You Don't Have a Productivity Problem. You Have a Setup Problem.
The full series
- The Prompt Is the Least Important Part
- Three Files That Replace a Chief of Staff
- You Don't Have a Productivity Problem. You Have a Setup Problem. (you are here)
- Stop Telling AI What to Do. Tell It What Done Looks Like.
- The _MANIFEST.md: Five Minutes That Save Hours
- Plugins Are Composable. Stack Them.
- The Monday Briefing That Runs Without You
- Custom Skills: Build Once, Run Forever
- AI as a Team, Not a Chatbot
- Cowork Is Not Chat. Stop Using It Like Chat.
One practitioner tracked every Cowork session across seven weeks and four hundred tasks, from sales prep and client reports to content drafts and meeting briefs. The sessions that produced client-ready work on the first pass all shared one trait. Not better prompts. Better setup.
The sessions that required three rounds of revision shared a different trait. They started cold, with no instructions loaded, no project context, and no connected tools. Just a blank window and a carefully engineered prompt trying to carry the full weight of context on its own.
The prompt alone can't carry that weight.
The layer most people skip
Article 1 in this series covered Global Instructions, the universal layer that loads before you type a single word. Your name, your role, behavioral rules for the agent. That layer matters, and it's where most people stop. But Global Instructions are generic by design. They apply to every session regardless of the project, the client, or the deliverable.
Cowork has a second instruction layer that almost nobody uses.
Folder Instructions.
When you open a Cowork session pointed at a specific project folder, Folder Instructions load on top of Global Instructions automatically. They carry the project name, the client's terminology, the deliverable format, the current deadlines, and any constraints specific to that engagement. Every session with that client starts warm, without a single word of context typed into the prompt.
Three layers. Not one.
Global Instructions set universal behavior for every session. Folder Instructions set project-specific context for that client. The prompt sets the specific task. Each layer is more specific than the last, and each loads automatically before the conversation begins.
Most users build layer one, skip layer two entirely, and try to compensate with layer three. They type everything the agent needs to know about the project into every prompt, repeating the client name, deliverable type, tone, format, and deadline from scratch every session.
That is the setup problem.
It's invisible because it feels like working.
What Folder Instructions change
A product manager running three concurrent projects creates three project folders. Each gets a Folder Instructions file with the product name, the target user, the sprint goals, the spec format the engineering team expects, and the key metrics for the quarter. When the PM opens Cowork in the payments folder, the agent already knows that the deliverable is a product requirements document, that the team uses Linear, and that the success metric is transaction failure rate below 0.3%.
No explanation needed in the prompt. The folder already told it everything.
A consultant serving four clients creates four client folders. Each Folder Instructions file captures the client's industry, their internal jargon, the report template they prefer, and the stakeholder names that belong in the executive summary. When you open a session in the Meridian Capital folder, Claude already knows that this client calls their quarterly review a "portfolio health check" and expects a two-page executive brief with risk flags.
We've seen content drafts that used to require three rounds of revision land on the first try after Folder Instructions were added.
Not because the AI got smarter between sessions. Because it started each session knowing things that we've watched users forget to mention, mention inconsistently, or describe differently every time.
Setup means
Setup means Global Instructions load identity and behavioral rules before the session begins.
Setup means Folder Instructions load the client's name, project goals, and terminology the moment you select a project folder.
Setup means your calendar, Slack, and Drive are already connected, so the agent pulls live data instead of working from stale information that was copy-pasted three days ago.
Setup means you type 10 words into a prompt and receive client-ready output, since three layers of context already did the rest.
Thirty minutes. Once. Then every session starts warm.
The tool connection that compounds
Context files and instructions tell the agent who to work for and what the project needs.
Connectors tell it what is happening right now.
Go to Settings> Connectors, then browse the directory. Connect Slack, and the agent scans your channels for follow-ups and surfaces decisions buried in threads. Connect Google Drive, and the agent reads your existing documents and matches their formatting without you having to describe it. Connect Google Calendar, and the agent knows your meetings for the week before you type a word.
We've seen practitioners connect a single tool and watch their morning planning drop from twenty minutes of manual scanning to a one-line prompt. The /productivity:start command handles the rest, checking your connected tools, organizing priorities, and building a task list for the day.
We've tracked this across roles and industries. The numbers land in the same range.
One connected tool makes your instruction layers work harder. Two connected tools make them compound, and the architecture rewards addition over replacement.
The cold start tax
Every Cowork session without Folder Instructions incurs a cold-start tax.
It doesn't appear on any dashboard. The first paragraph of every prompt gets spent re-explaining the client, the project, and the format. The agent spends its first pass absorbing what it should have known before the conversation started, and the output reflects that gap. Close but not tuned. Roughly right, but missing the details that make a deliverable ready to send.
Multiply that tax across five sessions a day and five days a week.
It shows up in thirty percent of every deliverable you rewrite before sending. It shows up in the second and third rounds of revision that wouldn't exist if the agent had started warm.
We've seen it manifest as a slow, creeping frustration that leads people to conclude the tool isn't good enough. We've investigated those cases, and the tool was never the problem. The context was always the fix.
The fix isn't a better prompt for next time. It's a five-line file in your project folder that loads automatically and corrects permanently. You never type it again.
Do This Today
- Pick your most active project or client.
- Create a folder for them inside your
projects/directory. - Create a Folder Instructions file inside it with five sections: client name, project goals, key terminology, deliverable format, and current deadlines.
- Connect one tool you use daily (Slack, Google Drive, or Notion) via Settings > Connectors.
- Open a Cowork session pointed at that project folder. Type
/productivity:startand see what comes back.
---
Next in the series: most people write prompts like a micromanager. Step one, step two, step three. The best delegation looks nothing like that. We cover the three-question framework that turns any task prompt into a clean handoff, and the feature that lets Claude stop and ask rather than guess.
Read Part 4: Stop Telling AI What to Do. Tell It What Done Looks Like. →