Home / AI Hub / Projects and Agents for Repeat Work
AI Hub

Build it once for the work you do over and over.

If you're re-explaining the same context to AI every single time, that's a setup problem, not a prompting problem. Most tools let you save a standing "project" or "agent," a place that already knows your team, your format, and your voice, so you're not rebuilding the prompt from scratch every week.

Set it up once with real detail. What comes out is only as good as what you put in at the start.

ELT status updates
Load in your team's goals, current projects, and last month's update as examples. Then each week you're feeding it bullet points, not writing the summary from a blank page.
Customer update emails
Give it two or three past emails you were happy with, plus the tone you want (direct, warm, no corporate padding). It'll hold that voice far more consistently than you rewriting it fresh each time.
Reviewing reports
Tell it what you're actually checking for, numbers that don't reconcile, missing context, claims without evidence, and have it flag those first. You still read the whole thing, it just tells you where to look hardest.
Setting This Up, Step by Step

Both Claude and ChatGPT let you build a standing workspace that holds context between conversations. The setup is a few minutes once, and it pays off every time after.

In Claude
  1. Click "Projects" in the left sidebar, then "New project."
  2. Name it for the specific recurring task, "ELT Status Update" or "Customer Emails," not just "Work."
  3. Open Project Knowledge and upload or paste your reference material: last month's update, your team's goals doc, two or three past emails you were happy with.
  4. Go to the project's custom instructions and add the tone you want, the format, and anything it should never do (no corporate padding, no exclamation points, whatever's true for you).
  5. From now on, every new chat you start inside that project automatically has all of it loaded, so you're not re-pasting context each time.
In ChatGPT
  1. Click "Projects" in the sidebar and create one, named for the specific recurring task.
  2. Add files to it: past examples, your team's goals, style notes.
  3. Open the project's instructions field and describe the tone and format you want.
  4. If you want something reusable across your whole team instead of just you, build a Custom GPT instead: go to "Explore GPTs," then "Create," and configure it with the same instructions and reference files.
  5. A Custom GPT can be shared with a link; a Project stays private to you.
Load the Reference Material With Real Detail

Don't just say "write like me." Paste in two or three actual examples of the output done well, plus one you'd consider a miss and why. Specific beats descriptive every time, the model learns your voice from examples far better than from an adjective.

Use It, Then Correct It Back Into the Project

The first few outputs won't be perfect. When you edit one, save the corrected version back into the project's files or notes: "here's the version I actually sent, use this as the new example." The project gets better each time you do this, instead of you making the same correction over and over in every new chat.

Setting one up

Give it your team's context once: who's on it, what you're working on, the format you want things in, and two or three examples of the output done well. From then on, each new request only needs what's different this time.

The point isn't to remove yourself from the work. It's to stop spending the first ten minutes of every request re-explaining things it should already know.

AI Hub · Go Deeper
AI Agents, and How to Build Your First One →
AI Hub
Using AI as a Thinking Partner →
AI Hub
How to Think About AI →