCOMES
You'll leave with:
A practical framework for safely exposing prompt customization to customers
Hands-on experience building validated prompt schemas and constraints
Mental models for evaluating when to lock down vs. open up prompt access
Strategies for balancing customer control with system reliability
PRE-WORKSHOP REQUIREMENTS
Hardware: Laptop with internet access and a development environment set up
Software: AI coding tool (Claude Code, Cursor, Copilot, or similar) installed and authenticated. Familiarity with basic CLI workflows is helpful.
Materials: No prior experience required. Come ready to explore how to turn risky prompt access into reliable, customer-controlled systems.
WORKSHOP ACCESS & ELIGIBILITY
This workshop is exclusively available to attendees of the main event and cannot be registered for separately. Workshop access is offered on a first come, first served basis with limited seats available.