DEV Community: Gary Xia

Rethinking Coding Education for the AI Era

Gary Xia — Tue, 21 Apr 2026 14:29:09 +0000

Today marks the 8-month anniversary of my tryst with Windsurf, Replit, and Vercel v0. I thought it fitting to commemorate this ill-fated romance with some thoughts on AI's place in the coding education landscape, as well as a learning aid I've been ideating on for some time.

· · ·

Some Background

When I first started coding with freeCodeCamp and The Odin Project over a year ago, I was blissfully unaware of coding agents and how sorely deficient my skills were in comparison to theirs. And I say 'blissfully unaware' not as a turn of phrase, but as a literal nod to how much simpler life was when it was just me, an editor, and some lesson notes.

Because there's nothing more demoralizing than the realization that with one prompt, a fancy algorithm can spit out entire features that took you weeks building.

They say that comparison is the thief of joy, and in this case I didn't scoff.

· · ·

The Problem

The issue with AI in the coding education landscape today is that it's a step function - you either use it or you don't. There's virtually no middle-ground except say, system prompts or context injection in an attempt to 'nerf' the model.

If you choose to use it, then you tread a very fine line between learning and pseudo-learning. In my experience, the period in which I leaned heavily on coding agents, coincided with the period in which I learned the most but retained the least.

It's like not going to classes all semester, then flipping through the slides the night before the exam. You think you got it, but you really don't.

If you choose not to use it, then you come off as being 'resistant' or 'old-school', and you do miss out on some objective benefits - like having a personal Stack Overflow on steroids and without the snobbish attitude.

· · ·

A Tenuous Middle-Ground

The solution?

I think everyone's handling it differently, but personally, I've gone back to reading documentation, code snippets from framework or service providers, and light tutorials.

This revised approach has clearly been more fruitful. But trying to maintain this approach with existing AI tools has been genuinely frustrating. Tweaking system prompts or injecting instructions inline to tone down the solutions, just doesn't seem like a permanent fix.

But at the very least, it provides a cardinal direction.

In my mind at least, the solution is to bifurcate our treatment of coding professionals and those who are learning to code. After all, knowledge acquisition lights up a fundamentally different part of our brain than task execution, even if the underlying networks are the same.

· · ·

The Systemic Issue

What I realized is that most startups and major AI companies are chasing professional developers. Their target customer is the enterprise team, not people who are learning to code. This is understandable given the revenue dynamics and longer term AI game-plan.

But I'm betting on the fact that learning to code will still be important despite how the role of coders will change in the age of AI.

To that end, perhaps there's a gap that's being left unfilled.

Namely, "AI Constraint".

· · ·

The AI Constraint Layer

My understanding is that the "AI Constraint" layer has 3 main levels:

Level 1: the model itself, i.e. training data, weights, etc
Level 2: the harness, i.e. the intermediary steps a user prompt flows through before being returned as a fully-formed response
Level 3: system prompts and context injection

Currently, individuals are only able to modify the response at Level 3, via system prompts or one-off context injection.

But it's somewhat annoying to have to save an instruction somewhere and paste it into every LLM you interact with. Doubly so when there are dozens of models from tens of providers over many platforms that you interact with regularly. This is as true for assistants in code editors or CLI as it is for general-purpose LLMs in browser chat windows.

This does not feel native. Because that's not the problem these solutions were built to solve.

In addition, it's incredibly inefficient.

Manually "nerfing" a response for an advanced AI assistant that can spin up hundreds of lines of code so that it can answer a student's questions and provide only a few lines of rudimentary code, is likely using the wrong tool for the job.

· · ·

A Better Solution

What I would like to see is something designed from the outset to help people learn how to code. NOT to make professional programmers more productive.

Those objectives are similar only in that they both involve code.

Fundamentally different problems need fundamentally different solutions.

· · ·

Product Vision

Level 3 doesn't interest me. It exists, it kind of works, and it doesn't move the needle.

Level 1 and Level 2 on the other hand, are positively titillating.

For Level 1, I'd like to see fine-tuned models that are primed to be teachers instead of practitioners. They don't need system prompts or context injections that warn them not to give the student the answer. Those models can be standalone - i.e. listed for free access on the AWS BedRock Marketplace - or incorporated into an IDE as a model provider.

Now to Level 2. I don't see the core IDE infrastructure changing too much. The way that a user prompt in a chat window gets delivered to an agent or LLM, then passes through the system before getting returned, remains largely the same. The ability for LLMs to be aware of a user's project files, and be able to connect to MCP servers also remain unchanged.

The critical departure from current IDEs however, is the "Interface/UI" aspect. That's also the critical piece to consider when deciding between Monaco, Code-OSS, or Theia-IDE as the base infrastructure.

· · ·

Features

What I'm envisioning is that in addition to a sidebar to find files, a main editor window, and a side-panel AI Assistant, there will be an "always-on" AI agent that constantly monitors the user's cursor position, inputs, and text highlighting.

The objective is to create a hyper-aware AI harness that predicts the user's intentions and knowledge gaps.

