00Intro
The end of explaining everything from scratch
For a long time, using AI at work meant opening a chat and explaining the same context again and again:
- Who is the client?
- What are we selling?
- What should the output look like?
- What happens next?
Anthropic’s role-based plugins point to a different model. Instead of starting from a blank conversation, Claude inherits a predefined way of working: skills, commands, workflows and connections to the tools where the work already happens.
Each plugin is a set of plain text files — instructions, commands and connectors — published in an open repository. Not an app, not a new interface. A packaged way of doing a job.
The interesting part isn’t that there are eleven plugins. It’s that AI work is becoming structured.
01From prompts to operating procedures
A prompt finishes a task. A system repeats it.
A good prompt helps AI complete a task. A good system helps AI complete the same class of tasks again and again — with the same quality. These are two different things.
You can write this every time:
Research this company, find relevant information, prepare questions for a sales call and draft a follow-up.
Or you can define the process once — and let AI work inside it:
Prompt
Good for one result. Starts from zero every time.
System
Built once. Refined over time.
The more repeatable the work, the more valuable a structured AI workflow becomes.
02Grow · Revenue & customer acquisition
Eleven plugins, grouped by the work they structure
Eleven identical descriptions in a row don’t help anyone decide anything. So I grouped the plugins by the kind of work they take over: Grow, Build, Run and Organize.
01Sales
What it can structure
If your team cannot describe how a good sales call is prepared today, AI will not create that process for you.
02Marketing
Marketing suits AI workflows well: much of the work is repetitive, but it still needs judgment. Good candidates:
- campaign structure and briefs
- adapting content for each channel
- SEO checks
- messaging variants
- email sequences
- repurposing one piece into many
What I would not hand over to the system: final brand positioning and the choice of strategy. Those stay with people.
03Build · Product & knowledge
Turning messy ideas into structured decisions
03Product management
The plugin turns a problem into a spec, synthesises user research and updates the roadmap.
One of the most valuable uses of AI isn’t creating more ideas. It’s turning messy ideas into structured decisions.
04Data
AI can speed up exploratory work a lot — but it does not fix bad data.
05Bio research
A good example of how specialised AI workflows can become: choosing a scientific problem, quality control for single-cell RNA data, building Nextflow pipelines.
It isn’t “install the bio-research plugin”. It’s that the same architecture works for extremely specialised work — including yours.
04Run · Operations around the work
Structure the repetitive part, keep the judgment
These plugins don’t replace an accountant or a lawyer. The legal one is set up with US defaults, so outside the US its rules need to be adapted first.
05Organize · The layer most people underestimate
From managing tasks to keeping continuity
10Productivity
The plugin keeps a task list, updates it and remembers people, projects and your shorthand. That alone isn’t new. It gets interesting when AI stops managing individual tasks and starts keeping continuity between them:
Part II · Practical setup
How to use these plugins without wasting time
Installing a plugin is the easy part. The harder question is whether it actually improves the way you work. A useful plugin removes friction from an existing process — it doesn’t add another layer of tools to manage.
AChoose
Pick the task that already costs you time
Not the most impressive use case. Not the most complex one. The task you already repeat every week.
BMap
Define the process before adding AI
Before installing anything, write down four things:
A simple workflow might look like this:
That small step changes how you use AI. Instead of asking the model to “do something useful”, you give it a defined role inside a system.
CConnect
Connect only the tools that remove real manual work
Plugins become more valuable when they work with the systems where your information already lives — CRM, email, documents, analytics, project management. But more integrations don’t automatically mean a better workflow. Connect a tool only when it removes a real step.
Illustrative example
An integration adds more complexity than it saves.
Not worth itA real, repeating manual step.
Worth connectingAutomation should reduce friction, not move it somewhere else.
DGuard
Keep human approval where judgment matters
Not every step should be autonomous. Human review matters most when the output affects contracts, financial reporting, client communication, public brand content, sales outreach or strategic decisions.
Often safer and more useful
Risky
Generic plugins are a starting point. The real value appears when you add your own rules:
- use the company tone of voice
- don’t invent statistics
- give sources for research
- approval before publishing
- never send outreach automatically
- exclude existing customers
- personalise only with verified information
- no destructive queries
- flag missing data
- separate facts from assumptions
- identify the jurisdiction
- highlight uncertainty
- a review draft, not legal advice
This is how a generic AI capability becomes part of an actual business process.
ETest
Test the workflow on real work
Don’t judge a plugin by one impressive demo. Run it through 5–10 real tasks and score it:
One excellent result proves very little. A workflow becomes valuable when it works predictably.
FMeasure
Measure the workflow, not the output
Compare before and after:
If the output looks impressive but the process still takes the same time, you have a demo — not an automation.
Implementation checklist
Ten steps from idea to a working workflow
- Chooseone repetitive workflow
- Mapinput → process → output → next action
- Installthe plugin closest to that workflow
- Connectonly the services that remove manual steps
- Define qualitywhat a good result must contain
- Add guardrailswhat AI may and may not do
- Set approval pointswhere a human reviews the result
- Teston 5–10 real examples
- Measuretime, corrections and steps — before and after
- Improveonly then automate the next part
Don’t start by asking what AI can automate.
Start by asking which process is worth improving.
06Don’t automate the chaos
Four questions before you install anything
Before installing a plugin
- Is the process repeatable?If it happens once every six months, automation probably isn’t the priority.
- Can you describe what “good” looks like?AI needs an acceptance criterion.
- Does the workflow have reliable inputs?Garbage context produces polished garbage.
- Where does human approval belong?Not every step should be autonomous.
The best AI workflow doesn’t remove the human. It removes the unnecessary work around the human.
07What to automate first
The 3R test
Take any task and ask three questions. Three yeses make it a strong candidate for an AI workflow.
Weekly content repurposing
- repetitive
- describable
- time-consuming
→ strong automation candidate
Choosing company positioning
- low frequency
- high ambiguity
- time-consuming
→ keep it human-led
08The actual stack is bigger than the AI
A plugin alone doesn’t solve anything
For AI to do real work, six layers have to work together:
AI becomes operational when all six layers work together.
09Conclusion
Start with one workflow, not eleven plugins
The temptation with AI is always to add more: more tools, more agents, more plugins, more automation. The better question is:
Which repetitive process would create the most value if it simply worked better every week?
Then build the next layer.
Don’t automate more.
Build better systems.
Sources: Anthropic, knowledge-work-plugins (open repository, Apache-2.0); Claude Help Center, “Use plugins in Claude”. Plugin names and features as published at the time of writing.