Article
How to get the most out of advanced AI tools
Building collaborative AI-assisted workflows with ChatGPT Desktop or Claude Desktop
Who is this article for?
Employees and managers who mainly use ChatGPT or Claude for conversations, questions and answers.
What is the purpose of this article?
To offer a way of thinking that helps you get more from advanced AI tools: incorporating them into workflows where people and AI work together, with shared information, tools and a defined way of working.
From individual conversations to a shared workflow: a change in perspective
Many of us already use AI at work. We ask it to draft an email, upload a document for summarising, discuss an idea or get help analysing data.
A chat can already be collaborative: we ask, the system suggests, we make corrections and together we develop a result. The shift comes when we create a defined AI-assisted workflow for a specific task. As part of that workflow, we extend the work to external systems and the files and folders relevant to completing the task.
This approach brings together three things:
A defined process with clear tasks
We know what we want to achieve, the stages of work, who handles each task and where the employee and AI pause to think together, review or make a decision.
Connections to external tools and systems
With appropriate connections and permissions, the system can gather information and take actions as part of the workflow. This can speed up the work and reduce copying, data entry and manual switching between systems.
Shared files on your computer or in the cloud
The employee and AI work on the same documents, spreadsheets and outputs in an environment both can access. AI can create and update a file, the employee can review and edit it, and the work continues from there.
The change in perspective is to see AI as a partner in carrying out a process within our working environment. Conversation supports planning, direction and joint thinking; tools make actions possible; and shared files hold the outputs. Connecting these elements lets us complete a whole piece of work across multiple tasks and conversations, following a method we've defined together.
What is an AI-assisted workflow?
A workflow is a sequence of actions that leads to a result. Examples include preparing a quote, creating a product listing on a website, analysing sales or producing a report.
In an AI-assisted workflow, some actions are carried out by the employee, some by the system and some through collaboration. That division can change as the work progresses.
AI might gather information and prepare a draft. The employee reviews it, adds knowledge that wasn't in the documents and chooses a direction. The system updates the output, runs checks and prepares the next stage.
This is what symbiotic working means here: the person and AI complement each other within a defined way of working.
The employee brings knowledge of the business, experience, preferences and responsibility for decisions. AI contributes the ability to process information, write, compare and take actions through its available tools. Their conversation is part of the work. Sometimes a question the system raises, or a correction the employee adds, is what makes a better result possible.
To work this way over time, the system needs three things: access to information and tools, instructions for how to work and organised context for the task.
System connections: access to information and actions
In everyday work, our information is spread across folders, email, spreadsheets, a customer management system and the business website.
Connections let AI use this information and, where appropriate, take actions. For example, it could read product details, find a document or create a draft in the website's content management system.
As you explore these possibilities, you'll encounter a few terms:
Connectors
These provide access to a particular service or system, depending on the connector's capabilities and permissions.
MCP
A standard through which AI systems can connect to information sources and tools. You don't need to understand the technical details to understand the benefit.
Plugins
Packages that can include connections, working instructions and additional capabilities.
The names, availability and setup vary between products. What you need to know is how to connect to the system you use. In the workflow design conversation described later, you can start by asking AI to guide you through connecting to a system. Some cases will also require technical help or permission from a system administrator.
Note: Giving AI access to your systems involves information security considerations you need to understand.
Skills: preserving your way of working
Suppose you've explained to AI how to write a product description: what to include, the style to use, which details to check and what to do when information is missing. After a few attempts, you reach a result that works for you. Now you want to use the same method for the next product.
A Skill is a reusable set of instructions and resources that defines how to perform a particular task. It can include stages, writing rules, examples, templates and checks. In some cases, it also includes tools or code that help carry out the work. Read about Skills in Claude's documentation
You can create a Skill for an entire workflow or divide the work between several Skills. For example, one could define how to prepare a product listing and another how to transfer the information to the website. A Skill can also define collaboration with the employee: when to show a draft, which questions to ask and when to stop for approval.
Saving instructions doesn't guarantee perfect execution every time, but it provides a consistent basis for working and checking results. When you discover an improvement to the method, you can update the instructions and apply it to future tasks.
Projects: bringing context and materials together
A project provides a framework for gathering the work, instruction files and relevant knowledge in one place. It lets you create a workspace for a defined process, including workflow instructions (Skill files), background materials and related conversations.
A workspace can draw on files uploaded to the project, a folder on your computer or information from connected systems. Connections and Skills may also be configured outside the project itself, but I recommend keeping them together.
