From AI as a Tool to AI as a Teammate: A Four-Level Framework for Using AI
Janet Gehrmann
Since joining Dust, a number of people have asked me what it’s been like. Dust is AI native, and thus I’ve been asked a number of times:
What does it actually mean to use AI well?
Most people are already using AI. They use it to summarize documents, rewrite emails, generate images, brainstorm ideas, or answer questions. But there is often a gap between using AI and using it in a way that meaningfully changes how work gets done.
In my experience, many people are still using AI primarily as a form of information retrieval or content generation. They give an AI system a prompt, receive an answer, and then manually take it from there.
That is a useful starting point. But it is only the first step.
I think of AI usage as progressing through four levels:
- Generation and transformation
- Connected information retrieval
- Bidirectional action
- Automated and scheduled workflows
The framework offers a way to understand AI’s capabilities and where there is room to go further, without assuming every task needs to reach level four.
Step one: Use AI to generate or transform something
The most basic way to use AI is as a tool for creating or editing content.
You might ask an AI system to:
- Draft a blog post
- Rewrite an email
- Edit a social media post
- Summarize a document
- Create an image
- Translate text
- Brainstorm ideas
- Turn rough notes into a presentation outline
At this level, the AI is responding to the information you provide directly in the conversation. You give it an input, and it produces an output.
For example:
“Rewrite this customer email to sound warmer and more concise.”
Or:
“Create an illustration of a robot organizing a messy office.”
This is already valuable. AI can help people move faster, get past blank-page syndrome, and improve the quality of everyday work.
But the AI only knows what you give it. It does not have access to the rest of the context surrounding the task.
Step two: Connect AI to another source of information
The next step is to let AI retrieve the information it needs from connected sources, within the access and permissions you’ve given it.
Instead of pasting an entire document, you could ask AI to search your company’s knowledge base. Instead of manually compiling customer information, you could ask it to look in your CRM. Instead of asking for a generic competitive analysis, you could connect it to your company’s research, strategy documents, and past sales calls.
This changes the role of AI. It is no longer just generating an answer based on a prompt. It is finding and synthesizing relevant context.
For example:
- “What did we promise this customer during their last three calls?”
- “Find all of our documentation about onboarding enterprise customers and summarize the recommended process.”
- “Which open opportunities are showing signs of expansion potential based on recent customer conversations?”
At this level, AI becomes much more useful because it can work with the information that already exists inside an organization. It can help people navigate the growing volume of documents, messages, databases, and meeting notes that would otherwise be difficult to search manually.
Still, the interaction is mostly one-directional. AI reads information and gives you an answer. You then decide what to do with that answer.
Step three: Let AI take action across systems
The third step is when AI becomes bidirectional.
AI does not only retrieve information from a system. It can also push something back into that system.
For example, an AI agent might:
- Read a sales call transcript and update the CRM
- Review a support conversation and tag it by topic
- Analyze a contract and create a task for the legal team
- Read meeting notes and draft follow-up emails
- Identify an issue in a customer conversation and create a ticket
- Review a document and add comments or suggested edits
The difference is important. AI is no longer simply answering a question about work. It is participating in the workflow itself.
Consider the difference between these two prompts:
“Summarize this customer call.”
And:
“Summarize this customer call, identify the next steps, update the opportunity record, and draft a follow-up email for my review.”
The second request connects multiple actions. The AI is still operating within a defined workflow, and a person may still review the result, but the output is no longer just text in a chat window. It is an update to the systems where work actually happens.
This is often where organizations begin to see a much larger return from AI. The value is not just in producing better words. It is in reducing the number of manual handoffs between tools and people.
Step four: Automate and schedule the process
The fourth step is to turn the workflow into something that runs automatically.
Instead of asking AI to perform a task every time, you define when it should happen and what it should do.
For example:
- Every morning, review new customer conversations and flag urgent issues
- Every Friday, summarize the team’s progress and identify blocked projects
- When a new sales opportunity is created, research the account and prepare a briefing
- When a support ticket is marked as resolved, extract any recurring product feedback
- Every month, analyze usage patterns and identify accounts that may need attention
At this stage, AI is no longer just a tool that someone actively opens. It becomes part of the operating system of the organization. When you share these AI workflows with your team, AI becomes multiplayer: colleagues can build on what already works rather than each starting from scratch.
The workflow might be triggered by a schedule, a new document, an incoming message, a database change, or an event in another application. AI can gather context, make a judgment, produce an output, and take action, according to the rules and permissions that have been defined.
This is where the idea of an AI teammate starts to become more concrete. The system is not waiting for someone to formulate the perfect prompt. It is carrying out a recurring responsibility.
A practical way to think about your own AI usage
When people ask how they should start using AI, I do not think the answer is simply “use it more.” A better question is:
What level is this task currently operating at, and is there a natural next step?
If you are using AI to rewrite an email, could it also retrieve the relevant customer context?
If it can retrieve that context, could it draft the follow-up and update the CRM?
If that workflow happens repeatedly, could it be triggered automatically?
Not every task should be fully automated. Sometimes the best use of AI is a one-off brainstorm or a quick edit. Human judgment remains essential, especially for sensitive, ambiguous, or high-stakes work.
But many people stop at the first level without realizing that the other three are possible.
The most important shift is to stop thinking of AI only as a place to ask questions. AI can also be connected to the information we rely on, the systems where we work, and the recurring processes that keep organizations moving.
That progression, from generation to retrieval to action to automation, provides a useful schema for learning how to use AI more effectively.
The question is not only:
“What can AI help me create?”
It is also:
“What information can AI access, what systems can it update, and what work can it carry forward without me having to start from scratch every time?”
That is where AI begins to move from a helpful tool to a meaningful part of how work gets done.
If you’re thinking about how to take AI further in your organization, get in touch with Dust →