Stop Moving
The Work.
Six features, one workspace.
A solo owner works as analyst, writer, designer, and project manager at once. Six Claude features, shipped or updated over the past few months, cover those roles inside one workspace: interactive dashboards, native docs with slides and tables, phone notifications, highlight comments, revamped projects, and account memory. The creator says most people have not touched them, and some accounts lack access.
Every demo starts by adding a business folder to the chat: business overview, priorities, brand guide, and marketing files. Outputs come back on brand because the brand guide sits in the folder. He calls the folder optional.
The prompt asks for an interactive dashboard from YouTube analytics, built to say what to do next, not only what happened: best videos and topics, the year's trend in views and watch time, what top performers share, date filters, metric sorting, three recommendations, the brand guide for styling, and nothing written to the folder. Input: a zip of one year of analytics. Run: 19 commands in 15 steps, about 10 to 15 minutes.
A headline stating six videos carried the year, 2.9 million views, a yearly chart with a channel average line, full beginner courses outperforming every other format, traffic sources, and a table sortable by average view duration and click rate. The same method applies to CRM, Instagram, or sales exports.
A follow-up prompt added a chat tab with suggested prompts labeled plan, fix, and risk. The tab needs AI-powered artifacts switched on under Settings, Capabilities, Visuals. Published artifacts stay private until shared. His question on good watch time with low click rate returned seven videos with at least 20 percent watched and click rate under 3.5 percent, flagged as packaging problems. Each question counts toward usage.
The share button invites people by email. Recipients with Claude accounts open the artifact from the invitation and use the chat tab from their own account.
Typing /doc opens a document editor beside the chat: bullet lists, checklists, code blocks, date pickers, dividers, dropdowns, headings, fonts, layouts, and tabs. Checklists are interactive, and Claude reads which items are checked, so “finish the uncompleted tasks” works without a recap. Export goes to Google Docs, Markdown, PDF, or Word.
Typing /slides turned the doc into a branded deck: one idea per slide, big text, ten slides at most, with speaker notes, a present mode, and a full editor for text, shapes, and tables. He puts the build under five minutes.
His account showed no native sheet. A /doc request for a table built a production tracker with one row per task from script to upload, plus owner, due date, and status columns. The editor got assigned by name, pulled from the business overview. He calls the method hacky.
A scheduled task, created from chat, checks AI news daily against his business. The task states a default: most days the honest answer is no video today. Advanced settings hold a notify option sending a one-line summary by push, email, or Slack only when a run finds something worth telling. Same account on phone required. The test run returned: quiet day, nothing worth a video. The task screen reads: requires this computer to be on.
Highlight text, add a comment, and send the comment to Claude or hold the comment. For a batch, leave send unchecked on each one, then type “check my comments and make the changes.” The rewrite touches only the marked sections. Works on native docs and Markdown, and per the creator on PDFs.
A coordinator sits on the left, threads on the right, sorted as waiting on you, idle, or resolved. Settings hold a goal (ongoing or fixed deadline), the coordinator model, memory, and instructions. Demo: four jobs (past analytics, competitor scan, outline, deck) split into threads, with the deck idle until the outline finished. Result: 18 slides published. An auto memory reaches every thread.
Settings, Memory holds toggles for sensitive topics (off on his account), memory from chats, and search of past chats. Type a fact to add one, or import memories from another provider with a copy-paste prompt. Memory spans chats, projects, and scheduled tasks. The video carries a mid-roll ad for his paid program.
The Concept.
You lose hours carrying work between tools. Export the numbers, paste them into a sheet, screenshot the chart, rebuild the deck, re-explain the change. Keep the work in one place. Turn raw exports into a readout ending in decisions. Write one brief and build every output from the brief. Point at the line you want changed. Split mixed jobs into parallel threads. Set alerts to stay silent until something changes. The hours you spent moving files go to the decisions inside them.
Export, paste, screenshot, rebuild, re-explain. A stale number ships when one copy gets missed.
One export, one brief, every output rebuilt from the brief. The alert stays quiet until a number moves.
Your week runs on copies.
Your week runs on copies. A CSV from Instagram, pasted into Google Sheets. A chart screenshot, dropped into Slides. A plan in a doc, retyped as a task list. Then a client asks for one change, and you open four files to make the same change four times. Miss one file and a stale number ships.
