Directory + guides · No pay-to-play
The factory, without the sales deck.
An agent-native software factory takes a signal in one end — an issue, a spec, a customer report — and produces deployed, verified software out the other. Agents execute. Engineers define intent and review results.
108 hand-curated listings, 8 guides, and 4 sourced case studies. Assessments are editorial. Placement is not for sale. The publisher founded Trevize; that conflict is on methodology.
The stack
All listingsWhat should be built
Intent
How work enters the factory: issues, specs, customer reports, and the human decisions that define done.
How work is routed
Orchestration
The control plane that sequences agents, assigns models, retries failed steps, and keeps a run moving without a human babysitting every tool call.
Who writes the code
Execution
The coding agents and harnesses that read a repository, change files, run commands, and produce a reviewable diff.
Where agents run
Sandboxing
Isolated computers, preview environments, and network policy so an agent can act without inheriting a developer laptop.
What must pass
Verification
Tests, reviews, hooks, and merge gates that decide whether output is software or just a plausible patch.
What the factory did
Observability
Run history, cost, cycle time, and audit trails so a team can manage a portfolio of agent work instead of a pile of chat logs.
Directory
Browse and filter 108 listings, one YAML file each. Every page carries a
verifiedAt date and an editorial take. Start with a
roundup or a comparison if
you already know the job.
- Agent Development Kit
Google’s open kit for building Gemini-centered agents.
- AgentOps
Observability aimed at agent runs, not only chat completions.
- Agno
Python multi-agent framework (formerly Phidata).
- Aider
Git-native CLI agent that commits as it works.
- Amazon Q Developer
AWS’s coding assistant and agent, billed into the AWS account.
- Amp
Sourcegraph’s agent-native coding product, successor energy to Cody.
- Augment
Enterprise coding agent that sells context over huge codebases.
- Blaxel
Agent sandbox/compute with a hibernation-shaped billing story.
- Bolt
StackBlitz’s in-browser agent that builds apps in WebContainers.
- Braintrust
Eval-first platform: datasets, experiments, and logging.
- Browserbase
Headless browsers as a service for agents that must click.
- Claude Code
Anthropic's coding agent across terminal, IDE, desktop, and browser.
- Cline
Open VS Code agent that drives the editor like a human.
- Cloudflare Agents
Agents SDK on Durable Objects — this site’s own job runtime.
- Cloudflare Sandbox
Isolated code execution on Cloudflare, next to Workers.
- CodeQL
GitHub’s semantic code analysis, query-based and CI-native.
- Coder
Self-hosted workspaces on your infrastructure, Terraform-defined.
- CodeRabbit
High-volume AI pull-request review as a product.
- Context7
Up-to-date library docs injected into coding agents.
- Continue
Open IDE agents, acquired by Cursor, source still public.
- CrewAI
Role-playing multi-agent framework with a commercial control plane.
- Crush
Charm’s glamorous terminal coding agent.
- Cubic
AI code review product aimed at merging faster with fewer comments.
- Cursor
An agent-native IDE with cloud agents that run in isolated VMs.
- Dagger
Programmable CI/CD engine agents can call as a tool.
- Daytona
Workspaces for agents, with a more persistent posture than E2B.
- DeepEval
Pytest-flavored evals for LLM systems.
- DeepSource
Static analysis SaaS with Autofix, now talking to agents.
- Devbox
Nix-powered reproducible shells for humans and agents.
- Devin
Cognition’s cloud software engineer; Windsurf is now Devin Desktop.
- DSPy
Stanford’s framework for programming — not prompting — LMs.
- E2B
Firecracker microVMs as a sandbox API for AI agents.
- Ellipsis
AI reviewer and fixer that can open follow-up PRs.
- Factory
Droids across the SDLC, sold as a software factory.
- Fern
OpenAPI-to-SDKs and docs, an alternative to Stainless.
- Firecracker
AWS’s microVM monitor that sandbox vendors wrap.
- Fly Machines
Fast microVMs you API-control, used as agent computers.
- Galileo
Evaluation and observability for production AI systems.
