Here are a few ideas for useful thunks you could create and use at work. You don't have to do something complicated for AI to be useful. In fact, quite the opposite. AI is often most useful in simple and commonplace work scenarios.
Each of these is available as a copyable sample in the product.
Each example is an instance of a category of thunks that corresponds to how people are used to working. If you were asked how you organize the tasks of a project, you might have one of the following responses:
Each task is an entry in a task list -- we show you an example of a thunk for ad-hoc task lists
Each task is represented by a file, with a collection of these files in a project folder -- we show you an example of a thunk used to extract information from expense receipts.
Each task is represented by a data row, with many such rows in a spreadsheet -- we should you an example of a thunk used to extract information about a spreadsheet listing clinical trials.
Each task is an entry in a workflow or project system like Trello or Asana -- we show multiple examples of thunk scenarios in this category, including managing customer support tickets, managing sales leads, and hiring an employee.
Thunk.AI supports all of these conceptual models of a project. In all cases, the project is represented as a "thunk". Whatever your preferred conceptual model, a thunk always needs to store the tasks and their data. A thunk needs to allow multiple users to access it and collaborate. Within a thunk, AI agents take on the work of creating and coordinating tasks automatically, and actually doing useful work on the tasks on behalf of the users.
Manage an ad-hoc task list
As the name suggests, this kind of thunk doesn't really have much upfront planning. Each task is heterogenous and different. You add tasks to it and your AI agent starts working on them. It's as simple as that.
Manage team expenses
There are many cases where information has to be extracted from files (documents or images) and then decisions have to be made based on that information. A classic example is when employees submit expense receipts. This example shows how image files representing expense receipts can be easily read.
Perhaps more interestingly, the subsequent step of "approval" of the receipts is also a good opportunity to showcase some things AI does particularly well. No two receipts are identical and the approval rules often require interpretation.
If you are experimenting with a copy of this sample, consider providing a receipt in a different language (AI models can translate). Consider applying a policy that says alcohol cannot be expensed, and then see if the AI model recognizes if specific items on a restaurant bill are alcoholic.
Watch the video below
Extract information about clinical trials
There is a broad class of work projects that involve data gathering and synthesis for a collection of entries. These projects often fit the pattern of a spreadsheet of tasks, with the new information gathered being added in extra spreadsheet columns.
In this particular example, we consider a thunk used by a medical research lab to collect deeper information about clinical trials. They already have an initial spreadsheet of trials with a URL for each trial. They then want to automatically collect more specific information by reading the details of the trial from the URL, and then doing some broader web research about the drugs being used in each trial. This example shows the Thunk.AI feature called "AI formulas" which are like spreadsheet formulas described in English.
Here's the link to this sample
Or watch the video below
Process intake forms
This sample thunk shows how to use AI to automate a very common pattern of work that spans and integrates multiple systems. Input comes from an external system. It is processed by AI agents in the thunk. And then as a final step, the AI agent communicates with an external service to share the result of processing.
The specific example use an external Typeform for data submission. The AI agent logic validates the submitted data, reads information from submitted URLs, checks the information against some requirements, makes a decision, and then communicates that decision via a Google Chat message.
The Thunk.AI platform is extensible. Each thunk can be configured to accept inputs from other systems (eg: Typeform). And it can also be configured to add new AI "tools" --- these are capabilities that you can give your AI agents to do things that are specific to your environment (like sending Google Chat messages).
This is a great example to show that intelligent thunks can participate in and add value in an end-to-end process that involves other systems or other human processes.
Here's the link to this sample
Or watch the video below
Hire an employee
Here's an example of a process that takes work items through a "funnel". This is very common in sales (a "sales leads funnel"), marketing (a "marketing leads funnel"), operations (a "support tickets funnel"), and hiring (a "candidate funnel").
Thunks support plan workflows that repeat a pipeline of tasks for each item in a collection. In this sample, it is a collection of candidates for an open job position. While the particulars of the thunk you use will be specific to your business, this general pattern is a good one to understand and apply.
Here's the link to this sample
Or watch the video below
Manage a funnel of sales leads
This is another classic "funnel" workflow used in every organization that has a sales team. This sample shows you how Thunk.AI lets you pair AI agents with every member of your sales team while your "owner" AI agent manages the entire sales process.
Process customer support tickets
This is another classic process used in every organization that has a customer support team. This sample shows you how Thunk.AI lets you pair AI agents with every member of your support team while your "owner" AI agent manages the entire process. Variants of this thunk can also be applied to bug tracking systems for product teams, help request systems for IT, and any other "track work by assigning tickets" systems.
Here's the link to this sample
Or watch the video below