The standard advice says find a better app. Another app adds a fifth copy. “6 New Claude Features Your Business Needs” points another direction: the dashboard, the doc, the deck, the tracker, and the alert all live in one workspace, built from the same files and edited in place. The features are the easy part. The hard part is deciding what each output is for before you ask for one.
An export becomes a page ending in three decisions.
One doc every other output is built from.
Highlight the line, write the change, batch the round.
One project, one goal, mixed work in parallel threads.
Alerts on a condition, silence by default, standing facts in memory.
One source. Many outputs.
Every output in this guide comes from the same two inputs: the raw export and the brief. Change a number upstream and every piece downstream gets rebuilt from the fix. Patch a deck by hand and the copies drift apart again.
Turn the export into a decision.
A chart of last month tells you what happened. A readout tells you what to do next. Ask for the readout in four parts: what performed best, how the trend moved over the period, what the top performers share, and three specific next moves. Add filters and sort controls so a reader answers their own follow-up questions.
In “6 New Claude Features Your Business Needs,” one prompt plus a zip of YouTube analytics produced an interactive channel review after 19 commands in 15 steps. A second prompt added a chat tab for live questions against the same data.
Export one raw file you already pull monthly: CRM, social, sales, or site analytics.
Request an interactive dashboard with the four sections and three recommendations at the end. Name your brand file for styling. Tell the model to write nothing to your folder.
Match three totals against the raw export before you read a single recommendation.
Switch on AI-powered artifacts under Settings, Capabilities first. Ask for three suggested questions: plan, fix, risk.
Invite only people with accounts and permission to see the numbers.
Three totals match the export, and you acted on one recommendation within seven days.
Write the brief first. Build the rest from the brief.
The demo chain ran in one direction: a doc plan, then a ten-slide deck from the doc, then a production tracker from the same context. Every output inherited the same facts, so a fix in the brief fixed the next build.
Native docs carry checklists the model reads, so “finish the unchecked tasks” works without a recap. Export to Word, PDF, Markdown, or Google Docs only when the work leaves for someone else.
Start with /doc and one paragraph: goal, audience, deadline, and three must-include facts.
One task per line, one owner per task.
Use /slides on the brief: one idea per slide, big text, ten slides at most, speaker notes on.
A doc table with task, owner, due date, and status. Keep math in a real sheet.
Change the brief first, then rebuild the downstream piece. Never patch the deck alone.
A changed number got fixed once, in the brief, and every output showed the fix.
Point at the line. Skip the recap.
Typing “the section under the table where you said” costs a minute and invites the wrong edit. Highlight the passage, leave a comment, and send. The rewrite touches the marked text and nothing else.
For a full review, leave send unchecked on each comment, then ask for one pass: check my comments and make the changes. One round, one rewrite, one read.
Read the whole draft once before you mark anything.
Each comment names what you see, what to change, and why.
Hold up to ten comments per round with send unchecked.
Check my comments and make the changes. Touch only the commented sections.
Log comments per round. The count drops every round, or the brief needs work.
Round two held fewer comments than round one, and no rewrite touched an unmarked section.
Split the job when the work types differ.
The project demo handed one coordinator four different jobs: mine past analytics, scan competitor videos from the last month, write the outline, and build the deck. The coordinator opened separate threads, each with its own context. The deck thread sat idle until the outline landed, then published 18 slides.
Projects hold the goal line, the files, the instructions, and an auto memory every thread reads. The coordinator flagged only the outline gaps for a human answer.
Output, cadence, and one metric. Ongoing or fixed deadline.
Upload what a new hire would read for this job. Nothing else.
Research, analysis, writing, building. Different types become separate threads.
State dependencies in the prompt: the deck waits for the outline.
Open every resolved thread's output. Resolved means finished, not approved.
Every thread finished, and the dependent thread started only after its input landed.
Stay quiet until something changes.
A scheduled task without a notification rule turns into one more feed. The demo task stated a goal and a default: most days the honest answer is no video today. The phone showed one line: quiet day, nothing worth a video.
