AI Coding Assistants

Executive Summary

AI coding assistants have matured into a large, fast-churning ecosystem of autonomous agents that go far beyond autocomplete. Unlike GitHub Copilot's inline suggestions, these tools edit files, run terminal commands, browse the web, and execute complex multi-step tasks with human oversight.

This report analyzes ten active AI coding assistants across terminal, IDE, and cloud deployment models — and records the category's first consolidation casualties.

Key Findings:

• OpenCode leads by an order of magnitude — 172K+ GitHub stars and a claimed 7.5M monthly developers under new parent company Anomaly^[1]
• Heavyweights joined — OpenHands (76K+ stars, $18.8M Series A), Goose (48K+, donated by Block to the Linux Foundation's Agentic AI Foundation), Crush (25K+, Charm), and Kilo Code ($8M seed, GitLab right-of-first-refusal)^[2][3][4][5]
• The consolidation wave is real — Roo Code shut down in May 2026; Aider has shipped no feature release since August 2025; Sweep's last release was February 2026^[6][7][8]
• Licensing diversified — the "all Apache 2.0" era is over: OpenCode and Cline are MIT, Goose is Apache, Crush is FSL (source-available), and Claude Code is proprietary with a public repo^[9]
• Foundation governance arrived — the Agentic AI Foundation (Anthropic, Block, OpenAI) now anchors MCP, Goose, and AGENTS.md^[10]

Strategic Planning Assumptions:

• By 2027, automated PR review agents will be standard in 40% of enterprise CI/CD pipelines
• By 2027, at least two more of today's ten active tools will shut down or be acquired (Roo Code was the first; GitLab holds a Kilo ROFR through August 2026)
• The February 2026 prediction that combined category stars would exceed 500K by year-end has already been passed — the ten active tools sum to ~640K as of June 2026

---

Market Definition

AI coding assistants are autonomous agents that help developers write, edit, and manage code through natural language interaction. They differ from autocomplete tools (like Copilot) by supporting multi-step tasks, file system operations, and terminal execution.

Inclusion Criteria:

• Operates as an autonomous agent (not just autocomplete)
• Supports multiple LLM providers OR is the official tool from a model provider
• Active development (commits within last 30 days)
• A traction signal (100+ GitHub stars, notable backing, or significant community attention) — promising new projects are deliberately included early

Exclusion Criteria:

• Full AI-native IDEs (Cursor, Windsurf) — these are a separate category
• Pure autocomplete tools (GitHub Copilot standalone)
• Closed-source enterprise agents (Devin, Factory Droid, Auggie) — open-source self-hosted alternatives are covered in background agents. Amp, formerly excluded on these grounds, joins this cycle: its free tier and terminal-first developer focus make it a direct peer
• Foundation-lab CLIs other than Claude Code (Codex, Gemini CLI, Grok Build) — covered in Foundation Lab Coding Agents

Membership notes (June 2026): Aider and Sweep technically strain the active-development criterion — Aider ships only maintenance patches and Sweep has not released since February 5, 2026. Both are retained this cycle with honest maturity labels and flagged for review next refresh.

---

Comparison Matrix

Departed

• Roo Code — shut down May 2026 (~3M installs at shutdown); the team pivoted to Roomote, a Slack-based operational agent. Official migration paths: Cline (its upstream fork origin) or Kilo Code (preserves .roomodes configs).^[6]

---

Product Profiles

Aider

The terminal pioneer, now in maintenance mode^[7]

• 46K stars; model-agnostic across 75+ providers with git-native workflows
• No feature release since v0.86.0 (August 2025); leaderboards last updated November 2025
• Still excellent for what it does — but the innovation has moved elsewhere

---

Amp

The opinionated multi-model frontier agent^[11]

• Spun out of Sourcegraph as Amp, Inc. in December 2025 — Quinn Slack and Beyang Liu run it as a frontier agent lab
• Multi-model by design: Amp picks the model per task (Opus for smart mode, GPT-5.5 for deep/rush, Gemini for review) — you don't
• Free tier with $10/day allowance (ads dropped March 2026); pay-as-you-go with zero markup on provider pricing
• Closed source; pass-through pricing on frontier models adds up — cost is the most common community complaint

---

Claude Code

Anthropic's flagship — the revenue and capability benchmark^[9]

• 131K+ stars; terminal, IDE, web, mobile, and Slack surfaces; $2.5B+ run-rate revenue
• Background sessions, scheduled Routines, subagent orchestration, computer use
• Claude-only by design; $20–125/seat/mo across tiers; proprietary license (public repo)

---

Cline

VS Code's approval-first agent — now a funded company^[12]

• 63K stars, ~4.3M VS Code installs; $32M seed + Series A (Emergence, Pace)
• Plan/Act modes, CLI 2.0 for headless CI/CD, JetBrains early access, enterprise tier
• The official migration home for former Roo Code users

---

Continue

Repositioned around source-controlled AI checks^[13]

• 33K stars; the pivot: markdown-defined AI checks that run on every PR as GitHub status checks
• IDE extensions and cn CLI agent remain, but PR enforcement is the new center of gravity
• ~$5.1M raised; no Series A as of June 2026

---

Crush

Charm's glamourous terminal agent^[4]