For example, if a user creates a new file app/api/query/route.ts, then the action of clicking into the blank document and pausing for 5-10 seconds, should trigger an inline popup with 4-5 choices for the user to select.

These choices could look like "Are you stuck on…", "Would you like to know more about…", "Is X what you are trying to accomplish", etc. Basically a 'Hover Tooltip', but users can click on these suggestions to trigger an automatic prompt returning an answer in the AI assistant side-panel.

I also want to nerf inline suggestions so that instead of spitting out the entire structure of a function, a user-customizable delay will trigger only the next keyword or symbol needed to complete a piece of code, and not the entire line or function.

I think there will be many more features and micro-features, but I think that's enough to get started with. To enable even these simple feature additions to traditional IDE's, I think will require significant refactoring of any open-source IDE codebase.

· · ·

Remaining Decisions

What remains undecided is how this product will be delivered.

It could be an extension in the VSCode Marketplace. It could power a learn to code website that's more dynamic in lesson generation and being hands-on than any code learning platform out there. It could be a standalone IDE that users download and use in place of VSCode. It could also be all 3.

I think it will take some time to untangle this and finalize the feature set and figure out which delivery formats to prioritize.

A good place to start might be sketching venn diagrams of the product variations, finding the intersection of these sample spaces, then beginning work on only the features and codebase needed for that intersection.

· · ·

Concluding Thoughts

This has been listed as "Project X" in my notes for a while now.

But I think I'll finally give my bastard son a name.

{ Raisin.IDE }

Why? Because 'raisin' sounds like the French for 'reason', and in the age of limitless knowledge, the only thing left to teach is reason.

Cheesy? Absolutely, and that's the way I like it.

✌️

Originally published on Stackademic. Cover photo by cottonbro studio from Pexels.

CommunityPulse: Automating Lead Gen for AI Agencies with n8n Agents and Bright Data

Gary Xia — Mon, 01 Sep 2025 07:00:08 +0000

This is a submission for the AI Agents Challenge powered by n8n and Bright Data

What I Built

A tool to build Qualified Leads funnels at zero cost. AI automation agencies get alerted when businesses get bad reviews, so they can be the first to call and offer a free damage control SaaS product (also built in n8n using the "Google Business Profile" verified node), which serves to establish rapport so they can offer bespoke services down the line.

The tool identifies prospects that are likely to be interested in automated damage control, by filtering for businesses with overall_ratings > 4.0 (likely pay attention to reputation management), reviews with ratings <= 3 (likely to impact customers visiting their page), owner_response = is_null (likely need to write a response, so you can offer to save them the work).

The tool uses a hosted n8n Chat interface so users can indicate preferences conversationally, generate sales scripts tailored for specific business and recent reviews, and delivers them to users via Email or Slack. The workflow uses n8n's AI Agent node with tool chaining, including Bright Data tool calls, to allow for flexible requests while still providing sufficient guardrails to optimize for good user experience.

Demo

Video Walkthrough

(Actual demo from 5:40-8:40)
Please excuse the poor audio!

Chat & Slack Delivery

Email Delivery

n8n Workflow (GitHub Gist)

https://gist.github.com/myopicOracle/163fed9602906eb99e50da02211481d3

Technical Implementation

Node-by-Node Breakdown

1. Chat Trigger

Starts the workflow in response to a user message from the chat UI. Passes the message and session context downstream.

2. Chat Trigger Agent B1

Acts as the central AI orchestrator—interprets user input, manages workflow steps, and decides when to call each tool node. Ensures a smooth, structured conversation.

3. Create Snapshot Tool

Requests business review data for the user’s chosen country. Output is a unique snapshot_id for later retrieval.

4. Loading Display Tool

Keeps users informed during data processing, displaying a loading message and GIF while waiting for results.

5. Retrieve Snapshot Tool

Checks snapshot status and retrieves review data once available, enabling the workflow to proceed at the right time.

6. Extract Top Reviews Tool

Identifies and summarizes the top 5 negatively reviewed businesses from the collected data, preparing options for user selection.

7. Generate Sales Script Tool

Creates a customized phone outreach script and a suggested public review response for a selected business. (Currently disabled.)

8. B1 LLM

Handles language understanding and generation, supporting the agent’s conversational intelligence.

9. B1 Chat Memory

Maintains conversation history and workflow context, allowing for continuity and effective state management throughout the session.

10. Send Email Tool

Sends generated scripts to the user’s email, if requested.

11. Slack Message Tool

Delivers scripts via Slack, when selected by the user.

How Nodes Are Orchestrated

Agent B1 is the command center: It interprets user input, maintains workflow state, determines which tool node to call and when, and coordinates the entire workflow with strict sequence control.
Data flows from the chat entry, through the agent, then to data collection, extraction, script generation, and finally to the chosen delivery node.
Memory and LLM nodes guarantee context awareness and conversational quality across multiple user interactions.
Delivery nodes are only triggered upon explicit user request, ensuring privacy and control over script sharing.

