# Ca · Coding Agent — element 5 of 58

> Your first ten engineers. Turns intent into shipped software.

- **Group:** 2 · Build
- **Necessity:** Essential
- **Price band:** $$ · $30–150/mo
- **Maturity:** Stable
- **Edition:** v2026.Q3 · verified 2026-08-06

## Leading tools (v2026.Q3)

- **Claude Code** — agentic depth, most-loved
- **Cursor** — the AI-native IDE
- **OpenAI Codex** — parallel agents, benchmark top
- **GitHub Copilot** — the enterprise default
- **OpenCode** — open-source, any model

## Our take

The single highest-leverage element on the table. If you hold only three elements, this is one of them.

## Combines with

Fm, Dp, Cq
- Guide: [Best coding agent for solo founders (2026)](https://elems.ai/best/best-coding-agent-for-solo-founders.html)

## The top 5 — deep dossier (verified 2026-08-06)

Claude Code, if you can only hold one — it leads every satisfaction survey that exists (46% "most loved", 91% CSAT), tops the Terminal-Bench 2.1 harness leaderboard, and reached a $2.5B run-rate faster than any dev tool in history. Cursor is the pick when IDE speed matters more than agentic depth; Codex when you want maximum autonomous throughput on a $20 plan; Copilot when enterprise procurement decides; OpenCode when you refuse lock-in.

1. **Claude Code** (Anthropic) — Pro $17–20/mo · Max $100–200/mo · Team from $20/seat · Enterprise seat + API usage. Best for: Deep, multi-file agentic work — architecture, refactors, ship-a-feature-from-a-sentence — from terminal, IDE, CI, or phone. Why: Leads every independent satisfaction metric: 46% "most loved" vs Cursor's 19% and Copilot's 9%, 91% CSAT, NPS 54 (JetBrains AI Pulse, Jan 2026); 44% of developers choose it specifically for complex tasks; #1 on the Terminal-Bench 2.1 harness leaderboard (83.8%, Jun 2026). $2.5B annualized run-rate by Feb 2026 — the fastest-scaling dev tool ever. Watch: Cost at heavy use — power users outgrow Pro fast and land on Max tiers. Copilot still out-adopts it inside 10,000+-employee orgs. [https://claude.com/product/claude-code](https://claude.com/product/claude-code)
2. **Cursor** (Anysphere → SpaceX (closing Q3 2026)) — Hobby free · Pro $20/mo · Pro+ $60 · Ultra $200 · Teams $40/user. Best for: Founders who live in an IDE and want the fastest inner loop — tab completion, inline agent edits, and background cloud agents in one polished editor. Why: The fastest-growing software company on record: $1B ARR (Nov 2025) → $3B (May 2026), 1M+ paying users, 50,000 enterprise customers — and the subject of the largest startup acquisition ever, SpaceX's $60B all-stock purchase agreed June 16, 2026. Watch: Post-acquisition model-routing risk: Cursor depends on Anthropic/Google models and the new owner has its own (Grok). Pricing and limits have changed repeatedly; large complex changes still favor Claude Code/Codex. [https://cursor.com](https://cursor.com)
3. **OpenAI Codex** (OpenAI) — Bundled with ChatGPT: Plus $20/mo · Pro $100–200/mo · Business $25/user; token credits since Apr 2026. Best for: Maximum autonomous task volume per dollar — parallel cloud agents, background automation, and near-benchmark-top models on a $20 entry. Why: The volume play: user base exploded from under 1M weekly (Feb 2026) to ~9M (Jul 2026) after the GPT-5.6 launch; business usage grew 6x Jan–Apr 2026; sits within a point of the harness-leaderboard top (Codex + GPT-5.5, 83.1% Terminal-Bench 2.1). Watch: Capacity strain moves limits under your feet — context was cut from 372k to 272k tokens during the July surge. OpenAI models only; the IDE-native experience trails Cursor. [https://openai.com/codex/](https://openai.com/codex/)
4. **GitHub Copilot** (Microsoft / GitHub) — Free tier · Pro $10/mo · Pro+ $39 · Business $19/user · Enterprise $39/user; usage-based AI credits since Jun 2026. Best for: Teams standardized on GitHub — cheapest paid entry, deepest PR/CI integration, easiest org-wide rollout, 90% of the Fortune 100. Why: The installed-base king: 4.7M paid subscribers (+75% YoY, Microsoft FY26 Q2), ~42% paid share, 56% adoption in 10,000+-employee orgs. Multi-model (OpenAI, Anthropic, Google) with enterprise policy controls nobody else matches. Watch: Lowest satisfaction of the leaders — 9% "most loved", and 29% workplace usage against 42% paid share implies shelfware seats. The June 2026 usage-based billing shift produced 10–50x bill spikes for agentic power users. [https://github.com/features/copilot](https://github.com/features/copilot)
5. **OpenCode** (SST (open source, MIT)) — Free, MIT-licensed · bring any API key or existing Claude/ChatGPT/Copilot subscription · optional pay-per-use Zen gateway. Best for: Lock-in-averse and cost-sensitive builders who want one agent harness over 75+ model providers — including open-weight and local models. Why: The definitive open-source agent: ~190k GitHub stars, ~900 contributors, top write-in of the State of AI 2026 survey (389 mentions). As near-frontier open models close the benchmark gap at a sixth of the cost, a provider-agnostic harness captures that arbitrage in a way no closed subscription can. Watch: No vendor support — community only, and quality tracks whatever model you plug in. Unmanaged token spend on premium models can exceed a flat $20 plan. [https://opencode.ai](https://opencode.ai)

### How to choose
- If Your bottleneck is hard engineering — architecture, gnarly refactors, multi-repo work → Claude Code. 44% of developers pick it specifically for complex tasks; pair heavy use with the Max tier rather than fighting Pro limits.
- If You live in an editor and the inner loop is everything → Cursor — accepting the post-SpaceX model-routing uncertainty as the price of the polish.
- If You want maximum autonomous throughput per dollar, or background agents running in parallel → Codex on ChatGPT Plus at $20 — the cheapest seat at the benchmark top table.
- If Procurement, compliance, or a GitHub-standard org decides for you → Copilot Business — the 90%-of-Fortune-100 path, with the honest caveat that its own users love it least.
- If You refuse lock-in, run high token volume, or want open-weight models → OpenCode — free harness, any provider, and the open-model price-performance keeps improving quarterly.

### The field (30 more)

Devin / Devin Desktop, Amp, Aider, Cline, Kilo Code, Zed, JetBrains AI / Junie, Antigravity, Jules, Kiro, Factory (Droid), Warp, Trae, Qodo, Tabnine, Continue, Goose, OpenHands, Crush, Mistral Vibe, Kimi Code CLI, Grok Build, Qwen Code / Qoder, CodeBuddy, Pi / Omp, Open Interpreter (fading), Gemini CLI (sunsetting), Roo Code (dead), Amazon Q Developer (sunsetting), Windsurf (renamed)

Full dossier data: https://elems.ai/e/ca.json

---
Source: [elems.ai](https://elems.ai/e/ca.html) — the periodic table of the AI-led startup. Data: https://elems.ai/elements.json (CC BY 4.0, cite elems.ai).