• 25K stars in under a year; LSP-aware context, MCP support, mid-session model switching
• Continues the original OpenCode Go codebase after the contested 2025 split with the SST side
• FSL-1.1-MIT license (source-available, not OSI); token-hungry per community reports

---

Goose

Local-first agents under foundation governance^[3][10]

• 48K+ stars; CLI + desktop app; any LLM including local models; MCP-native extensions
• Donated by Block to the Linux Foundation's Agentic AI Foundation (repo moved April 2026)
• Vendor-neutral by construction — the safest long-term governance story in the category

---

Kilo Code

The multi-surface gateway play^[5]

• ~20K stars, 3M+ users; VS Code, JetBrains, CLI, Slack, and cloud agents
• 500+ models with zero-markup pass-through pricing; Teams at $15/user/mo
• $8M seed led by GitLab co-founder Sid Sijbrandij; GitLab holds an acquisition ROFR through August 2026

---

OpenCode

The adoption runaway^[1]

• 172K+ stars — more than the next two tools combined; claimed 7.5M monthly developers
• Terminal, desktop, and IDE surfaces; 75+ providers; Zen (per-token) and Go ($10/mo) paid tiers
• Now built by Anomaly; MIT-licensed; listed on OpenAI's "Codex for Open Source" program

---

OpenHands

Delegate the whole ticket^[2]

• 76K+ stars — the most-starred open agent platform; CLI, GUI, cloud, and SDK
• Issue-to-PR autonomy: writes code, runs commands, browses, opens PRs end-to-end
• $18.8M Series A (Madrona, November 2025); MIT core with source-available enterprise directory

---

Refact.ai

Self-hosted agents for teams that can't use the cloud

• 3.5K+ stars (BSD-3-Clause); VS Code, JetBrains, and more, backed by a self-hostable server
• On-prem deployment and fine-tuning on your own codebase — the compliance play
• Held the #1 open-source spot on SWE-bench Verified at its June 2025 submission; small team, funding undisclosed

---

Sweep

JetBrains-native, but stalled^[8]

• 67K JetBrains installs, 4.7-star rating; YC-backed ($500K, 3-person team)
• No plugin release since February 5, 2026 — flagged for membership review next cycle
• If you need JetBrains-native today, also weigh Cline's JetBrains early access

---

Gap Analysis

Key gaps across the market:

• Terminal-only tools lack team features — Aider, Crush, and Goose have no shared/team layer; Kilo and OpenCode are filling it
• Closed-source pressure is rising — Factory Droid tops Terminal-Bench, Amp ships an ad-supported free tier; the open category competes on trust and BYOK economics
• Sustainability is the new differentiator — funding (Cline, OpenHands, Kilo), foundation governance (Goose), or platform scale (OpenCode, Claude Code) now separate survivors from the stalled

---

Strategic Recommendations

By Use Case

By Buyer Profile

Individual developers: → OpenCode or Crush in the terminal; Cline in VS Code. All free with BYOK.

Startups (< 20 engineers): → Claude Code if standardized on Claude; OpenCode for flexibility; OpenHands when you want agents working tickets in the background.

Scale-ups and enterprises: → Claude Code (enterprise tier, Bedrock/Vertex deployment), Cline Enterprise (SSO, RBAC), or OpenHands (self-hosted + cloud). Weigh vendor sustainability — this category just had its first shutdown.

Open-source-first teams: → Goose (foundation-governed) is the strongest neutrality story; OpenCode and OpenHands for scale; note Crush's FSL license is source-available, not OSI.

---

Market Outlook

Near-Term (2026)

• Consolidation continues: watch the GitLab/Kilo ROFR (expires August 2026), Sweep's silence, and Aider's maintenance mode
• Closed-source CLIs (Factory Droid, Amp, Auggie) force open tools to compete on benchmarks, not just openness
• The Agentic AI Foundation becomes the neutral home for category infrastructure (MCP, AGENTS.md, Goose)

Medium-Term (2027-2028)

• Automated PR review agents standard in 40% of enterprise CI/CD pipelines
• Surviving tools converge on multi-surface (terminal + IDE + cloud + chat) with hosted gateways
• Expect 2-3 acquisitions as platform vendors buy distribution

Long-Term (2028+)

• The "coding assistant" framing dissolves into general agent harnesses with coding as one skill
• Foundation-governed, model-neutral tooling becomes the enterprise default for trust reasons

---

Bottom Line

The category grew up fast and is now consolidating. The market splits four ways:

Adoption leaders:

• OpenCode — the open-source standard by raw numbers
• Claude Code — the capability and revenue benchmark, at the cost of model lock-in

Funded challengers:

• OpenHands for ticket-level autonomy, Cline for the IDE approval workflow, Kilo Code for multi-surface gateway economics

Neutral infrastructure:

• Goose under Linux Foundation governance — the trust play

Fading first wave:

• Aider (maintenance), Sweep (stalled), Roo Code (gone)

The strategic question has shifted from "which tool is best?" to "which tool will still be shipping in two years?" Favor tools with funding, foundation backing, or platform-scale adoption — and keep your workflows portable (BYOK, MCP, AGENTS.md) so switching stays cheap.

---

Research by Ry Walker Research • methodology

Disclosure: Author is CEO of Tembo, which builds coding-agent orchestration in an adjacent category.