New vertical

AI Agent and Coding Agent Plans Compared

Compare task-oriented AI agent plans such as Devin, Claude Code, Codex, Replit Agent, Manus and Genspark. The useful comparison is not just monthly price. It is concurrency, long-running work, browser access, code execution, filesystem access and human approval flow.

Agent plans should be compared by capability

Agent products sell completed work, not just tokens. V1 adds the category and surfaces; curated data and scrapers can follow without disturbing API pricing.

Coding Agents

IDE, CLI, repo indexing, terminal execution, PR edits and background tasks for Cursor, Windsurf, Copilot, Claude Code, Codex and Replit Agent.

IDE / CLIRepo contextShell access

Autonomous Agents

Agents that browse, plan, call tools and ask for approval when needed. Useful for Devin, Manus, Genspark and ChatGPT Agent style products.

BrowserLong taskApprovals

Workflow Agents

Automation agents for ops, sales, research and internal workflows. The comparison is integrations, triggers, logs, permissions and handoff.

IntegrationsHandoffAudit log

Priority products

Cognition

Devin

Autonomous software engineering agent. Track task capacity, concurrency and approval flow.

CodingAutonomous

Anthropic

Claude Code

CLI coding agent. The plan relationship with Claude Pro / Max / Team Premium must be explicit.

CLIRepo

OpenAI

OpenAI Codex

Coding agent bundled with paid ChatGPT tiers and developer workflows.

CodingChatGPT

Manus

General-purpose task agent. Track credits, browser, files and long-running work.

General agentCredits

Catalog tracker

V1 tracks public products, units and capabilities. Verified exact prices will move into plans and usage_prices.

Product

Status

Unit

Best for

Devin

Cognition

autonomous codingbrowser

available

seat, task, compute

Subscription / task capacity

Delegating end-to-end software tasks to a long-running agent.

autonomous codingbrowsershell

Claude Code

Anthropic

CLIrepo context

available

messages, prompts, subscription

Bundled with paid Claude tiers / usage limits

Repo-aware coding from a terminal workflow.

CLIrepo contextterminal

OpenAI Codex

OpenAI

coding agentrepo tasks

available

subscription, tasks

Bundled with paid ChatGPT tiers

Coding tasks inside the OpenAI / ChatGPT ecosystem.

coding agentrepo tasksChatGPT integration

Replit Agent

Replit

app generationdeployments

available

credits, seat

Subscription / usage credits

Building and deploying small apps in one workspace.

app generationdeploymentsdatabase

Manus

browserfiles

available

credits, tasks

Subscription / credits

General web and research tasks with file outputs.

browserfilesresearch

Genspark AI Agent

Genspark

researchbrowser

available

credits, tasks

Subscription / credits

Research and content tasks that need browsed sources.

researchbrowserpresentations

GitHub Copilot Coding Agent

GitHub

GitHub issuespull requests

available

seat, requests

Copilot subscription / enterprise controls

Teams that already live in GitHub pull requests.

GitHub issuespull requestsrepo context

Cursor Background Agents

Cursor

IDErepo context

available

requests, credits, seat

Subscription / usage pool

Developers using Cursor as their primary editor.

IDErepo contextbackground tasks

Windsurf / Cascade

Cognition

IDEagentic coding

available

credits, requests, seat

Subscription / usage pool

Agentic coding inside a dedicated AI IDE.

IDEagentic codingrepo edits

Lovable

app generationfrontend

available

credits, projects

Subscription / credits

Shipping web app prototypes quickly from prompts.

app generationfrontendSupabase integration

Fields to track

Monthly / annual / seat price

Included credits or compute hours

Concurrent agents

Browser and filesystem access

Code execution and deployment

Human approvals and enterprise controls

Why this is its own vertical

Agent plans are sold by tasks, credits, compute hours or seats. Folding them into chatbot subscriptions would hide the thing buyers actually compare.