Glean’s next-generation prompting enables everyone to leverage the full potential of AI without complex prompt engineering
Today, Glean is introducing next-generation prompting with its suite of advanced features that enable users to create multi-step prompts using natural language to automate everyday workflows, as well as share and discover prompts in their company’s own prompt library. By allowing all employees to create both simple and more complex prompts with natural language, Glean is removing the necessity for coding skills, prompt engineering expertise, and complex traditional workflow automation tools, bringing the powerful capabilities of AI into the everyday tasks of everyone at work.
Glean’s next-gen prompting features include the following:
- Advanced prompts. Advanced prompts are new multi-step prompts that automate complex analysis and workflows by employing multi-step RAG for more intricate reasoning. Users can create advanced prompts with natural language, instructing Glean to conduct multiple retrieval and reasoning steps, and building on the output of each preceding step.
- Prompt builder. Prompt builder is a new studio creation tool for simplifying prompt engineering, allowing users to create, edit, test, iterate on, and save prompts. Users can easily link to and share prompts with their teammates, and teammates can continue building and iterating on their own versions of those prompts.
- Prompt library. Every organization will have its own prompt library, including suggested prompts from Glean, as well as the prompts that the organization’s own users have saved and shared in their organization’s prompt library. All employees can discover helpful prompts for their daily work, and teams, departments, and companies can curate the best prompts for everyone. Explore our ever-growing online prompt library to see prompts already in use today.
Read on for further details on each feature and how they work!
Prompt builder and advanced prompts
Glean’s prompt builder enables everyone to create both simple and advanced prompts using just natural language. For general search, research tasks, content drafting or document analysis, a simple prompt is the perfect solution. Simple prompts perform single retrieval actions, such as searching for documents and reading their snippets or summarizing the full contents of specified documents.
However, some tasks require multiple steps and retrievals to accomplish, something that a simple prompt can’t complete in one shot. This is where advanced prompting comes in.
For example, let’s say a customer support resolution process requires the following steps:
- Read the Zendesk ticket to understand the customer issue
- Search for related help center articles
- Search for recently closed tickets with similar issues
- Search for Slack messages where CSMs coached customers through the issue
- Generate a response with three suggestions on how to resolve the issue
This workflow usually would require multiple prompts sent to an AI assistant to resolve piecemeal, or multiple hand-coded, complex, deterministic logic steps in a workflow system. Advanced prompts enable users to instead isolate individual steps each with their own retrieval and reasoning, and then link them together to form a complex process. This both enables each step to be more accurate as it is focused on highly-targeted retrieval results, and each step to build on and utilize the results from all previous steps, so insight can grow throughout the process.
For this support workflow, the objective would be a high-level output that the user wants to achieve, written in natural language. In this case, it would be ‘Generate a response with three suggestions on how to resolve the issue’.
Glean then needs to be instructed with the appropriate steps, or separate actions, which load information into the memory of the Assistant, accessible by any subsequent step. Some examples of actions are:
- Glean Search. Search through all your company’s knowledge in Glean and load the most relevant results into the Assistant’s memory. You can issue search requests in natural language, or for further precision, you can choose to explicitly define search query templates. In this example, Glean Search could help find related help center articles and closed tickets with similar issues.
- Read documents. Read the full content of any document in Glean and load it into the Assistant’s memory. You can explicitly specify which documents to read or provide natural language instructions for the Assistant to retrieve documents found in previous steps. In this example, the read documents action can initially read through the Zendesk ticket to process and summarize the customer issue.
- Generate response. Generate response generates a response from all the information currently in memory using natural language instructions. It’s usually the final step in an advanced prompt sequence and always outputs natural language text to the user. In this example, the generate response action can take all the information loaded into the memory and generate a complete response with suggestions on resolving the issue at hand.
Advanced prompts enable users to be as customized and specific as they’d like—when creating prompts, users can create input fields by enclosing information in [[double brackets]]. For example:
- Find Gong calls with [[customer]] from the past month
- Find Slack conversations from last week involving our rollout of [[feature]]
- Find our go-to-market presentation for [[Q3]]
Bracketed fields will automatically be outlined and added to the Fields section at the top of the prompt builder. Users can then iterate on the prompt and test it with different values for the input fields to ensure that the prompt will execute consistently.
Prompt libraries encourage AI fluency
Another great part about building successful prompts? They’re all easily shareable and discoverable through links and the prompt library. Each organization has its own prompt library, where prompts can be readily distributed, experimented with, and discovered by everyone.
Users can take tried and true prompts already being used successfully by their teammates for their own workflows, removing the need for each individual user to manually engineer new prompts that make their work more efficient. Popular prompts are automatically suggested to users based on their role and what other similar users found useful. When people create new and useful prompts, the prompt library will drive awareness and adoption throughout the entire organization. Glean also provides suggested prompts for common workflows in the prompt library to help your teams get started right away.
AI that works for all
A key part of our vision for AI involves ensuring that it's useful for everyone. We’re beginning this journey today by empowering everyone from all parts of an organization—engineering, sales, support, and beyond—to harness natural language in a flexible, no-code system to automate their daily workflows and then share their learnings, so that everyone in the company can expand their horizons about what is possible with AI.
Here are more examples of how to use Glean’s prompting capabilities to deliver immediate value to daily workflows:
As of today, all Glean customers have access to the prompt builder, prompt saving and sharing, as well as the prompt library. Advanced prompts will be rolling out in private beta to select customers soon.
Ready to accelerate your workflows, drive collaboration, and uplevel every employee in your organization through next-generation prompting? Learn more by exploring the types of workflows you can accomplish through Glean in our online prompt library, and sign up for Glean GO: Beyond the box to see demos of sales, engineering, and support employees putting these prompts to work. If you're looking for an enterprise-ready Work AI platform and would like to learn more about Glean, sign up for a personalized demo today!