Notifications fire by push, email, or Slack only when a run finds something worth telling. Account memory holds the standing facts every chat, project, and scheduled task reads: format rules, names, time zone.
Before the schedule: notify me only when [metric] moves [amount] or [event] happens.
No change is a good outcome. One line, no report.
Open the task's advanced settings, switch on notifications, and sign in to the same account on your phone.
Click run now and confirm the notification lands.
Role, clients by segment, format rules, time zone, words to avoid. Client numbers stay in project files.
Most runs stayed silent, and every ping named a change worth acting on.
One month, three clients.
Nia, a first-year freelance marketing consultant. Three small-business clients: a bakery, a physical therapy clinic, and a landscaper. Every month she exports Instagram and Google Analytics CSVs, pastes them into Google Sheets, and screenshots charts into Slides. About 4 hours per client, 12 hours a month. Last quarter two reports went out with a number fixed in the sheet and stale in the deck.
One dashboard per client from the month's exports. Three totals checked against the CSV. Three recommendations, a chat tab, shared with owners who hold accounts.
A “Monthly Report” doc per client: goal, three must-include numbers, checklist with owners.
An eight-slide deck and a task table, both built from the brief. A changed number gets fixed in the brief, then rebuilt.
One comment round on each deck, ten-comment cap, send unchecked until the last mark.
Project “Client Reports.” Goal: three reports by the fifth business day, track client replies. Threads: analytics, competitor posts, report draft, deck.
Scheduled task on the first business day at 8:45 a.m. ET. Notify only when reach or site visits drop 20 percent or more month over month. Quiet otherwise.
Five standing facts: role, client segments, report format, Eastern Time, words to avoid. Client numbers stay in the project.
Planning estimate: about 4 hours per client down to about 75 minutes, 12 hours a month down to about 4. Sample first month: 9 totals checked, 1 mismatch fixed before sending, 14 comments in round one, 5 in round two, 1 alert fired. Planning figures. Log three months and replace every number.
Seven prompts.
Create an interactive dashboard from the attached [export]. Show what performed best, how [metric 1] and [metric 2] trended over [period], and what the top performers share. Let me filter by date range and sort by every metric. End with three specific recommendations, each tied to a number from the data. Use my brand file for styling. Do not add anything to my folder. List the three totals you used so I check them against the export.
Add a second tab to this dashboard called Chat. Let me ask questions about this data with an AI model, and show three suggested questions: one plan, one fix, one risk.
/doc Write a brief for [job]. Goal, audience, deadline, and these three facts: [facts]. Add a checklist with one task per line and an owner for each task.
/slides Turn the brief above into a deck I present live. One idea per slide, big text, ten slides at most, speaker notes on every slide.
Check my comments and make the changes. Touch only the commented sections. List each change in one line.
I need four things for [job], each a different kind of work: [1], [2], [3], [4]. Run each as its own thread at the same time. [4] waits for [3]. Tell me when all four finish and list anything you need from me.
Create a scheduled task for [cadence] at [time] ET. Goal: tell me whether [condition]. Most runs end with no change, and no change is a good outcome. Notify me only when the condition is met, in one line.
What the demo skips.
The creator says some accounts lack access. Screens and slash commands differ by account. Learn the order, not the buttons.
The dashboard chose its own sections. Check three totals against the raw export before you trust a recommendation.
He calls the workaround hacky. No formulas appear on screen. Keep the math in a real sheet.
Every question counts toward usage, and shared viewers ask questions too. Watch the meter for the first month.
Viewers need accounts, and client numbers travel with the link. Ask before sharing and strip personal details.
The task screen reads: requires this computer to be on. A sleeping laptop means no run and no ping.
Account memory feeds every chat, project, and task. Keep client facts in project files, not account memory.
A mid-roll ad pitches his paid course. Prove the free features on one report first.
Seven days. One report.
Export one raw file you pull every month. Build the readout and check three totals.
Add the chat tab. Ask three questions you would normally answer in a sheet.
Write the brief in a native doc with a checklist and owners.
Build the deck and the tracker from the brief. Nothing else.
Review the deck with batched comments, ten max, one round.
Set one scheduled task with a condition and a quiet default. Notifications on. Run now once.
Add five standing facts to memory. Count the files you opened this week and compare against last week.