{"id":3664,"date":"2026-03-19T08:39:29","date_gmt":"2026-03-19T08:39:29","guid":{"rendered":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/2026\/03\/19\/open-swe-captures-the-architecture-that-stripe-coinbase-and-ramp-built-independently-for-internal-coding-agents\/"},"modified":"2026-03-19T08:39:29","modified_gmt":"2026-03-19T08:39:29","slug":"open-swe-captures-the-architecture-that-stripe-coinbase-and-ramp-built-independently-for-internal-coding-agents","status":"publish","type":"post","link":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/2026\/03\/19\/open-swe-captures-the-architecture-that-stripe-coinbase-and-ramp-built-independently-for-internal-coding-agents\/","title":{"rendered":"Open SWE Captures the Architecture That Stripe, Coinbase and Ramp Built Independently for Internal Coding Agents"},"content":{"rendered":"
\"\"<\/div>\n

\"\"<\/p>\n

Stripe built Minions. Ramp built Inspect. Coinbase built Cloudbot. Three engineering organizations, working independently, arrived at similar architectural decisions for their internal AI coding agents. LangChain noticed the convergence and open-sourced the pattern.<\/span><\/p>\n

Open SWE, released March 17, is an open-source framework built on LangChain\u2019s Deep Agents and LangGraph that provides the core architectural components for internal coding agents. The MIT-licensed project isn\u2019t trying to be another AI coding assistant<\/a>. It\u2019s a customizable foundation for organizations that want to build their own \u2014 the way Stripe, Ramp and Coinbase already have.<\/span><\/p>\n

The Convergence<\/b><\/h3>\n

What caught LangChain\u2019s attention was that these independently developed systems share the same architectural decisions. Isolated cloud sandboxes where tasks run with full permissions inside strict boundaries. Curated toolsets \u2014 Stripe reportedly maintains around 500 carefully selected tools. Subagent orchestration where complex tasks get decomposed and delegated to specialized child agents. And integration with existing developer workflows through Slack, Linear, and GitHub rather than requiring engineers to adopt new interfaces.<\/span><\/p>\n

The convergence suggests that production engineering environments have common requirements for AI coding agents. Open SWE codifies those requirements into a framework that other organizations can start from rather than rebuilding from scratch.<\/span><\/p>\n

How it Works<\/b><\/h3>\n

Open SWE composes on the Deep Agents framework rather than forking an existing agent \u2014 similar to how Ramp built Inspect on top of OpenCode. This gives organizations an upgrade path: When Deep Agents improve context management or token efficiency, those improvements flow through without requiring customizations to be rebuilt.<\/span><\/p>\n

Each task runs in its own isolated cloud sandbox \u2014 a remote Linux environment with full shell access. The repository is cloned in, the agent is granted full permissions within that boundary, and errors are contained. Multiple tasks run in parallel, each in a separate sandbox. The framework supports pluggable providers, including Modal, Daytona, Runloop and LangSmith.<\/span><\/p>\n

The toolset is deliberately small \u2014 roughly 15 curated tools covering shell execution, web fetching, API calls, Git operations, and integrations with Linear and Slack, plus Deep Agents\u2019 built-in tools for file operations, search, planning, and subagent spawning. Stripe\u2019s insight was that tool curation matters more than quantity. Open SWE follows that principle and allows organizations to add tools for internal APIs or testing frameworks as needed.<\/span><\/p>\n

Context engineering happens through two channels. An AGENTS.md file at the repository root gets injected into the system prompt \u2014 the equivalent of Stripe\u2019s rule files, encoding conventions, testing requirements, and architectural decisions. The full Linear issue or Slack thread history is assembled and passed to the agent, so it starts with rich context rather than having to discover everything through tool calls.<\/span><\/p>\n

Workflow Integration<\/b><\/h3>\n

Open SWE integrates through the surfaces where developers already work. Mention the bot in a Slack thread, and it replies with status updates and PR links. Comment @openswe on a Linear issue, and the agent reads the full context, acknowledges with a reaction, and posts results as comments. Tag @openswe in PR comments on agent-created pull requests, and it addresses review feedback and pushes fixes to the same branch.<\/span><\/p>\n

Each invocation creates a deterministic thread ID, so follow-up messages route to the same running agent. You can send messages mid-task \u2014 change the spec, add a requirement, redirect the approach \u2014 and the agent integrates your feedback into its active session without restarting.<\/span><\/p>\n

When the work completes, the middleware automatically opens a PR if the agent hasn\u2019t already. The agent runs linters, formatters, and tests before committing. Organizations can extend this with deterministic CI checks, visual verification, or review gates as additional middleware.<\/span><\/p>\n

\u201cWhen Stripe, Ramp, and Coinbase independently arrive at the same architecture for internal AI coding agents, you take notice. Open SWE brings together the important elements of isolated sandboxes, curated toolsets, subagent orchestration, and workflow integration through surfaces developers already use,\u201d according to Mitch Ashley, <\/span>VP and practice lead for software lifecycle engineering at<\/span> The Futurum Group<\/span><\/a>.<\/span><\/p>\n

\u201cFor teams evaluating whether to build their own, the rebuild calculus has changed. These are production-validated patterns, and organizations that reconstruct them from scratch are absorbing delay and risk the market has already worked through.<\/span><\/p>\n

<\/b>Why This Matters for DevOps<\/b><\/h3>\n

Open SWE represents a different trajectory than the AI coding tools we\u2019ve covered in this series. Claude Code, GitHub Copilot, Gemini Code Assist and Cursor are products you adopt. Open SWE is a customizable framework. The distinction matters for organizations that need control over how agents interact with their codebase, internal tools, and workflows.<\/span><\/p>\n

The sandbox isolation model addresses security concerns we\u2019ve seen across the agent landscape\u2014full permissions inside strict boundaries. Blast radius contained. The same pattern is formalized in IronCurtain\u2019s architecture and VS Code\u2019s agent hooks, now implemented as a deployment model.<\/span><\/p>\n

The middleware architecture is the extensibility story. Deterministic hooks for CI gates, security scans, approval workflows \u2014 without modifying core agent logic. Configuration-level customization rather than fork-level modification.<\/span><\/p>\n

The framework defaults to Claude Opus 4.6 but supports any LLM provider. Organizations can configure different models for different subtasks \u2014 the same strategy-vs-tactics separation we\u2019ve seen in Gemini CLI\u2019s model routing and Random Labs\u2019 Slate architecture.<\/span><\/p>\n

Open SWE is available now at github.com\/langchain-ai\/open-swe<\/a>. A hosted version is available at swe.langchain.com<\/a> and requires an Anthropic API key.<\/span><\/p>\n

Read More<\/a><\/p>\n

\u200b<\/p>","protected":false},"excerpt":{"rendered":"

Stripe built Minions. Ramp built Inspect. Coinbase built Cloudbot. Three engineering organizations, working independently, arrived at similar architectural decisions for […]<\/p>\n","protected":false},"author":1,"featured_media":3665,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[5],"tags":[],"class_list":["post-3664","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-devops"],"_links":{"self":[{"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/posts\/3664","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/comments?post=3664"}],"version-history":[{"count":0,"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/posts\/3664\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/media\/3665"}],"wp:attachment":[{"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/media?parent=3664"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/categories?post=3664"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/tags?post=3664"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}