How do you design the workflow through conversations with AI?
Once we've chosen a process for an AI-assisted workflow, we create a central folder for it and a project connected to that folder. The first conversation in the project is the workflow design conversation. Here, we explain what we want to achieve, how the work is done today, what materials are available and where the employee's judgment is needed.
Together, we work out the stages, decide what the system will do independently and which tasks will be collaborative. We define when the system should stop for the employee's review and approval. If we need connections to external systems that AI can't yet access, we can use this conversation to ask for help and setup guidance. We then turn these decisions into the appropriate Skills. They preserve the agreed method so subsequent conversations can be used to carry it out.
I recommend pinning the design conversation at the top of the project, because you may want to return to it later to update the workflow.
How do you improve the workflow over time?
Even a well-designed workflow will encounter new situations. You may discover a missing step for checking dimensions, find that the description style needs changing or notice that the system asks for approval too often.
It helps to distinguish between a correction for a particular case and a change to the working method. If one product needs an unusual description, you can handle that in its conversation. If all product descriptions need additional information, update the Skill.
You can return to the saved workflow design conversation, explain what you've learned and ask for the instructions to be updated. Make sure the change is saved in the Skill itself and that it's clear which version is now in use. This gradually turns experience into a better working method. The employee continues to contribute professional understanding, and the system can apply it to future work.
How does it all work in practice? An AI-assisted workflow example
Let's look at a workspace for a business adding products to a Shopify store. This illustrates one possible structure; the actual connection depends on the tools and permissions configured.
The project will have two main Skills:
A Skill for the overall workflow
This defines how to move from source materials to a finished product listing: reviewing materials, clarifying missing details, writing together, preparing images and checking the result with the employee.
A Skill for working with Shopify and uploading products
This defines how to use the store connection, transfer product details into the right fields, upload images in the correct order and check the result. The Skill guides use of the connection; access to the store also requires an active connection and appropriate permission.
The local working folder on your computer would look like this, with the project connected to it:
- store-products/
- business-context.md
- skills/
- product-workflow/
- SKILL.md
- shopify-upload/
- SKILL.md
- product-workflow/
- products/
- ceramic-vase/
- supplier-details.pdf
- product.md
- images-source/
- original-01.jpg
- original-02.jpg
- images-final/
- ceramic-vase-01.jpg
- ceramic-vase-02.jpg
- ceramic-vase-03.jpg
- ceramic-vase/
This is an example structure for an environment that works with folders and files. Adapt the location of Skills and access to materials to the tool you're using.
A new folder for each product
For every new product, create a folder inside the products folder. Add the initial materials: supplier details, specifications, notes and source images. In this example, the original images are stored in images-source, and the finished images go in images-final.
The employee opens a conversation about the product and points AI to the folder. The system follows the workflow Skill: reviewing the materials, identifying what's missing and starting the collaborative work.
A product file developed through conversation
Together, you create product.md. This is a structured text file with headings and sections that brings together the information needed for upload:
- Product name, short description and full description.
- Relevant search terms.
- Technical details, materials and dimensions.
- Prices.
- Variants or options, such as colours and sizes.
- References to the final images and their display order.
AI can prepare a draft from the materials, while the employee adds details, refines the wording and chooses how to present the product. Missing details remain flagged for clarification until the necessary information is available.
The file is updated as the work progresses until it reflects the approved content.
Preparing images and their display order
With suitable image tools, you can work on the images and save the approved results in images-final. Each filename includes the product name and a number indicating the desired display order. For example:
ceramic-vase-01.jpgMain imageceramic-vase-02.jpgAnother angleceramic-vase-03.jpgClose-up
The upload Skill specifies using this numbering to set the image order on the website. The filenames express the intended order; the upload stage needs to apply and verify it.
Transferring to the store and checking the product
Once the content and images are approved, the upload stage can begin. AI uses the Shopify Skill and available connection to transfer the information into the system. You can define the workflow so the product is first created as a draft. The employee reviews the product page, prices, variants and image order, then approves publication after checking.
Throughout the work, AI reads materials, prepares content and uses tools. The employee contributes knowledge, chooses, corrects and approves. The Skills connect these actions into a repeatable method.
When the next product arrives, you open a new folder and conversation for it and use the workflow you've already developed. If you identify a need for a lasting change, return to the design conversation and update the relevant Skill.
I hope you found this article useful.
Best of luck,
Yaron Shaool