TechPrufer
Get started
Stripe
Stripe logoStripe · Feature trail

Stripe Projects and Agent-based Development

21 posts · Nov 2024 – Sep 2026

What this trail covers

This post continues the evolution of Stripe Projects and Agent-based Development, detailing Stripe's homegrown coding agents, 'Minions,' which are responsible for merging over a thousand pull requests weekly. It highlights how these agents write code from start to finish, with human review, and introduces the second part of this series which will delve deeper into their workings. This post focuses on the technical implementation and architectural details of how these agents operate end-to-end, specifically detailing the creation of a 'Payment Method Factory' that orchestrates agents using over 100 reusable prompts, hardened by observing implementation runs, and kept current by feeding back learnings from each run.

All Stripe trails
  1. Stripe’s Payment Method Factory: Orchestrating agents for repeated, custom integrationsThis post details the technical implementation and architectural details of Stripe's 'Payment Method Factory,' an agentic system for orchestrating repeated, custom integrations. It describes how over 100 reusable prompts are used, how agents observe implementation runs to harden the process, and how learnings from each run are fed back to keep the system current. This builds upon the existing capability of Stripe Projects and Agent-based Development by focusing on a specific application of coding agents for integration workflows.
  2. How Stripe is designing Checkout for AI agentsThis post details how Stripe is designing Checkout for AI agents, introducing the WebMCP (Web Checkout Machine Communication Protocol). This protocol allows AI agents to interact with Checkout programmatically and securely, enabling them to initiate and complete purchases. The post discusses the challenges of enabling agents to perform actions within a browser-based checkout flow and outlines the technical considerations for secure agent interactions with Checkout.
  3. Harbor: Stripe’s AI-assisted prototyping toolThis post introduces Harbor, Stripe's internal AI-assisted prototyping tool. It details how Harbor compiles prototypes in the browser, how the AI agent inspects its own work, and how comments remain anchored when the page is rewritten. This contributes to the broader theme of agent-based development and developer productivity by providing a concrete example of an AI tool used for prototyping.
  4. AI didn't write our SDK. It changed how we built it.This post details how AI agents, specifically 'Minions,' were used in the development of Stripe's extensibility SDK. It emphasizes that AI did not write the SDK but significantly changed the development process by assisting with prototyping, drafting specifications, and providing persistent guidance. The post highlights the iterative process of using AI-generated code, human review, and refinement to bring ideas to a reviewable shape faster, focusing on the workflow and tooling that enabled this accelerated development cycle.
  5. Reading the signals: Helping your AI-built payment integration understand business intentThis post details how Stripe's AI-built payment integrations can better understand business intent by reading signals. It discusses the challenges of translating high-level business logic into concrete payment flows and introduces techniques for enriching AI-generated code with contextual understanding. The author highlights the importance of providing AI agents with more than just low-level instructions, emphasizing the need for them to grasp the 'why' behind a transaction. This involves analyzing various signals, such as customer behavior, transaction history, and business rules, to inform the AI's decision-making process. The post suggests that by effectively interpreting these signals, AI-built integrations can achieve a higher degree of accuracy and alignment with business objectives, moving beyond simple code generation to intelligent automation.
  7. Agentic payments for marketplaces: How a smart fridge can shop safelyThis post introduces the concept of agentic payments for marketplaces, using a smart fridge as an example to illustrate how AI agents can securely handle transactions. It discusses the engineering challenges and potential solutions for enabling agents to initiate and complete payments within marketplace environments, focusing on security and user experience.
  8. Meet Stripe's Knowledge AI PlatformThis post introduces Stripe's Knowledge AI Platform, an AI agent platform designed for diverse non-coding knowledge work. It connects employees to over 1,000 internal tools and skills, enabling secure, enterprise-scale productivity. The platform handles tasks ranging from quick queries to complex, multi-day projects, showcasing a significant advancement in internal AI tooling for knowledge management and task automation.
  9. From hobby to real world: integrating the hard stuff with agentsThis post details the challenges and solutions for integrating AI agents with real-world systems, moving beyond happy path scenarios. It discusses friction points like error handling, state management, and debugging when agents interact with complex services. The author shares practical advice on how to approach these challenges, emphasizing the need for robust error handling and state synchronization mechanisms when building agentic workflows that interact with external APIs and services.
  10. What it feels like building with Stripe ProjectsThis post describes the experience of building a GitHub traffic dashboard using Stripe Projects, highlighting the rapid materialization of infrastructure components like Auth0, Supabase, Vercel, Prisma, deployments, environments, and scheduled jobs within an hour. It emphasizes the shift from 'AI coding' to real infrastructure orchestration and discusses the emerging challenges of AI-assisted infrastructure, including operational visibility, ownership clarity, environment management, and understanding rapidly orchestrated systems.
