Tool guides

Which AI tool does each job?

Seven jobs a team does with AI, and the tool each vendor offers for it.

7 jobs3 families of tools25 dated facts4 manualsLast checked 2 October 2026

Who this page is for, and how to use it

For

Anyone about to hand a job to an AI tool, and anyone asked which one the team should use.

Use it to

Find the tool each vendor offers for a job, see how finished it is, and open the vendor's own page.

How

Find the job down the side: drafting, holding the case, testing a prompt, building, handing off, browsing or a scheduled check.
Read across. Each cell names the tool, its status tag and one fact, and links to the page the fact came from.
For the tools in a manual, the cell links to it: five moves with the real prompt, three settings and the traps.

Seven jobs a team does with AI, down the side, and the tool each of three families offers for it, across. Each cell names the tool, its status in the vendor's own word where it gives one, and links to the vendor's page.
The job	Claude	ChatGPT and Codex	Google
Think it throughDraft, reason and research, with sources	Research Research runs many searches that build on each other, across the web and your connected sources, and returns a cited report in minutes. In the manual Claude at the desk	Deep research Deep research plans, reads many sources and writes a report with citations, and you can review and edit its plan before it starts.	Gemini API, grounded in Search With the Gemini API's Google Search tool the model runs the searches itself and answers with citations that link each span of text to its source. In the manual Google AI Studio and Jules
Hold project knowledgeKeep the case's documents and rules in one place	Projects A project is a self-contained workspace with its own chats and knowledge base, where you upload documents and set the project's own instructions. In the manual Claude at the desk	Projects A project keeps related chats, files and instructions together so every chat in it starts with the same context, and a shared project uses project-only memory. In the manual ChatGPT and Codex	Gemini Notebook NotebookLM was renamed Gemini Notebook on 16 July 2026, and each notebook is getting a secure cloud computer that runs code for analysis grounded in your sources.
Test a promptTry a prompt and its settings before any code	The Console playground The Console's Workbench is now the playground, which keeps your current draft in your browser and has no saved prompts or prompt versions. In the manual Claude at the desk	The Playground OpenAI's prompt engineering guide says to use the Playground to develop and iterate on prompts. In the manual ChatGPT and Codex	Google AI Studio AI Studio's playground takes a system instruction and a chat with a chosen model, with run settings at the side and a Get code button. In the manual Google AI Studio and Jules
Build in the repoRead, change and test the code	Claude Code Claude Code is a coding agent that reads a codebase, edits files, runs commands and uses git. In the manual Claude in the repo	Codex OpenAI's Codex CLI runs Codex in a repository from your terminal, where it inspects code, edits files and runs commands. In the manual ChatGPT and Codex	Google Antigravity Google Antigravity comes as a desktop app, Antigravity 2.0, a command line tool and an IDE, and in the desktop app agents work within a project's folders and repositories.
Hand off a taskLet an agent work while you are away	Claude Code on the web A cloud session runs on an isolated machine and keeps going after you close the laptop: start one with `claude --cloud "task"` and pull it back with `claude --teleport`. In the manual Claude in the repo	Codex Cloud Codex Cloud runs coding tasks on OpenAI-managed machines, each in its own workspace, and keeps going while your computer sleeps. In the manual ChatGPT and Codex	Jules experimental Jules is an experimental coding agent that works on a GitHub repository in a virtual machine while you move on, and shows a plan for you to approve before it changes code. In the manual Google AI Studio and Jules
Act in a browserClick through a screen as a person would	Claude in Chrome Claude in Chrome is a browser extension that reads pages, clicks, types, fills forms and moves between tabs, and its Chrome side panel is in beta. In the manual Claude in the repo	ChatGPT Work Work, which replaced agent mode, browses with a cloud browser or the desktop app's own, and signs in to sites through a form the model cannot read. In the manual ChatGPT and Codex	Google Antigravity Antigravity 2.0 is a desktop app for managing agents that can interact with Chrome and create artifacts and implementation plans.
Run on a scheduleRepeat a check without being asked	Scheduled tasks A scheduled task is a saved prompt that Claude runs hourly, daily, weekly, on weekdays or on demand, in the cloud, so your computer can be off. In the manual Claude at the desk	Scheduled tasks ChatGPT runs a prompt once, on a schedule, or when a new Gmail message, a Slack message or GitHub pull request activity arrives. In the manual ChatGPT and Codex	Google Antigravity In Antigravity 2.0 the command `/schedule` runs an instruction once at a set time or on a recurring schedule, as a scheduled task.
Prices, limits and modelsThey change every month, so none is printed here	Plans and prices Models	Plans and prices Models	Plans and prices Models
Four more names a team meets, beyond the three families
CursorBuild in the repo	cursor.com Cursor describes itself as a coding agent that helps you understand your codebase, plan and build features, fix bugs and review changes.
GitHub Copilot cloud agentHand off a task	docs.github.com Copilot cloud agent researches a repository, plans and makes changes on a branch in its own short-lived GitHub Actions environment, and you create the pull request when you are ready.
KiroBuild in the repo	kiro.dev Kiro turns a feature into a spec of requirements.md, design.md and tasks.md, then runs the tasks and shows the status of each.
GitHub Spec KitBuild in the repo	github.com Spec Kit is GitHub's open source toolkit that gives AI coding agents a set process: a constitution once per project, then specify, plan, tasks, implement and converge for each feature.

The tag is the vendor's own word for how finished a thing is, and it changes what you can promise a team. No tag means the vendor's page gave none.

A tool's name links to the vendor's own page, the one its fact was checked against. The number beside a fact in a manual leads to the same page and its date.

Four manuals put the tools in the airline team's hands

Each one is five moves with the real prompt or command, the three settings that matter, its traps, and every fact it used with its source and date.

5 moves · 31 dated factsClaude at the deskChat, Projects, skills, connectors and scheduled tasks, as the people who write the spec use them.5 moves · 31 dated factsClaude in the repoClaude Code in the terminal, VS Code, the cloud and Chrome, as the people who build use it.5 moves · 36 dated factsChatGPT and CodexChatGPT's projects and scheduled tasks for the spec, and Codex in the repository, on GitHub and in the cloud, as the team uses them.5 moves · 32 dated factsGoogle AI Studio and JulesAI Studio to try a prompt before any code, and Jules for the chores the engineers hand off, as the team uses them.

Start where the spec gets written, at the desk.

Claude at the desk Or grow the spec in the lab