Claude Code · Devin · Sourcegraph Amp · Cline · OpenHands · Roo Code · Replit Agent · Bolt.new · Lovable · v0 by Vercel.
One question: which one is right for your stage?

1. Claude Code Anthropic · official terminal-native autonomous agent · trillion-$ AI lab

Anthropic's official terminal-native autonomous coding agent — the operator's daily driver and the agent SideGuy itself is built with. Lives in your terminal, reads your repo, edits files, runs tests, ships commits. Best-in-class at multi-file refactors and shipping a full feature end-to-end from a one-line spec. AI-baked-in (Claude is the substrate, not a feature) — the gap vs bolted-on AI compounds every Claude release. Two trillion-$ companies (Anthropic + Google substrate) wired together for one operator output.

2. Devin Cognition AI · category-defining autonomous SWE · hosted-agent flagship

The category-defining autonomous SWE — Cognition's hosted agent that takes a Linear/Jira ticket and ships a PR with no human in the middle. Browser-based 'employee' UX with its own VM, browser, terminal, and IDE. Best for teams that want ticket-to-PR full automation as a workflow, not a CLI. Devin pioneered the autonomous-agent category in late 2024 and remains the brand-defensible enterprise pick when leadership wants 'a junior engineer that runs in the cloud' as the org metaphor.

3. Sourcegraph Amp Sourcegraph · enterprise-scale autonomous agent · code-graph-grounded

The enterprise-scale autonomous agent built on Sourcegraph's code intelligence graph — purpose-built for very large codebases (1M+ files) where context retrieval is the bottleneck. Amp pairs autonomous agentic execution with Sourcegraph's symbol graph (call sites, type definitions, cross-repo references). When the agent needs to understand 'how is this function used across 47 services?' it walks the graph instead of guessing from embeddings. The right pick for enterprise eng orgs already running Sourcegraph who want autonomous agents grounded in real code intelligence.

4. Cline Open-source · VS Code agent · self-host friendly · BYOK

The open-source VS Code agent for self-hosted teams that want autonomy without sending code to a vendor. Runs as a VS Code extension, BYOK any model (Anthropic, OpenAI, Bedrock, Azure, local Ollama, vLLM), zero vendor cloud in the data path unless you choose one. The cleanest exit ramp from Devin / Cursor pricing for shops with ops capacity. Active community, MIT-licensed, fork-friendly (Roo Code is its most popular fork).

5. OpenHands Open-source · formerly OpenDevin · research + self-host autonomy

Open-source autonomous coding agent (formerly OpenDevin) — the research-grade self-host answer to Devin. Born as the open-source response to Cognition's Devin, OpenHands now runs as a polished agent platform with browser + terminal + code-edit + planning capabilities. Best for research teams running SWE-Bench experiments, university labs, and engineering orgs that want autonomous agents fully on their own infra with no vendor in the data path.

6. Roo Code Open-source · Cline fork · multi-mode agent personas

The multi-mode fork of Cline that ships specialized agent personas (Architect / Coder / Debugger / Ask). Roo Code (formerly Roo Cline) takes Cline's VS Code agent foundation and adds explicit mode-switching — Architect mode plans high-level designs, Coder mode implements them, Debugger mode triages failures, Ask mode answers questions. The right pick for teams that want autonomy with explicit cognitive-mode separation instead of one monolithic agent prompt.

7. Replit Agent Replit · cloud-native autonomous builder · prototyping leader

The cloud-native autonomous builder for greenfield prototypes and full-stack scaffolding inside Replit's runtime. Replit Agent doesn't just edit code — it provisions the runtime, the database, the deploy target, and ships a working URL from a prompt. Best agent for one-shot full-stack scaffolds, idea validation, and non-developer founders who need a working prototype today. Trade-off: you're locked into Replit's environment.

8. Bolt.new StackBlitz · AI-native web app prototyping · browser runtime

StackBlitz's AI-native web app builder that ships live in the browser via WebContainers. Bolt.new lets you describe a web app and watch the agent code + run + iterate it in a real Node.js runtime running entirely in your browser tab. Zero-install, zero-deploy-config prototyping. Best for AI-native web app prototypes, demo builds, hackathons, and validating UX ideas before investing in a real codebase.

9. Lovable Full-stack web app builder · designer-friendly · built-in deployment

The designer-friendly full-stack web app builder with built-in deployment and Supabase integration. Lovable targets non-developer founders and designers who want to ship a real working full-stack app (frontend + auth + DB + deploy) from natural-language prompts. Tighter design polish than Bolt for production-leaning prototypes, deeper than Replit Agent on the design + UX layer. The right pick when the buyer is the founder/designer, not the engineer.

