AI coding tools directory
Every coding tool we cover, grouped by what it does, with a link to our most useful page on it. Deciding between two tools? See Compare. Planning a whole setup? Start with how to pick an AI coding agent stack, and check what these tools really cost.
Editors / IDEs
The tools developers live in all day. The right question is not just model quality; it is how well the editor handles context, diff review, navigation, and team rollout.
Cursor
The AI-native editor most teams compare first when they want agent behavior directly inside a VS Code-style workflow.
Start with the three-way editor comparison →GitHub Copilot
Still the GitHub-native default, but agent mode, MCP support, and usage-based billing changed the evaluation criteria in June 2026.
See the billing breakdown →Windsurf
The flat-rate editor challenger that deserves a real pilot when teams want capable AI help without GitHub-style usage sensitivity.
Read the Windsurf breakdown →JetBrains AI
Best evaluated through the lens of existing JetBrains-heavy teams, code intelligence depth, and enterprise workflow fit.
Compare JetBrains AI with Amazon Q →Zed AI
A fast open-source editor with a different bet: speed, collaboration, and AI without inheriting the full VS Code extension universe.
Read the Zed AI breakdown →Replit Agent
Useful when the browser-native dev environment is part of the value, especially for prototypes and hosted app iteration.
Read the Replit analysis →CLI agents
Terminal agents are where prompt quality meets hard reality: tests, git state, shell commands, and bounded execution on real repositories.
Claude Code
Strong at developer-supervised terminal work, especially when you care about controlled edits, reviewable diffs, and explicit task scoping.
Compare Claude Code with Codex CLI →OpenAI Codex
Best understood as an execution agent, not just a model name, with growing relevance in headless and CI-style workflows.
Read the Codex CI guide →Opencode
The open-source Go-based terminal coding agent from the SST team. Model-agnostic, audit-friendly, and built for teams that want BYOK control over the full agent loop.
Read the opencode breakdown →Aider
The safest terminal-first open-source default for many teams because its workflow stays close to git and code review discipline.
See the open-source agent guide →Continue.dev
Open-source editor extension for teams that want to choose their own models, now under Cursor ownership.
Read the Continue.dev analysis →Cline
One of the clearest open-source choices for teams that want visible approvals, model portability, and a repo-shapable workflow.
Read the Cline breakdown →Autonomous agents
Tools that take a ticket and come back with a pull request. Judge them by how many of their changes survive review.
Devin
The best-known managed “AI software engineer” product, now easier to trial and more interesting when its testing evidence actually reduces review burden.
Compare Devin with OpenHands →OpenHands
The open-source alternative for teams that want autonomy experiments without locking into a single managed runtime.
Read the OpenHands comparison →SWE-agent
Best treated as benchmark-informed patterns for issue-oriented task execution, not proof that unsupervised software delivery is solved.
Read the SWE-agent guide →Open-source coding models
Model choice matters most when you control the stack. Once you do, context plumbing, eval discipline, and tool integration usually matter more than leaderboard screenshots.
Hermes (Nous Research)
The open-weight instruction-tuned model family that has become the standard BYOK recommendation for serious coding stacks. Strong instruction following, fine-tuning-friendly, and available at multiple sizes.
Read the Hermes coding breakdown →Ornith
Coding-focused fine-tuned models optimized for agentic context maintenance and tool-use consistency. The family to watch for teams that evaluate open-weight coding models seriously.
Read the Ornith model breakdown →Pi / Inflection AI
Less central to day-to-day coding stacks today, but still useful to track when teams benchmark assistant behavior against broader conversational styles.
Read the Pi analysis →Mistral
Worth watching for sparse MoE efficiency and long-context pressure on proprietary tools, especially in cost-sensitive deployments.
See the open-model comparison →Llama
Still a serious option when ecosystem support and self-hosting flexibility matter more than having the newest closed model features first.
Read the open-source guide →Protocols
The standards that decide how agents reach tools and context, and how they hand work to each other.
MCP
The model-to-tool context layer developers now need to understand if they want safe, inspectable access to files, APIs, and local systems.
Read the MCP + A2A guide →A2A
More useful when you need explicit delegation between agents than when you just need one agent to use a few tools well.
Read the MCP + A2A stack guide →Frameworks
Frameworks decide how much orchestration complexity you own. They are worth using only when traceability, state, and retries justify the extra moving parts.
LangGraph
The best fit when you need explicit stateful flows, durable execution, and enough control to debug the unhappy path.
Compare LangGraph with CrewAI and AutoGen →CrewAI
Still attractive for role-based orchestration, but should be judged on observability and failure recovery rather than demo readability.
Read the framework comparison →AutoGen / AutoGPT lineage
Important mostly as a reminder that agent conversations alone are not architecture; execution control and debuggability still decide production value.
Read the framework comparison →