With traditional software, products were built on determinism. You shipped a feature, QA tested it, and expected it to behave the same way tomorrow as it did yesterday. It was a static system.
AI broke this model.
AI features are probabilistic, not deterministic. Which means you need to learn in production, with real customers and their real data, not just in staging.
AI products need constant tuning. Updating prompts, adjusting model settings, token rate limits, and more. Every adjustment can’t become a multi-day engineering ticket.
And as AI helps speed up development, we need a way to safely target, release, rollout, and manage those features to customers.
Your product is no longer static. It's alive.
Introducing Reforge Launch
Today, we're introducing Reforge Launch - AI-native feature flags and configuration.
Launch is the fourth product in our suite for AI-native product teams.
We started with Reforge Insights to aggregate and analyze customer feedback.
Added Reforge Research to capture new insights through AI interviews and surveys.
Then recently released Reforge Build, AI prototyping for product teams.
Launch completes the cycle, enabling builders to go from insight to launch.
Reforge Launch treats your product as a living system. You can test new models with specific customers. Adjust prompts without deployments. Roll out AI features gradually. See exactly what's working where. All at fundamentally lower cost than alternatives.
AI-Native Feature Flags & Dynamic Configuration
Pre-AI: Traditional feature flags were binary. A feature was on or off. Engineers wrote the flags, QA tested them, and PMs filed tickets for changes. That made sense when software was deterministic—same input, same output.
The Problem: AI products need constant tuning. Updating prompts, adjusting model settings, token rate limits, and more. Every adjustment can’t become an engineering ticket. The right values can vary by customer, task, or context. You’re no longer flipping a switch—you’re tuning a system.
The Solution: Reforge Launch provides feature flags and configuration for an AI world:
Agent-ready: AI coding assistants can automatically wrap new features in flags. No manual wiring required.
Self-cleaning: Automatically detect stale flags. Retire or convert them into proper configurations.
Custom objects: Store more than booleans. Full objects such as prompts, model parameters, or experiments—in type-safe JSON.
Templated content: Enable personalized prompts, dynamic messages, or quick CMS-style content updates. Marketing, product, and engineering can all iterate safely without deploys.
What You Can Do With Launch
Prompt management: Store prompts in Reforge Launch, not your codebase. Refine prompts in real-time. Customize prompts for specific accounts. No engineering required.
Gradual model migration: Roll out a new model to 1 % of users. Track behavior. Expand or revert instantly.
Feature management: Tailor features, AI capabilities or messages by region, plan, or user persona, or specific customer segments.
Developer Experience
None of this matters if it's painful to use. Reforge Launch is designed for developers who expect their tools to be as intelligent as the AI features they're building, while giving product managers real-time control over the AI experience customers actually see.
Type-safe SDKs for your stack: Native libraries for Python, Node.js, Ruby, Java, Go, and React — with real, generated TypeScript types. Launch automatically generates type-safe code from your configuration, so your editor knows every flag, every config key, and every schema. Never mistype a flag name again. Never ship a missing JSON key.
MCP-powered editors: The Model Context Protocol (MCP) lets your coding environment talk directly to Launch. In Claude Code, Cursor, or any AI-powered IDE, your assistant can query flags, generate charts of evaluation data, read audit logs, find stale flags marked for cleanup, and even open a pull request to remove them automatically. Your AI tools finally understand your configuration as deeply as you do.
CLI for automation:
reforge flag create,reforge config set,reforge rollout. Script your deployments. Let your CI/CD pipeline manage rollouts. Your AI coding assistant can interact with flags programmatically, not just through code comments.Simple integration: Replace hardcoded values like
const ModelTemp = 0.3with fully typed accessors such asReforge.llmWeatherBot.temperatureor templated calls likeReforge.llmWeatherBot.systemMessage({ city: "Boston" }).
Accelerating Product Velocity
The companies winning with AI are the ones that adapt fastest.
When Anthropic releases a new Claude model, can you test it with beta users tomorrow or do you need to wait for next month’s release? When a customer reports your AI giving incorrect answers, can you adjust its behavior while they're still on the phone? When OpenAI changes pricing, can you instantly update token limits to maintain margins?
Building with AI is like building a living system. They evolve. They drift. They surprise us. Our tools must enable us to be equally dynamic.
Getting Started
Reforge Launch is available today. You can get started for free.
