Guide · Verified 2026-10-14 · Verified

What Are AI Coding Agents? The 2026 Landscape Guide

What you’ll learn in this article

“Should I use Claude Code or Cursor?” “I keep hearing about Warp and JetBrains Air. What are they?” The AI coding agent market has been reshuffled almost every month since the start of 2026. This guide organizes the current picture on a single page, based on the monthly tracking we have run since May 2026. For what happened in any given month, see the monthly reports.

What are AI coding agents?

An AI coding agent is an AI tool that takes a bundle of development work and carries it through on its own: planning, changing code, and running tests. Traditional autocomplete was built on the premise that “the human writes and the tool helps.” Agents are built on the opposite premise: “the agent writes, and the human directs and verifies.” In May 2026, the research firm Gartner also renamed its evaluation report for this category from “AI Code Assistants” to “Enterprise AI Coding Agents.”

The working environment for running agents is sometimes called an ADE (Agentic Development Environment). Warp started using the term with its Warp 2.0 announcement in June 2025, and it has not yet become a common term across the industry. In this article, we call the market as a whole “AI coding agents.”

What matters is that the axis of competition in this market changed three times during 2026.

1. Spring (through May): the fight over terminals and IDEs. Which screen do you run agents on?
2. Summer (June-August): cloud and parallelism. Isolated VMs, hundreds of parallel agents, delegating from mobile.
3. Fall (September onward): always-on agents and governance. Agents that start working on their own from Slack or a schedule, while organizations control cost and risk.

The landscape: five camps

1. Model providers (Anthropic / OpenAI / Google / xAI)

• Claude Code / Cowork (Anthropic): The de facto reference point. On September 16, 2026, Cowork was merged into Claude chat (rolling out over several weeks starting with Pro/Max, with Team and Free to follow), and the act of “using an agent” as a separate step disappeared. You just talk to Claude normally and multi-step tasks run. For non-engineers, it is currently the product with the lowest barrier to entry.
• Codex (OpenAI): Its distinguishing feature is that the CLI, desktop app, IDE extension, and cloud tasks are all covered by a single ChatGPT subscription. The Codex desktop app arrived on macOS on February 2, 2026, added Windows on March 4, and was merged into the ChatGPT desktop app on July 9. As of October 9, the latest CLI is 0.162.1, and since 0.161.0 on October 7 the default model is GPT-6.1 Sol. The pricing page says that Plus ($20/month) and above covers the web, CLI, IDE extension, iOS, and cloud integration, while Free and Go ($8/month) are positioned for lightweight work (the pricing page does not say whether cloud tasks are available on Free or Go, as of 2026-10-14). Codex is in the Leaders category of Gartner’s May 2026 evaluation.
• Google Antigravity: Google entered in earnest by announcing Antigravity 2.0 at Google I/O in May. Its strengths are Gemini and Google’s search assets (the built-in SEARCH_WEB tool). Individuals can start on the free Individual plan ($0/month), with higher limits tied to paid Google AI plans: Google AI Pro ($19.99/month) comes with standard rate limits, and AI Ultra (two tiers, $99.99/month and $199.99/month) raises them (US prices, confirmed 2026-10-14).
• Grok Build (xAI): Launched in May. With the addition of Workflows on July 23, it implemented what is currently one of the largest-scale designs: fanning out to hundreds of parallel agents, then verifying and consolidating (128 by default, up to 1,024 agents). Its GitHub repository, open-sourced in July (Apache-2.0), gained stars quickly after release and had passed 27,000 as of October 2026.

2. IDE-based (Cursor / JetBrains)

• Cursor: It has almost completed its shift from an IDE company to an agent-platform company. Building on cloud agents that run in isolated VMs (February), it added in June a feature that runs subagents in dedicated VMs plus an iOS app, and with “Projects” in September it became an always-on service that starts autonomously from Slack, schedules, and PR events. Cursor Start, a plan launched in July exclusively for India (₹649/month, about $7), was also the first demonstration of an economic structure in which restricting users to in-house models (Composer and Grok 4.5) makes a price of roughly one third of the then-current Pro plan ($20) viable. Cursor then overhauled its pricing on October 13, 2026: the pricing page now lists Free, Plus $8, Premium $30, Super $100, and Ultra $200 (per month), and Teams from $30/user (confirmed 2026-10-14).
• JetBrains Air: A reversal that only a veteran vendor could make: redefining 26 years of IDE assets as an “operating system for managing agents.” In September it announced a three-product lineup (Air in IDEs / Air Teams / Air Governance). It manages multiple agents in one place, including Claude Agent, Codex, Junie, Copilot, and OpenCode, and you can also add any agent that supports ACP (Agent Client Protocol).

