AI Agent for Every Website

Source: Dev.to

Introduction

This will be the last blog for 2025, and I wish you a happy New Year in advance.

Sam Altman, the founder of ChatGPT and OpenAI, said this year that “the next revolution will be AI agents – AI agents building websites, writing content, posting and scheduling content, and doing more.”

I read this year’s 2025 YC funding statistics blog (link), which states that 359 of the 420 companies were AI‑based startups. Building AI agents is the next top skill to learn in software development.

This certainly opens up the way to grab your next job or to build your next idea.

How I Learn New Technology

I often get asked how I learn new technology in programming, and I usually say:

• YouTube – my favourite channel for beginners is Hitesh Code Lab
• Online courses – Udemy (or any other platform) is my go‑to source
• Reading – I consistently read a lot of online articles, blogs, and newsletters (like this one 😁)

AI agents are the future. We will all be using AI agents in the next few years – let me connect the dots:

Feel free to judge 😂

What Will Future Websites Look Like?

One thing is certain: as both consumers and creators, we will be dealing with plain English prompts and explaining our ideas in a single chat input. A few layers will be added—voice agents, image‑reading/perceiving LLMs—turning a computer into a Jarvis that can talk, understand, and perform actions.

A quick imagination exercise suggests that almost every future website will have some AI‑chat component. For example:

• SaaS websites – AI chatbots to answer queries and agents to handle manual work.
• Documentation sites – AI chat to answer doubts instantly.
• Health / Finance / CRM / ERP / E‑commerce sites – LLM agents running in parallel to perform tasks on users’ behalf.
• Education platforms – Customised AI solutions for better learning (personalised quizzes, courses, deep research, content scraping).

Aggregators and social‑media platforms such as ThereIsAnAIForThat, Twitter, Reddit, and YouTube help me understand and predict the future.

“Online is the future. If you know how to use online channels and social media, the money follows with a lot of opportunities.”

AI tools that help content creators are earning massive MRR—not because they’re a trend, but because they work and are needed.

People connecting people—small communities to massive‑traffic sites like Twitter and Medium—generate revenue because of the human network.

Example: BoringCashCow

BoringCashCow is a simple aggregator site with ~50 k visitors per month. It illustrates how small online businesses can generate massive revenue/profits. There are countless YouTube videos about how these “boring” sites earn millions; the secret is that they solve at least one real problem for people.

AI‑Powered React CRM Template

One of our clients for the React CRM template asked, “Why don’t we have a simple AI chat input that takes my prompts and makes changes in the existing template?”

That’s why I added v0.dev and lovable.dev links to the React CRM template, enabling users to purchase and customise it with an AI website builder.

An amazing way to help businesses is to provide a solid base and then let the client/user make changes as needed.

Freedom Is Loved!!

Giving people freedom with tools always works for me—it builds trust. That’s why I believe platforms like gettemplate.website are equally important for the future.

GetTemplate provides online website templates for inspiration as well as a first footstep toward building an online business.

What GetTemplate Offers

• 80+ FREE website templates – copy‑paste the code or download the repository to build your own.
• 10+ production‑level backend & database‑integrated APIs (e.g., CRM, Docs, SaaS builder templates) to quickly start an online business.

I’ve worked with 50+ clients over the past five years, and most of them have similar requests: a portfolio site, a CRM, an ERP, API automations, forms, etc. This commonality inspired the launch of GetTemplate.

Get your source code → Build your idea, quickly and easily

Providing freedom builds trust, and that’s loved by almost everyone.

Building an AI Agent – How?

AI agents consist of three core components:

1. Memory – stores context and past interactions.
2. Task orchestration – decides which sub‑agents should act and in what order.
3. Monitoring – watches the execution and handles errors or retries.

Think of the parent agent as a manager that decides the overall process. The child agents are like junior salespeople or developers that perform specific tasks on behalf of the manager.

People often miss this concept: AI agents are a system, not just a team of independent bots. The system decides, performs, and monitors tasks.

Example:

• The main ChatGPT interface is a single‑agent that takes user questions and answers them.
• A secondary agent could be a web scraper that gathers information from the internet, feeding the results back to the main agent for synthesis.

By chaining agents together, you can build powerful, autonomous workflows that handle complex, multi‑step problems end‑to‑end.

Overview

The goal is to create a local version of Perplexity that can:

1. Retrieve the latest data (weather, finance, news, etc.) via web scraping.
2. Feed the scraped information to an LLM to generate accurate answers.

Architecture

1. Web‑Scraping API – Scrapes any data from a given URL.
2. Internet‑URL Scraping API – Retrieves data from public web pages.
3. LLM (AI Agent) –
   • Accepts the user’s question.
   • Calls the scraping APIs to gather relevant information.
   • Combines the scraped data with the original prompt.
   • Produces the final answer.

Workflow

flowchart TD
    A[User submits prompt] --> B[AI agent parses prompt]
    B --> C[AI agent formulates web query]
    C --> D[Web‑scraping API fetches data]
    D --> E[Scraped data + original prompt]
    E --> F[LLM generates answer]
    F --> G[Answer returned to user]

1. User adds a prompt →
2. AI agent understands the prompt →
3. AI agent creates a query for the internet →
4. Web‑scraping API fetches data using the query →
5. Feed the scraped data (plus the initial prompt) into the LLM →
6. LLM returns the final answer to the user.

Comparison

• This pipeline mirrors how services like Claude and Perplexity operate under the hood.
• The “agentic” behavior is essentially the same: the LLM decides what to fetch, retrieves it, and then answers.

Next Steps

That’s all for today. See you in the next session!

— Shrey