AI Agents
Tools that plan and execute multi-step tasks with access to files, browsers and APIs.
13 tools
- ChatGPTThe most widely used assistant, with image generation and a large plugin ecosystem.Official: chatgpt.comfrom $20/mo
- ClaudeAssistant known for long-context reasoning and writing that needs less editing.Official: claude.aifrom $20/mo
- GeminiGoogle's assistant, tightly integrated with Search, Gmail and Workspace.Official: gemini.google.comfrom $20/mo
- Claude CodeTerminal-based coding agent that runs commands and edits files directly.Official: claude.com/product/claude-codefrom $20/mo
- Codex CLIOpenAI's open-source terminal coding agent, rewritten in Rust.Official: developers.openai.com/codex/clifrom $20/mo
- Claude CoWorkClaude working on your files in the background instead of waiting for prompts.Official: claude.com/coworkfrom $20/mo
- Gemini CLIGoogle's open-source terminal agent, with the most usable free tier of the group.Official: github.com/google-gemini/gemini-cliFree tier available
- opencodeOpen-source terminal coding agent that runs on whichever model you point it at.Official: opencode.aiFree
- PiA minimal terminal coding harness you extend in TypeScript instead of configuring.Official: pi.devFree
- Copy.aiGo-to-market workflows that automate sales and marketing content.Official: copy.aifrom $49/mo
- herdrA background server that owns your agents' terminals, and knows which one is blocked.Official: herdr.devFree
- CrushA terminal coding agent from the team that makes terminal interfaces look good.Official: charm.land/crushFree
- gooseOpen-source agent that installs, executes and tests — not just suggests code.Official: goose-docs.aiFree
Agents are the newest and least settled category here. The pitch is that instead of answering a question, the tool carries out a task: reads files, browses, calls APIs, writes code, checks its own work. When it goes well the leverage is real. What separates agents from every other category is that they act rather than suggest, and that changes what you need to think about before adopting one.
What to look at when choosing
Permissions are the specification
An agent is defined by what it is allowed to touch — your filesystem, a browser session logged into your accounts, an API key that can spend money, a repository it can push to. The useful question is not what it can do but what happens when it does the wrong thing. Prefer tools that ask before irreversible actions and that let you scope access narrowly, and start with the smallest permission set that lets it be useful.
The failure mode is quiet, not loud
Agents rarely crash. They report success on a task they did partially, or plausibly, or in a way that looks right until you check. That is much harder to catch than an error message, and it is the single biggest practical cost of running one. Assume every result needs verification, and pick tasks where verifying is cheaper than doing.
Cost scales with steps, not with questions
A chat exchange is one call. An agent working through a task may make dozens, re-reading context each time, and a loop it fails to break out of can burn through a budget quickly. Look for tools that show step counts and spending, and that let you cap a run — visibility here is not a luxury.
When you probably do not need one
If the task has fixed steps you can write down, a script or an existing automation tool will do it faster, cheaper and identically every time. Agents earn their cost on tasks where the steps depend on what is found along the way — and they are a poor fit anywhere a wrong action is expensive to undo.