Describe the task. Get the right AI workflow.¶
No API key required. Start with the AI app you already use.
This site is free, runs in your browser, and works on whatever AI subscription you already pay for — ChatGPT, Claude, Gemini, Grok, Perplexity, GitHub Copilot, or even "I don't know what I'm using."
Describe a task. Pick the app you already have. Get a complete, copy-pasteable workflow plus a level-up ladder that shows you the next useful capability — saved prompts, projects, memory, custom assistants, skills, scheduled tasks, and (only if you want) developer tools.
No command line, no install, no signup beyond the AI app you already use.
Pick the right starting point¶
You do not need to read this site front to back. Start with the row that matches you.
| If you are... | Do this first | You should have by the end |
|---|---|---|
| An absolute beginner | Open the Task Builder, leave Beginner Mode on, and describe one real task. | A prompt to paste, exact clicks, and a safe fallback using the AI app you already have. |
| Learning Claude specifically | Open the standalone Claude site and follow Day 0 before exploring advanced features. | Account instructions, one Project, a reusable prompt kit, and a safe path to Skills, connectors, and Claude Code. |
| Trying to get more value from one subscription | Run the 7-mission learning path. | A portable AI profile, reusable prompt, Project/Gem/Space, citation habit, and evaluation routine. |
| Choosing between ChatGPT, Claude, Gemini, Grok, Perplexity, or Copilot | Use the Capability map and Mastery hub. | A deliberate tool choice for each recurring workflow instead of defaulting to whichever tab is open. |
| Building a real agent or coding workflow | Start with Agent Factory, Evals & safety, and Starter kits. | A scoped agent spec, tool allowlist, eval set, red-team probes, and launch checklist. |
| Responsible for a team or sensitive data | Read Team path, Safety baseline, and Human-in-the-loop. | A permission model, review gates, data boundaries, and off-switch before any automation runs. |
-
Start the Task Builder
Describe a task. Pick the AI app you already use. Get a recommended workflow, a copy-paste prompt, exact "what to click" steps, a level-up ladder, and a fallback if your plan doesn't have a feature.
-
Claude
Standalone step-by-step Claude setup for beginners, then Projects, memory, Artifacts, Skills, connectors, Claude Code, and expert reliability habits.
-
Try the 7-mission learning path
Seven 30-minute missions that take any AI subscription from "I have it" to "I'm getting full value." Works for ChatGPT, Claude, Gemini, Grok, Perplexity, Copilot, or anything else.
-
Master the AI you already have
Beginner-to-expert tracks for ChatGPT, Claude, Gemini, Grok, Perplexity, GitHub Copilot, and any other AI chat. Each track ends with a guided exercise.
-
Capability map
Cross-product comparison: chat / projects / memory / files / custom assistants / skills / scheduled actions / agents / coding agents / API. Plan-availability tags and fallbacks per cell.
-
No-code automations
Native scheduled actions in ChatGPT, Claude Cowork, and Gemini — plus the universal calendar-reminder fallback for everyone else. Manual-first; safety-first.
-
Memory & preferences
A portable AI profile you can paste into any tool. Templates for global memory, project memory, and the conversations you'll have with your AI about its own memory.
-
Beginner examples
Thirteen guided workflows you can copy today: difficult email, meeting prep, study plan, trip plan, weekly report, research, custom assistants, scheduled tasks, and more.
-
Compare AI surfaces
Chat vs. Project vs. Custom GPT/Gem vs. Copilot/coding agent vs. Skill vs. Automation vs. Agent. A decision matrix that maps your task to the right shape.
-
Browse recipes
41 complete agent recipes, each with a job statement, prompt, tool allowlist, evals, red-team probes, and a safe-launch checklist.
-
Download starter kits
Nine copyable kits — universal spec, Claude Code, Codex, Gemini Antigravity, OpenAI Agents SDK (Python + TS), MCP server (Python + TS), local script.
Three example outputs from the Task Builder¶
These are real shapes the Task Builder produces. Click any to jump to a worked example.
Weekly research summary¶
Input: "Every Monday I want a one-page summary of the past week's newly-deposited preprints in my field, with three highlighted papers and questions for journal club. I have ChatGPT and Claude. Inputs come from a folder of PDFs I save during the week. Output goes to my Obsidian vault. Manual run only."
The Task Builder routes this to a Project (ChatGPT or Claude) plus a prompt plus a manual playbook — not an agent, not a scheduler.
It produces: a system prompt, a Project description, a memory/preferences block ("MIND-diet research, NEJM citation style, no full-text reproduction"), an output schema, a 5-case eval set, three red-team probes (e.g., "what if a PDF is image-only?"), and a Monday playbook with the exact 60-second steps.
→ See this output as a worked recipe: Literature triage
Repository maintenance¶
Input: "Keep
my-org/my-repohealthy: dependency updates, lint fixes, and tiny refactors that pass CI. I have Claude Code and GitHub Copilot. Anything that opens a PR has to be reviewed by a human."
The Task Builder routes this to a coding agent workflow (Claude Code locally for the dev loop, GitHub Copilot cloud agent for hands-off PRs), with a CLAUDE.md / AGENTS.md scaffold, an HITL gate that forbids git push to main, an eval set ("PR diff is small," "tests pass," "no new dependencies without review"), and a maintenance cadence.
→ See this output as a worked recipe: Repo maintenance
Email and calendar triage¶
Input: "First-thing-in-the-morning brief: triage overnight email into Now/Later/Reference/Trash, and write a one-paragraph prep note for each meeting today. Connectors: Gmail and Calendar via ChatGPT. Drafts only, never send."
The Task Builder routes this to a Project + connectors (read-only) plus a draft-only refusal block so the agent never sends a reply on your behalf. It produces a system prompt, a connector permission spec, an eval set with at least one ambiguous "is this Now or Later?" case, and three red-team probes (prompt injection from email body, calendar invite spoofing, forward-this-to-X attempts).
→ See worked recipes: Email triage and Calendar meeting prep
What's inside¶
-
Start Here
Decide whether to build an agent at all, then pick a setup path.
-
Platforms
Setup guides for Claude, Gemini, ChatGPT, Grok, Codex, Copilot, MCP, and more.
-
Quickstarts
Five end-to-end recipes you can finish in 30–60 minutes.
-
MCP & connectors
Concepts, installing servers, writing your own server, security.
-
Browser & computer use
Anthropic, OpenAI, the
browser-uselibrary, and operating boundaries. -
Orchestration
Single-agent loops, multi-agent patterns, local-first, state and memory.
-
Evaluation & safety
Eval sets, red-team workflows, safety checklists, HITL, incident response.
-
Agent Factory
A repeatable workflow to design, rank, build, eval, port, and launch agents.
-
Templates
Reusable agent specs, prompts, eval rubrics, safety checklists, PRDs.
-
Reference
Glossary, model freshness, source map, source audit, roadmap, decision records.
Boundaries this guide enforces¶
- No real secrets in any example. All API keys, tokens, and personal info are placeholders.
- No background automations or schedulers run by default. Every cron / launchd / Task Scheduler example is opt-in and marked as such (ADR 0003).
- No tracking, analytics, or telemetry on this site. The Task Builder runs entirely in your browser; nothing leaves your machine.
- Every product page is dated and sourced. Specific model names, prices, and rate limits drift fast — see Model freshness.
License¶
MIT. All trademarks belong to their respective owners. This guide is independent and is not endorsed by Anthropic, OpenAI, Google, xAI, GitHub, or Microsoft.