How do you hand off work to AI Agents and your team in Sordit?

One Job, multiple Tasks

Imagine you’re heading into a meeting and want your inbox reviewed while you’re away. By the time you return, you’d like to know which conversations need your attention most and why.

In Sordit, you can create a one-off Job to handle this, with three Tasks.

Task 1: Review and summarise your inbox

The first Task is to read your recent email conversations and prepare a summary in Google Docs.

For this, we assign Agent L, powered by an everyday, cost-efficient AI model. Since the work involves processing a large number of emails, we don’t necessarily need our most powerful model.

Task 2: Identify the two most urgent conversations

Next, we want to know which conversations are genuinely important, which require immediate action, and why they deserve our attention.

This requires more judgement than simply summarising emails, so we assign Agent T, powered by a stronger reasoning model.

Agent T receives the previous Task’s work and uses it to evaluate the conversations and recommend the top two priorities.

Task 3: Review the recommendations

Finally, we assign this Task to you or another team member.

You review the two recommended priorities and decide whether the Agent’s judgement is correct. You can accept the recommendations or send them back for further consideration.

Each Task has its own working environment

When an AI Agent starts a Task, Sordit runs it in a dedicated cloud sandbox, providing the Task instructions, relevant context and authorised business tools.

The Agent can perform multiple actions within its Task. For example, Agent L reads Gmail conversations, retrieves messages, creates a Google Doc and verifies the document before completing its work.

When the next Task begins, Sordit passes along the relevant results and context from the previous work.

For Human Tasks, Sordit presents the instructions, previous results, supporting documents and any available decisions, so the person has the information needed to respond.

Decisions control what happens next

Not every Job follows a straight line.

For each Task, you can define whether the work should:

• Continue to the next Task.
• Go to another Task, returning to an earlier Task for revisions or jumping ahead to a later Task, like a decision tree.
• Complete the Job when the desired outcome has been achieved.

You can also define multiple decisions within a Task, each leading to a different next step.

In our inbox example, the final review offers two choices:

• OK: You’re satisfied with the priorities, so the Job is completed.
• Wrong judgement: You provide feedback, and Sordit sends the work back to Task 2 for Agent T to reconsider.

When work returns to an earlier Task, Sordit creates a new iteration, retaining the previous results and feedback. Once Agent T has revised its recommendations, the Job returns to you for another review.

Sordit keeps a record of the completed Tasks, decisions and revisions, so you can follow how the Job progressed from start to finish.