10. v0 by Vercel Vercel · shadcn/ui + Next.js component generator · ship-to-Vercel native

Vercel's component-generation agent for shadcn/ui + Next.js, optimized for shipping straight to Vercel. v0 generates component-grade React + Tailwind + shadcn/ui code that drops cleanly into Next.js apps and deploys to Vercel in one click. Less of a full-stack agent (Lovable / Bolt win that), more of a component-grade builder for teams already on the Next.js + Vercel + shadcn stack. The right pick when you want polished UI code that fits your existing Next.js codebase, not a separate prototyping environment.

🚀 If you're a Solo founder shipping a SaaS prototype this week

Your problem: You're a solo founder. Idea on Monday, working prototype by Friday. You need an autonomous agent that ships full features end-to-end without you babysitting every line. Cost matters but velocity matters more. Hosted-runtime agents (Replit / Bolt / Lovable / v0) compete with terminal-native agents (Claude Code) on different axes.

1. Claude Code — operator-grade terminal-native agent on Anthropic's frontier substrate — fastest end-to-end shipping if you're terminal-comfortable
2. Replit Agent — prompt-to-deployed-URL in one workflow — fastest if you want hosted-runtime + auth + DB + deploy bundled
3. Lovable — designer-friendly full-stack builds with Supabase + deployment baked in — best for polished UX prototypes
4. Bolt.new — browser-runtime web app prototyping with zero install — fastest demo-grade builds
5. v0 by Vercel — if you're shipping Next.js + Vercel + shadcn — component-grade UI in one click

👨‍💻 If you're a Series A/B startup with a 200K-LOC codebase wanting to accelerate feature delivery

Your problem: You have a real 200K-LOC codebase. Engineers are productive but the backlog grows faster than the team. You need autonomous agents that take a Linear/Jira ticket and ship a PR your team can review — async, parallel, repo-aware. The agent must understand your conventions, not invent new patterns. See the sister AI Coding Tools (IDE assistants) megapage for the live-editing layer that pairs with autonomous agents.

1. Devin — ticket-to-PR full automation with hosted async runs — ship 5-10 PRs in parallel without per-engineer babysitting
2. Claude Code — terminal-native agent that ships full features end-to-end — pair with Devin for async + interactive coverage
3. Cline — self-host + BYOK if you can't send code to vendor cloud — VS Code-native autonomy
4. Roo Code — Architect / Coder / Debugger persona separation if you want explicit cognitive-mode workflows
5. Sourcegraph Amp — code-graph-grounded autonomous agent if your 200K LOC starts hitting context-retrieval limits

🏗 If you're a Mid-market enterprise with a 2M-LOC codebase wanting AI-assisted maintenance + new feature work

Your problem: You're a 50-500 engineer org with a 2M-LOC monorepo across 47 services. Most autonomous agents fail at this scale because embedding-based retrieval gets noisy and the agent hallucinates. You need agents grounded in real code intelligence (symbol graph, call sites, cross-repo refs) plus enterprise deployment options (BYOK, on-prem, audit logs).

1. Sourcegraph Amp — code-graph-grounded autonomous agent purpose-built for monorepo scale — the only honest answer at 2M LOC
2. Devin — hosted async ticket-to-PR works at this scale if Cognition's enterprise tier fits procurement
3. Claude Code — terminal-native agent with MCP tools + custom skills — strong at this scale with explicit context scoping
4. Cline — self-host + BYOK if regulatory mandate blocks public model APIs at this scale
5. OpenHands — fully self-hosted autonomous agent if you want zero vendor cloud in the data path

🏛 If you're a Enterprise CTO/VP Eng standardizing autonomous coding tooling org-wide (security review + procurement)

Your problem: You're standardizing autonomous coding agents across the org. Procurement requires SOC 2 + privacy controls + admin dashboards + license management + a vendor with SLA. Brand defensibility matters at this scale. The category is new — most autonomous agents are still indie-velocity tools, which makes vendor selection a real procurement risk.

1. Devin — Cognition is the brand-defensible category-defining vendor — procurement teams can validate the entity + funding + customer list
2. Sourcegraph Amp — Series D vendor with decade-old enterprise sales motion + on-prem option — procurement already familiar
3. Claude Code — Anthropic-backed (trillion-$ AI lab) — the substrate is the procurement story, MCP-native enterprise integrations
4. Cline — if self-host + BYOK + open-source inspectability is the procurement gate (no vendor lock-in story)
5. OpenHands — open-source autonomous agent if procurement requires fully-self-hosted with no vendor cloud in the data path