- Gemini CLI
Google’s open-source terminal agent for Gemini models.
- Gemini Code Assist
Google’s IDE coding assistant for Gemini, distinct from Jules.
- Giskard
Open testing and red-teaming for AI applications.
- GitHub Codespaces
Microsoft-hosted dev VMs that agents and humans already share.
- GitHub Copilot
Incumbent coding assistant with a cloud coding agent on GitHub.
- Gitpod
Open-source platform for automated, ready-to-code environments.
- Goose
Block’s local-first, open-source agent with recipes.
- Graphite
Stacked PRs plus AI review, aimed at a faster merge path.
- Greptile
AI review that claims to read the whole repo, not the diff alone.
- Hatchet
Open task queue aimed at durable, observable background jobs.
- Haystack
deepset’s open framework for pipelines, RAG, and agents.
- Helicone
Open LLM gateway with logging, caching, and spend controls.
- Inngest
Event-driven durable functions for product and agent jobs.
- Inspect
UK AISI’s open framework for evaluating language model agents.
- Jules
Google’s asynchronous coding agent that opens pull requests.
- Junie
JetBrains’ coding agent inside IntelliJ-family IDEs.
- Kilo
Open agent across IDE and CLI, plus a zero-markup model gateway.
- Langfuse
Open-source LLM observability you can self-host.
- LangGraph
Graph runtime for agents, from the LangChain project.
- LangSmith
LangChain’s commercial tracing, eval, and prompt hub.
- LangWatch
Observability and evaluation platform for LLM products.
- Letta
Stateful agents with memory, from the MemGPT lineage.
- Linear
Issue tracker that became an intent API for coding agents.
- LlamaIndex
Data framework for LLM apps that grew workflows and agents.
- Logfire
Pydantic’s observability product, including LLM/agent traces.
- Lovable
Prompt-to-full-stack app builder with a hosted computer.
- Mastra
TypeScript agent framework with workflows and evals attached.
- Mergify
Merge queue and automation rules for GitHub.
- Microsoft Agent Framework
Microsoft’s successor surface for AutoGen-style multi-agent apps.
- Mintlify
Docs platform that treats agent-readable documentation as a product.
- Modal
Serverless containers and GPUs that agents can treat as a computer.
- n8n
Open workflow automation that now grows agent nodes.
- Northflank
Workloads and sandboxes on your cloud, with an agent-aware pitch.
- OpenAI Agents SDK
OpenAI’s official Python/TS SDK for agent loops.
- OpenAI Codex
OpenAI’s coding agent across CLI, IDE, and cloud tasks.
- OpenCode
Open terminal coding agent; Kilo CLI is a downstream fork.
- OpenHands
Open software-agent runtime with a computer you can watch.
- OpenRouter
Multi-provider inference router used by many coding agents.
- Phoenix
Arize’s open observability and eval notebook for LLM apps.
- Plandex
Open terminal agent that plans large diffs in a sandbox.
- Portkey
AI gateway: routing, guardrails, and observability in one proxy.
- Prefect
Python workflow orchestrator used to schedule agent jobs.
- Promptfoo
Open evals and red-team for LLM apps and agents.
- Pydantic AI
Typed Python agents from the Pydantic team.
- Qodo
Generation and review agents aimed at test-aware pull requests.
- Ragas
Open metrics for RAG and LLM application quality.
- Replit Agent
Hosted computer plus agent that can build and deploy an app.
- Roo Code
Cline fork aimed at modes, automation, and teams.
- Semgrep
Fast static analysis you can teach, including AI-assisted rules.
- smolagents
Hugging Face’s small, first-class code-agent library.
- Sourcery
Refactoring and review assistant with a long IDE history.
- Spec Kit
GitHub’s toolkit for spec-driven development with coding agents.
- SpecStory
Capture AI IDE sessions so intent does not die in the chat.
- Stainless
SDK generation from an OpenAPI spec — a machine-readable intent.
- Steel
Open-source browser sandbox API for AI agents.
- SWE-agent
Princeton’s agent that made SWE-bench a product category.
- Tabby
Self-hosted coding assistant you run like an internal service.
