Source: raw/Understand_the_FULL_Claude_Ecosystem_in_One_Video.md (primary, for the four-layer map); raw/Claude_Chat_vs_Cowork_vs_Code_-_Which_One_Should_You_Use.md; raw/Use_Claude_Better_Than_99%_of_People_Beginner_to_Pro.md; raw/The_3_Most_Powerful_Claude_Features_Beginners_Don_t_Know_About.md; raw/Full_Claude_Tutorial_-_Beginner_to_Advanced_in_19_Minutes.md; raw/Full_Claude_Cowork_Tutorial_-_The_AI_That_Replaces_Your_Assistant.md

Creator: Futurepedia Platform: YouTube Duration: not recorded in the transcripts URL / Published:

Futurepedia’s six Claude tutorials from March to August 2026 build one beginner path: get chat right, add context through projects and connectors, turn repeat work into skills, then delegate in Cowork and build in Claude Code. The most reusable piece is the June ecosystem video, which sorts every Claude feature into four layers so a beginner knows where a new feature fits. This article consolidates the series, keeps what still applies, and dates what has changed.

Historical context (March to August 2026)

Every video predates the current wiki state. On 2026-09-16 Anthropic merged chat and Cowork into one Claude, so there is no longer a mode to pick on plans the merge has reached (see Claude Cowork). The current models are Opus 5.5, Sonnet 5.5 and Fable 5.1. The videos were recorded on Opus 4.x / Sonnet 4.x (March to June) and Opus 5 (August). Read the “chat vs Cowork” sections below as the mental model behind the merged product, not as UI instructions.

Key Takeaways

  • Chat to think, Cowork to delegate, Code to build (and, from June, Design to create). Futurepedia repeats this rule in three videos. Open chat when the work is the conversation, Cowork when you would hand the job to an assistant, and Code when the answer is “I wish a tool existed for this.”
  • The signal that you picked the wrong mode is assembly work. If you are copying chat output into docs or stitching pieces together, that job should have been delegated (July 2026). This test still works after the merge: ask for the finished deliverable, not the parts.
  • Chat and Cowork differ in three ways (July 2026): the finish line (chat stops when it has answered; Cowork stops when the goal is met, “2 minutes or 40”), the workspace (Cowork keeps a computer it builds on for the whole job), and scheduled tasks (Cowork runs without you prompting it).
  • Cowork and Code are one agentic architecture. “The only thing that changes is what gets loaded into it. Just your files and connected apps and it’s Co-work. Add a code repository and it’s Code” (July 2026).
  • Every feature sits in one of four layers: products, a universal layer (skills, connectors, plugins), a context and memory layer (projects, memory, CLAUDE.md), and an autonomous layer (computer use, Dispatch, scheduled tasks, routines). Table below.
  • Three prompting habits carry the most weight: dump all the context you have, ask Claude to interview you for the context you forgot, and push back on the first answer. Futurepedia’s view is that most older prompt-engineering advice “just isn’t necessary” now (August 2026).
  • Make skills from finished conversations. Iterate in chat until the output is right, then say “package that process up as a skill.” Claude then gets it right on the first try next time and calls it automatically when the prompt fits.
  • Leave account-level custom instructions blank and put instructions in projects. Futurepedia says account-wide instructions that cannot fit every topic “can lead to some weird responses” (April and August 2026).

The four-layer map (June 2026)

LayerFeatureOne-line job, in Futurepedia’s words
ProductsChat · Cowork · Code · DesignThink · delegate · build · create
Universal (works in chat, Cowork and Code)SkillsRepeat a task the same way (“expertise packages”, “a folder of instructions”)
ConnectorsConnect the tools you already use, with a per-tool permissions list
PluginsBundle skills and connectors around a role or tool (Figma, legal)
Context and memoryProjectsPersist context: instructions, files, daily-refreshed project memory
Account memoryLearn who you are across chats (chat only in June 2026)
CLAUDE.mdBrief Claude Code (or a Cowork folder) on how to work; /init writes the first one
AutonomousComputer useClaude operates your desktop (“still early”)
DispatchRun and redirect a Cowork session from your phone
Cowork scheduled tasksRun a task on a schedule, such as a morning brief
RoutinesTrigger Claude Code from schedules, GitHub events or API calls with your computer off

What has moved since June: projects in chat and Cowork were separate in June (“they don’t sync”, though a chat project could be imported) and are now one kind; account memory covered chat only in June and has covered Cowork since the merge (see Claude Cowork, October 2026 section).