  11. You can't whisper at an AI agentThis post details experiments with AI agents for steering code generation, focusing on the challenges of prompt engineering and the unpredictable behavior of LLM-based agents in real-world scenarios. It highlights that direct, precise instructions are often ineffective, and that agents require more nuanced guidance and iterative refinement. The authors share learnings from a dozen steering experiments, emphasizing the need for robust testing and understanding of agent limitations.
  12. Learnings from init to deployThis post details the learnings and friction encountered when integrating services with agents, building upon the previous post about Stripe Projects. It focuses on the practical challenges and real-world application of agent-based development, moving beyond the 'happy path' to explore the complexities of integrating various SaaS providers and handling real-world scenarios. The author shares insights gained from the 'init to deploy' lifecycle of agent-based projects.
  13. From init to deploy: building with agents and Stripe ProjectsThis post details how a developer advocate built an internal transcript editing application using Stripe Projects, Vercel, and OpenRouter. It focuses on the practical application of Stripe Projects in orchestrating various SaaS providers to create a functional application, moving beyond theoretical AI coding to real infrastructure orchestration.
  14. Formatting an entire 25 million line codebase overnight: the rubyfmt storyThis post details the development and rollout of rubyfmt, a Rust-based, zero-config, ultra-fast autoformatter, across Stripe's 25 million line Ruby codebase. It highlights a significant advancement in developer productivity and code standardization, demonstrating Stripe's commitment to maintaining code quality and consistency at scale, which complements the broader theme of agent-based development and infrastructure orchestration.
  15. Using AI agents to build Stripe-Salesforce integrations fasterThis post details how AI coding agents, specifically CLI-based ones, can accelerate the building of Stripe-Salesforce integrations. It addresses the challenge of agent hallucination on complex, evolving platform surfaces by introducing open-source grounding context files to flatten the learning curve and reduce errors, enabling developers to ship working integrations faster. The post implicitly discusses the integration challenges between two complex platforms (Stripe and Salesforce) and how AI agents, augmented with grounding context, can mitigate these.
  16. Provision a production-ready dev stack from your terminalThis post introduces the Stripe CLI's capability to provision production-ready development stacks from the terminal. It details how Stripe Projects creates real resources in the user's own cloud provider accounts, securely synchronizes agent-ready credentials, supports upgrades to paid tiers, and handles payments without leaving the terminal. This extends the existing capability by providing a concrete mechanism for developers to set up their entire development environment, including AI tools, through a command-line interface.
  17. Minions: Stripe’s one-shot, end-to-end coding agents—Part 2This post, Part 2, delves deeper into the technical implementation and architectural details of Stripe's 'Minions' coding agents. It focuses on how these agents write code from start to finish, enabling the merging of over a thousand pull requests weekly, with human review.
  18. Minions: Stripe’s one-shot, end-to-end coding agentsThis post introduces Stripe's 'Minions,' one-shot, end-to-end coding agents that merge over a thousand pull requests weekly. It details how these agents write code from start to finish, with human review, and focuses on the technical implementation and architectural details of their operation.
  19. Configuring Stripe using Terraform and AI agentsThis post details the use of AI agents to author Terraform code for configuring Stripe infrastructure. It contrasts this approach with direct API calls, highlighting the benefits of transparency, consistency, and auditability through code review workflows. The post implies a system where AI agents generate infrastructure-as-code, which is then reviewed and applied, offering a safer and more robust method for managing Stripe configurations.
  20. Building financial operations agents with Hypermode and StripeThis post details how to build financial operations agents using Hypermode and Stripe. It explains how to create AI agents that can handle payments, process refunds, generate invoices, and automate complex financial workflows without requiring coding. The post focuses on the practical application of these agents for financial operations.
  21. Stripe’s new AI Assistant in VS CodeThis post introduces Stripe's new AI Assistant in VS Code, detailing its technical implementation. It covers the use of retrieval-augmented generation (RAG) for context-aware code suggestions, the evaluation strategies employed to ensure accuracy and safety, and the measurement approaches used to track its performance and impact on developer productivity. The assistant is designed to integrate seamlessly into the developer workflow, providing intelligent code completions and suggestions.
  22. Adding payments to your LLM agentic workflowsThis post details the integration of Stripe's payment processing capabilities into LLM agentic workflows. It introduces the Stripe Agent Toolkit, which provides tools for agents to interact with Stripe's payment APIs, enabling them to perform financial transactions, manage metered billing, and streamline operations. The post discusses how to handle payment failures and ensure secure transaction processing within agentic systems, and provides examples of how agents can use these tools to automate tasks like creating invoices, processing payments, and managing subscriptions.