- Tabnine
Long-running completion vendor with enterprise model hosting options.
- Temporal
Durable workflow engine that can sequence factory steps.
- Tessl
Spec-first platform for building software with agents.
- Trae
ByteDance’s agentic IDE, free-tier aggressive, closed runtime.
- Trevize
Shared cloud workspaces where teams and agents close the loop.
- Trigger.dev
Background jobs for TypeScript apps, including long agent runs.
- Trunk
Unified linters, formatters, and CI flakiness as a merge gate.
- v0
Vercel’s UI generator that ships into a Next.js repo.
- Vercel AI SDK
TypeScript SDK for streaming model UIs and tool-calling apps.
- Vercel Sandbox
Vercel’s documented sandbox primitive for running untrusted code.
- Warp
Open terminal, multi-model agents, and factories in early access.
- WebContainers
StackBlitz’s in-browser Node runtime used as an agent sandbox.
- Zed
Open Rust editor with first-class agent panels.
Guides
All guides- Assembling a self-hosted stack
How to pick an open execution harness, a sandbox you can inspect, and gates you own — without pretending one vendor is the factory.
- Choosing an execution harness
How to pick the agent that writes the diff — IDE, CLI, or cloud worker — without a bake-off that measures the wrong thing.
- Designing the intent layer
How work enters a factory — specs, issues, and pass conditions — so agents have something to close against.
- Verification and quality gates
How to decide a diff is software — tests, review, evals, and merge policy — without trusting a green check the agent painted.
- What a software factory is
The agent-native definition, the older DevSecOps meaning, and the loop that actually has to close.
- Sandboxing and isolation
Why the agent needs its own computer, what isolation actually means, and which shortcuts turn a factory into a laptop with extra steps.
- Measuring factory output
What to count when agents write the diffs — and which dashboards are just token spend with a nicer chart.
- How to read a listing
What the fields mean, why verifiedAt is load-bearing, and how to treat an editorial take.
Case studies
All studiesProduction deployments with every scale claim tied to a primary source. Talk-circuit numbers we could not open on the organization’s own site are omitted.
- Shopify: River, a Slack-native agent on the Aquifer substrate
Shopify built River, an AI agent that only operates in public Slack channels. It reads code, runs tests, opens pull requests, queries the warehouse, and inspects production traces. Underneath, Aquifer is the internal platform (durable session, harness, sandbox, gateway, credentials proxy, observability). The May 2026 engineering post is explicit that the 2024 monorepo-and-Nix work was done so agents would have a substrate.
- Stripe: Internal minions and AI-assisted pull requests
Stripe runs internal coding agents it calls minions. In the Sessions 2026 developer keynote, API design lead Michelle Bu described minions that open production pull requests for human review, alongside a broader rise in AI-assisted PR creation and agent traffic to Stripe’s own documentation.
- StrongDM: A three-person AI team and non-interactive development
StrongDM’s AI team published a first-party account of a “Software Factory” in which humans write specifications and scenarios, and agents write and iterate on code without human-authored patches or human code review. The validation story is scenario satisfaction against a Digital Twin Universe (behavioral clones of Okta, Jira, Slack, and Google Workspace APIs), not a conventional green test suite.
- Uber: Minion, uReview, and the internal agent stack
Uber engineers have described Minion as an internal background-agent platform that takes toil (validation-heavy fixes, cleanup, similar interruptions) and returns a pull request for a human. That description comes from talks by Uber staff, not from an uber.com post named Minion. What Uber has published under its own domain is uReview, a GenAI review system that comments on diffs, plus adjacent tools such as FixrLeak. This page keeps those sources separate.
Compare
Meaningful pairs only. Not a generated cross-product.
Roundups
Inclusion is a filter on structured fields, not a rank.
Build log
This site was assembled by the same kind of factory it describes. The log records the pipeline, what broke, and what agents got wrong.
Finishing the brief — pages, sources, and a hundred listings
How this is run
Listings live as YAML in Git. Assessments are opinionated and sourced. We do not sell placement. The publisher founded Trevize; that conflict is on the methodology page, not buried in a footer.