With traditional software, products were built on determinism. You shipped a feature, QA tested it, and expected it to behave the same way tomorrow as it did yesterday. It was a static system.
AI broke this model.
AI features are probabilistic, not deterministic. Which means you need to learn in production, with real customers and their real data, not just in staging.
AI products need constant tuning. Updating prompts, adjusting model settings, token rate limits, and more. Every adjustment can’t become a multi-day engineering ticket.
And as AI helps speed up development, we need a way to safely target, release, rollout, and manage those features to customers.
Your product is no longer static. It's alive.
Introducing Reforge Launch
Today, we're introducing Reforge Launch - AI-native feature flags and configuration.
Launch is the fourth product in our suite for AI-native product teams.
We started with Reforge Insights to aggregate and analyze customer feedback.
Added Reforge Research to capture new insights through AI interviews and surveys.
Then recently released Reforge Build, AI prototyping for product teams.
Launch completes the cycle, enabling builders to go from insight to launch.
Reforge Launch treats your product as a living system. You can test new models with specific customers. Adjust prompts without deployments. Roll out AI features gradually. See exactly what's working where. All at fundamentally lower cost than alternatives.
AI-Native Feature Flags & Dynamic Configuration
Pre-AI: Traditional feature flags were binary. A feature was on or off. Engineers wrote the flags, QA tested them, and PMs filed tickets for changes. That made sense when software was deterministic—same input, same output.
The Problem: AI products need constant tuning. Updating prompts, adjusting model settings, token rate limits, and more. Every adjustment can’t become an engineering ticket. The right values can vary by customer, task, or context. You’re no longer flipping a switch—you’re tuning a system.
The Solution: Reforge Launch provides feature flags and configuration for an AI world:
Agent-ready: AI coding assistants can automatically wrap new features in flags. No manual wiring required.
Self-cleaning: Automatically detect stale flags. Retire or convert them into proper configurations.
Custom objects: Store more than booleans. Full objects such as prompts, model parameters, or experiments—in type-safe JSON.
Templated content: Enable personalized prompts, dynamic messages, or quick CMS-style content updates. Marketing, product, and engineering can all iterate safely without deploys.
What You Can Do With Launch
Prompt management: Store prompts in Reforge Launch, not your codebase. Refine prompts in real-time. Customize prompts for specific accounts. No engineering required.
Gradual model migration: Roll out a new model to 1 % of users. Track behavior. Expand or revert instantly.
Feature management: Tailor features, AI capabilities or messages by region, plan, or user persona, or specific customer segments.
Developer Experience
None of this matters if it's painful to use. Reforge Launch is designed for developers who expect their tools to be as intelligent as the AI features they're building, while giving product managers real-time control over the AI experience customers actually see.
Type-safe SDKs for your stack: Native libraries for Python, Node.js, Ruby, Java, Go, and React — with real, generated TypeScript types. Launch automatically generates type-safe code from your configuration, so your editor knows every flag, every config key, and every schema. Never mistype a flag name again. Never ship a missing JSON key.
MCP-powered editors: The Model Context Protocol (MCP) lets your coding environment talk directly to Launch. In Claude Code, Cursor, or any AI-powered IDE, your assistant can query flags, generate charts of evaluation data, read audit logs, find stale flags marked for cleanup, and even open a pull request to remove them automatically. Your AI tools finally understand your configuration as deeply as you do.
CLI for automation:
reforge flag create,reforge config set,reforge rollout. Script your deployments. Let your CI/CD pipeline manage rollouts. Your AI coding assistant can interact with flags programmatically, not just through code comments.Simple integration: Replace hardcoded values like
const ModelTemp = 0.3with fully typed accessors such asReforge.llmWeatherBot.temperatureor templated calls likeReforge.llmWeatherBot.systemMessage({ city: "Boston" }).
Accelerating Product Velocity
The companies winning with AI are the ones that adapt fastest.
When Anthropic releases a new Claude model, can you test it with beta users tomorrow or do you need to wait for next month’s release? When a customer reports your AI giving incorrect answers, can you adjust its behavior while they're still on the phone? When OpenAI changes pricing, can you instantly update token limits to maintain margins?
Building with AI is like building a living system. They evolve. They drift. They surprise us. Our tools must enable us to be equally dynamic.
Getting Started
Reforge Launch is available today. You can get started for free.