Where each product ran, by video date

Video dateWhat Futurepedia showedStatus now
2026-03-12Cowork only in the desktop app, Pro or Max, “research preview”. No memory across Cowork sessions. Scheduled tasks need the desktop app open; a missed run starts the next time you open it.Cowork runs on web and phone in the cloud; memory spans the account
2026-04-08Chat on the $20/month plan is “plenty for most people”; move up for heavy Cowork or Code use. Desktop app has three tabs: chat, Cowork, Code.Tabs replaced by the merged product
2026-06-24Cowork and Code “both run directly on your computer”; you point them at a folderSuperseded for Cowork
2026-07-27Cowork and Code on web and mobile in beta, Max plan and up only, Pro “soon”. Web/mobile Cowork gets a fresh cloud sandbox; a web or phone session can reach local files only while the desktop app is open on that computer. Code on web/mobile has two options: remote control of a session running on your computer, or cloud sessions that need a GitHub repo.Cloud Cowork reached Pro and Max; merge from 2026-09-16
2026-08-16Same split: cloud Cowork on the web, local-folder Cowork in the desktop app. For Code, start in the desktop app with a local folder and plan mode.Desktop is still where local-folder work happens

The Google Drive workaround from July and August still holds: put the files you want Cowork to use in a Drive folder, connect the Google Drive connector, and tell Cowork to read from and save to that folder. Then the files are reachable from every surface whether or not your computer is on.

Prompting habits

  • ICC: instructions, context, constraints (April 2026). Instructions say what to do. Context gives your role, objectives and background (“err on the side of more context”). Constraints set rules, tone, length and format, plus an example output where you have one. Order does not matter. Futurepedia’s before-and-after: “Recommend five ways to implement AI in my marketing agency” becomes the same instruction plus business details, plus “low cost, no technical expertise, numbered list with a one-sentence explanation each.”
  • Context interview. End the prompt with a request for Claude to ask you whatever it needs to know first. “It will know what it needs better than you do.”
  • Ground first, ask second. Separate searching from answering. First: “Research [tool] and make sure you’re up to date on all its capabilities.” Then ask your actual question. Futurepedia notes that web search is on by default but Claude does not always use it; in a script critique it called a real model name a placeholder until told to search.
  • Push back by default. “It’s better to assume it’s not right… rather than to assume it’s right all the time.” Ask for the gaps and flaws in your thinking and in its own.
  • Use one-sentence lists to triage. Ask for a short list with one sentence per item, then ask Claude to expand only the promising ones.

These map onto this wiki’s existing prompting material; see Module 2 — Prompting for the RCTF version.

Model and effort picks (August 2026, Opus 5 era)

Futurepedia’s rules, recorded before the September 5.5 releases:

  • Haiku: “basically never” used.
  • Pro plan: Sonnet as the daily driver, Opus for harder reasoning. Max plan: Opus most of the time, especially in Cowork and Code; Sonnet for some creative work.
  • Fable only when Opus is stuck: “way more token intensive for not that much extra.”
  • Effort: medium for most daily tasks, high for coding and analysis, extra high for agentic or multi-step coding sessions. Extra high on a simple task “will not only take way longer, but it doesn’t improve the result. It might even be worse.”

This matches the Sonnet-first routing and the medium default recorded in Which Claude Model for Which Job. For the current lineup and prices, use that article.

