Hemant Bhatt

June 8th, 2025

Building agentic projects that think for themselves

Let's dive into some practical agent ideas. Once you've got your creative wheels turning, you can head over to my other, more technical blog that talks about the technical details of agents. Vercel AI SDK supports Next.js, Vue.js, Node.js, and plenty of other frameworks. The age of rigid, brittle automation is ending. Welcome to agents that actually know how to improvise.

Email Assistant Agent

Drowning in emails? Got more messages than minutes in your day? Or maybe you just want to free up some mental space while your inbox sorts itself out. Time to build an email assistant. This digital helper will sort your emails, track ongoing issues, escalate problems when needed, and draft replies for your approval.

Tool list:

1. GetEmailsForToday()
   Description: This function will let the assistant get the latest emails that the user received in the last 24 hours.
2. GenerateIssue()
   Description: This can be a simple text file that the agent can maintain to keep track of the ongoing issue, or it can be an issue management software like GitHub that the agent can use to generate an issue based on the conversations for the day.
3. UpdateIssue()
   Description: This can be used to update existing issues in the system. An issue once generated can have updates based on the latest conversation.
4. GetOpenIssues()
   Description: This will be used to get the list of open issues which was generated by the agent previously.
5. DraftFollowUpEmail()
   Description: Based on the conversations, issues and their context, the agent should generate follow-up drafts with clients and vendors respectively.
6. DraftEscalation()
   Description: Based on the situations of the issues and their timelines, escalations should be drafted while keeping appropriate members in the loop.

How Will These Work?

Think of a sequence like this:

1. User executes the agent first thing in the morning - Because who doesn't love automation with their coffee?
2. The agent gets the latest emails from the last 24 hours - Time to see what chaos unfolded while you were sleeping.
3. The agent retrieves all open issues - Pulling up the digital equivalent of your mental sticky notes.
4. The agent compares emails with existing issues - Like a detective matching clues, it spots that Alice's delivery problem is still lurking around after 4 days.
5. The agent escalates stale issues - It drafts an escalation email to the delivery manager because Alice has been patient long enough.
6. The agent identifies new issues from fresh emails - Bob's software isn't working, and this looks like a brand new headache.
7. The agent generates a new issue for Bob's problem - Creates a proper ticket because scattered thoughts help nobody.
8. The agent drafts acknowledgment emails - Tells Bob "we got your message and we're on it" without you lifting a finger.
9. The agent queues follow-up emails for approval - Prepares gentle nudges for clients who've gone radio silent.

Documentation Builder Agent Assistant

Hate documentation? Just deployed a neat little feature but don't have the time to document it? Or maybe you want a rough draft that you can polish into perfection instead of staring at a blank page? Time for a docs builder agent. Because let's face it - writing documentation ranks somewhere between doing taxes and assembling furniture on the fun scale. But it's necessary, and this agent makes it painless.

Tool list:

1. GetLatestCommit()
   Description: This gives the agent the latest commit from the repository. Includes the old code, the new code, and the diff's text. Basically, it's like showing the agent exactly what you changed without having to explain it twice.
2. GetDocsList()
   Description: Returns the list of markdown files in the system. Think of it as the agent's table of contents for your documentation library.
3. UpdateDocFromFileName()
   Description: Accepts the updated documentation string and updates the markdown file. The agent's way of saying "here's your documentation, fresh and ready."
4. CreateDocumentationPR()
   Description: Based on the proposed documentation changes, create a new pull request in the documentation repository.

How Will These Work?

Think of a sequence like this:

1. A user commits a new set of code into the repository - This gives users quick access to frequently used parameters without the tedious journey to their profile page every single time.
2. The settings icon acts as a modal - Clean, accessible, and saves everyone from unnecessary clicking. Smart design.
3. After the commit is pushed to GitHub, the agent activates through GitHub Actions - Like a documentation fairy that appears whenever code changes.
4. The agent pulls the latest code changes and the documentation list - It's doing its homework, understanding what changed and what docs exist.
5. Based on the changes, the agent pulls the header documentation - Smart enough to know which docs need updating.
6. The agent generates the altered documentation for the header - It writes the first draft so you don't have to. The heavy lifting is done.
7. It creates a new pull request for the documentation updates - Organized, traceable, and ready for review. No chaos, just clean process.
8. The user reviews and accepts the PR - You get to be the editor instead of the writer. Much more pleasant.

The beauty is in the simplicity. Code changes trigger documentation updates automatically. No more "I'll document this later" promises that never get kept. The agent handles the grunt work, you handle the final polish.