iTranslated by AI
How Adding an "Ask AI" Feature to Task Management Significantly Improved the Developer Experience
I am developing a task management service as a personal project.
I added a feature to "let the AI handle this task" in that service, and it greatly improved the experience, so let me introduce it.
What I'm building
I am building a web service called tone.
The concept is "Team task management for humans and AI." In addition to a user-friendly interface for people, it is a task management service characterized by being accessible to AI via MCP.
This time, I added a feature to request tasks directly from the AI within tone.
Requesting tasks from AI
Here is the flow for requesting tasks from the AI.
First, you create a task normally. Write what you want to implement in the title and description.
First, create a task
Next, assign an "AI user" to this task.
AI users can be assigned just like human users, and they are configured with how they collaborate with the AI behind the scenes (described later).
Assigning an AI user to a task
Once an AI user is assigned, the "Run Task" button becomes clickable.
Click it to request the AI to execute the task.
The "Run Task" button is displayed when an AI user is assigned
After starting the execution and waiting for a while, the task status changed to "IN REVIEW".
This means it's now time for a human to review it, so let's take a look.
Tasks waiting for review are indicated with the IN REVIEW status
When checking the task details, the execution details are displayed as a conversation history, and if there were changes to the implementation, a Pull Request is created.
Once execution is complete, the history and a link button to the PR are displayed
Review the content, and if there are no problems, merge it and set the task to "Done."
And that completes the series of steps!
What's Great About This Experience?
1. Status can be visualized at a glance
When requesting work from multiple AI agents in parallel, it's easy to find yourself thinking, "Wait, what did I ask for just now...?"
By keeping everything in the form of tasks, the status is clear—"DOING" for work in progress and "IN REVIEW" for items waiting for confirmation—making it convenient to check the overall status at a glance.
Task status at a glance
2. Request tasks the moment you think of them, whether you're out or in bed
Have you ever come up with an idea while you're out or just before going to sleep?
Previously, I would make a note of it so I wouldn't forget, but now I create a task in tone and have the AI execute it right then and there.
By the time I open my computer the next morning, the work is already complete, and I can start my day by reviewing it.
tone is designed to be easy to use from a smartphone as well.
Being able to quickly give instructions even when you're not in a position to open your computer is an excellent experience.
tone is easy to use from a smartphone
3. Smooth flow from task identification to AI execution
For example, when developing a new feature, you might write down bugs and points for improvement as tasks while interacting with the product.
After listing them all, humans used to prioritize them and work on them in order of importance.
By turning them into tasks in tone, you can simply assign AI users from the top and request implementation, and all the work will proceed in parallel.
In reality, there may be cases where humans should handle parts involving core logic.
By leaving the simple tasks to the AI, we can focus on the essential parts that humans should be doing.
Accumulating a backlog in tone makes it easy to pass tasks to the AI
4. Other points
<Smooth Japanese input>
Global services often have issues with Japanese input (like submitting when you're just trying to confirm a kanji conversion).
tone is built to be used in a Japanese environment without stress, making it comfortable to use.
<Support for follow-ups>
AI agents are quite smart, but they occasionally head in a different direction than expected.
In such cases, try giving the task a bit more information or providing additional instructions as a follow-up.
tone allows you to follow up directly from the task.
If you don't get the desired result, you can supplement it with a follow-up
<Visualizing tasks for other members>
When you request a task from an AI, you naturally start writing a certain amount of information in the task title and description.
By writing things out like this, you can indirectly communicate the work status to other members, making collaboration easier. It's also convenient for looking back at it yourself later.
How to Set Up AI Users
Setting up an AI user is easy.
Create an AI user and configure the Cursor API key and GitHub repository information.
If you have used Cursor before, you should be able to set it up in about 3 minutes.
Please check this article for details.
Accompanying Tasks and Delegated Tasks
In an episode of fukabori.fm featuring t-wada, it was mentioned that Agentic Coding can be divided into "Accompanying" and "Delegated" types.
Agentic Coding can be divided into "Accompanying" and "Delegated" types
The "Accompanying Type" is a style where a human sits in front of the computer and proceeds with implementation steadily while giving instructions.
This method is suitable for core designs and complex features.
The "Delegation Type" does not require the human to be present; you entrust the task entirely to the AI, and the human steps in at the timing of the review.
This style is a very good fit for the feature introduced this time.
For content that you want to proceed on its own, turn it into a task and feel free to pass it over to the AI.
Summary
I wrote about making it possible to call the AI directly from a task.
I use it quite a bit myself; over the weekend, I used my commute time on the train to build and release a web service.
The development of tone itself is also progressing using this feature, and it is becoming an indispensable tool.
It's quite a great feeling to have tasks progress while you're away. Please give it a try! :)
Discussion