3. Terminal-based (Warp)

• Warp: The one that lowered market prices and the barrier to entry a notch by going open source in April. With its May pricing change, its cloud execution platform Oz moved to platform-credit billing per cloud agent run. With the “software factory” concept (Warp Factories), which entered early access at the end of August, it aims to automate the entire SDLC with agents.

4. Parallel runners / downstream tools (Emdash / Conductor / GitKraken Kepler)

This is the layer that runs multiple agents (Claude Code, Codex, Cursor, and others) side by side, reviews their diffs, and carries the work through to merge.

• Emdash: Free and open source. Runs many CLI agents, including Claude Code, Codex, and Cursor, in parallel with git worktree isolation.
• Conductor: A native macOS GUI with a diff-first review UI. From free (BYOK) to Pro at $50/month.
• GitKraken Kepler: A new category that handles the downstream stage, getting agent-written code all the way to production. Free during the preview period.

“Writing code got cheap. Who is responsible for delivering and merging it?” is the hot spot of late 2026, and this camp is going after exactly that.

5. SDLC platforms / runtimes (GitHub Copilot / Augment Cosmos / Vercel eve / Microsoft Conductor)

• GitHub Copilot: Because it owns the place where code lives (GitHub), it offers a cloud agent (now named Copilot cloud agent) that takes work assigned from an Issue and the like and produces a PR. On February 4, 2026, it announced Agent HQ, which lets you use Claude and Codex from inside Copilot (initially a public preview for Pro+ and Enterprise). The cloud agent’s Linear integration became generally available on July 23, and the Jira integration started in public preview on March 5. On October 1, Copilot CLI and the Copilot app (macOS / Windows) gained computer use, which operates desktop apps, in public preview. Pricing is Pro $10, Pro+ $39, and Max $100 (per month); the cloud agent is not available on Free. Copilot is in the Leaders category of Gartner’s May 2026 evaluation.
• Augment Cosmos: Promotes its spec-driven Intent into a platform covering triage, implementation, review, and deployment. The company’s Auggie CLI makes its quality case with numbers: 51.80% on SWE-bench Pro (using Claude Opus 4.5; for the same model, Cursor scored 50.21% and Claude Code 49.75%).
• Vercel eve: An open-source runtime built on the idea that “an agent is a directory.” According to Vercel’s CEO, agent-initiated deployments made up about 29% of deployments on Vercel when eve was announced in June, and reached about half in July (less than 3% a year earlier).
• Microsoft Conductor: “Deterministic routing,” in which a YAML declaration fixes the workflow structure at definition time. It is a contrarian bet against the prevailing trend of dynamic orchestration, aimed at enterprises that prioritize auditability and reproducibility.

Three structural shifts in 2026

1. A wave of free and open-source tools

Warp went open source (April), Emdash and Conductor are offered free, and GitHub Copilot for JetBrains added support for selecting the Codex agent (July, public preview). The “box that runs agents” is becoming hard to sell, and vendors are moving their monetization points to cloud execution, organizational governance, and the downstream stage (merge).

2. Costs become real

Microsoft cut most of its direct Claude Code licenses in some internal divisions and moved them to Copilot CLI by the end of June (reported in May). The case showed that enterprise cost management is still immature. Meanwhile, Anthropic held off (paused) its separate billing for the Agent SDK on its scheduled start date of June 15, and the market is still in a “customer acquisition phase.” From the user’s side, this is a period when vendors are handing out access at something close to a loss.

3. Reaching non-engineers

Cowork’s merge into chat (September), Cursor’s Slack integration and iOS app, and the cost visibility JetBrains Air Governance gives to managers. A path for “people who are not engineers to delegate work to agents” is where vendors are investing in common in late 2026. This publication will keep following this “delegator’s perspective” in our monthly tracking.

Which one should you choose? (Our view as of October 2026)

About updates

This article is based on the tracking we have accumulated since May 2026, and we add to it each time a monthly report is published. Prices and features change often, so check each vendor’s official page for the latest information before making a decision.

Last verified: 2026-10-14. For sources, see the source list in each monthly tracking article.