Four workflows from the series

  1. Weekly special, end to end (August 2026). In one chat, Futurepedia had Claude read the week’s sales sheets, propose a weekend special, write the copy and generate an image through a custom Higgsfield connector. After several rounds of refinement: “package that process up as a skill called weekly special.” Then in Cowork: run that skill every Friday morning. Claude has no image generator of its own, so the connector supplies one (Magnific and Runway are named as alternatives).
  2. YouTube script pipeline (April and May 2026). A scripting project with custom instructions plus four skills: a script critique that returns an artifact with notes after each section (with instructions not to rewrite anything), an intro writer that drafts three options, a title skill that gives 10 options or expands a seed title, and a description writer. Futurepedia says the four save “hours every single week.”
  3. Hook-writing skill with skill-creator (May 2026). Add Anthropic’s skill-creator from Customize → Skills → + → Browse. Turn off existing skills that might fire on the same prompt. Generate hooks, say which ones work and why, repeat, then ask Claude to create “a skill that would have gotten me this response on the first try.” Then upload a reference file (his hook-template database) through “edit with Claude.” The resulting skill held voice rules, banned structures, banned phrases and tone calibration.
  4. Calendar notes as instructions (March 2026, Cowork). Write what each meeting needs into the calendar event’s notes. Then give Cowork one short prompt: “Check my calendar for meetings and help me prepare… Read through the notes and follow all of the instructions.” With Gmail, Google Calendar and Claude in Chrome turned on, Cowork researched the web, wrote a one-page brief, built an editable PowerPoint and drafted a Gmail email, working on the brief and the deck in parallel. (The demo itself was a joke acquisition of Anthropic.)

Try It

  1. Sort your last ten chats into think, delegate and build. Any chat where you copied the output somewhere else was a delegate job. Rerun one as a single request for the finished file.
  2. Run a context interview on your next real prompt. Add: “Before you answer, ask me any questions you need to give the best result.”
  3. Turn one refined chat into a skill. Pick a chat you iterated on until the result was right and say: “Package this whole process as a skill called .” Test it in a fresh chat without naming it.
  4. Clear your account-level instructions and move them into one project for your main work stream. Check that project’s memory after a week and delete anything stale.
  5. Set up the Drive home base for cloud tasks. Create one Google Drive folder, connect the Google Drive connector, and tell Claude to read inputs from and save outputs to that folder.

Implementation

Tool/Service: Claude app (web, desktop, mobile), with Cowork and Claude Code, and the Higgsfield and Google Drive connectors in the examples. Setup: Settings → Capabilities: turn on “search and reference chats” and “generate memory from chat history”; “Import memory from other AI providers” brings memory over from ChatGPT or Gemini (see ChatGPT to Claude Migration). Customize → Connectors → Add → Browse connectors, or Add custom connector with a name and the vendor’s URL. Customize → Skills for built-in and custom skills. Cost: Futurepedia (April 2026) quotes the $20/month Pro plan as enough for most chat users, with Max for heavy Cowork and Claude Code use. Chat is on the free plan; Cowork and Code need a paid plan. He reports a 5-hour rolling usage window plus a weekly cap. Check current limits in Claude Cowork before you rely on them. Integration notes: Projects are walled off from each other by design: a project remembers everything inside it, but other projects and plain chats do not see it. Futurepedia keeps one project per work stream and asks one-off questions outside projects.

Open Questions

  • Video durations are not in the transcripts.
  • Futurepedia’s May 2026 video shows a Customize button on published artifacts that opens a copy in a new chat for the viewer to remix. This wiki has no first-party source for it, and it is unknown whether it survived the September merge.
  • The April video’s “inline visualizations” with clickable tabs are described only in passing; it is unclear whether they are the same feature as the inline visuals in Artifacts.
  • Research-mode numbers (“5 minutes, 169 sources” in April) are one run reported by the creator, not a benchmark.
  • The effort ladder (medium, high, extra high) is Futurepedia’s practice on Opus 5. Whether it carries over unchanged to Opus 5.5, where thinking is always on, is not tested in these videos.