Bright Data Verified Node

I mainly leveraged the "Filter dataset" and "Retrieve snapshot contents" modes of the Bright Data Verified Node as tool calls for my AI Agent, but they're also available as standalone nodes, and you can do stuff like: store snapshot, filter dataset, get dataset metadata, get snapshot data, get snapshot metadata, split snapshot, list datasets, list snapshot ids, download snapshot, get results, check batch status, extract from url, start batch extraction, unlock and extract url.

The Bright Data Tools sub-nodes for AI Agents were the key pieces to my workflow’s ability to provide timely, high-quality sales prospects. It enabled the automated retrieval of up-to-date Google Maps business reviews tailored to the user's selected country. When the workflow begins, once the user specifies the desired country, the agent validates and formats the country code, then passed it into the "Group Filter" parameter of the Bright Data node, which will generate a snapshot based on any number of filter parameters you set (although you can also apply single filters). Then the Bright Data node returns a snapshot_id that you can use to retrieve that specific snapshot downstream or save into memory.

By leveraging the Bright Data Verified Node, this workflow can programmatically gather, monitor, and process business review datasets without manual intervention. This integration is what empowers the automated review extraction, selection of target businesses, and ultimately, the hyper-personalized script generation for sales outreach.

In summary, the Bright Data Verified Node transforms raw online review data into an actionable sales asset, tightly integrated in the workflow for maximum automation, reliability, and value for business consultants and agencies.

Journey

n8n

I made the mistake of jumping right in without doing a deep dive on docs for each of the nodes I was using, or looking up related nodes to analyze trade-offs. Very quickly, I was putting out fires left and right without a clear path in mind. Looking back, that's also where a lot of the learning happened, but I wish I knew these things before I started:

n8n has a "Chat with Docs" modal in n8n's docs you can use to get far more accurate answers on syntax and trade-offs, because a lot of the answers you'll get from generic LLM's reference older variants of the same node, so it might give you advice that is no longer optimal.
Some of n8n's AI Agent and LLM/Tool calls use Langchain under the hood (n8n-nodes-langchain.llmChatOpenAI; @n8n/n8n-nodes-langchain.toolCode; @n8n/n8n-nodes-langchain.chatTrigger) so you can transfer over any knowledge you have over to get a better understanding of how to structure calls and prompt engineer your way to victory.
The browser AI Assistant can directly access your workflow's JSON (I think) as well as the n8n docs, so it's very helpful for debugging issues that could be attributed to several points of failure in your workflow. However, the n8n docs remain the ultimate source of truth when setting up your nodes, especially with optional parameters or new nodes.

Bright Data

Firstly, shoutout to @depapp for letting me know that we can just pull snapshots instead of buying the whole marketplace dataset. Before I found this out, I was trying to create CommunityPulse's functionality with entirely manual HTTP requests to several Google Places API endpoints. Massive pain the derriere. Why? Because the Places endpoint only returns a list based on a lat/long, so you first need to call the Geocode endpoint to transform user requests for city/country.

But the list that the Places endpoint returns doesn't contain any reviews! So you need to call the Reviews endpoint. Guess what? The Reviews endpoint only allows calls to be made with individual place_id URL params, so you need to loop over your array of extracted place id's and call the Reviews endpoint individually for each business- quick way to get blocked!

Bright Data for save: In addition to custom web scrapers you can set up quite quickly by using the Bright Data Agent to generate a starter script, there's also a massive collection of unlocked data in Bright Data's Dataset Marketplace. And it's pretty wild the range that it spans. Off the top of my head, I can think of several commercial B2B products (Bloomberg, MergerMarket, S&P CapitalIQ, MorningStar, etc.) that charge boatloads of money for people to access similar data that's available at a fraction of the price. 👀

Parting Thoughts

I went from putting out fires to drinking from the firehose, but after that initial hump (which many of you probably don't relate to), it gets really fun with the speed that you can prototype ideas and discard them without looking back. I've only barely scratched the surface on the number of integrations that n8n has, or all of the data that's at your fingertips with Bright Data, without exorbitant costs or advanced scripting skills. If you're still all-in on BeautifulSoup, would highly recommend checking out Bright Data's custom Web Scrapers and IDE that you can use to build and test queries.

Comet Review: The First True AI-Native Browser?

Gary Xia — Sun, 17 Aug 2025 05:59:06 +0000

A few days ago, Perplexity surprised (baffled?) the tech world with an unsolicited $34.5B bid for Chrome. While it makes sense in context of Google's need to divest after losing a massive antitrust case over their search engine, cited as "illegally exploiting its dominance to squash competition and stifle innovation", the move still seemed pretty bold. Until I actually spent time using their new Comet Browser and its built-in Comet Assistant.

I recorded a 6-minute demo to showcase what I think sets Comet apart. Not just as "yet another AI browser," but as a potential blueprint for how we'll all use the web in the near future.

Here's What Struck Me

1. AI-Powered Email Triage (with real multi-account support)

Comet's assistant can analyze multiple inboxes, summarize unread threads, and even propose and schedule meetings, all inside one interface. This is a big leap from the inbox-zero tools built for just Gmail or Outlook.

2. Multi-Tab Doc Analysis & Instant Code Generation

The "@" shortcut lets you reference files, tabs, or even parts of code/documents, and get summaries, explanations, or working code samples. Context switching drops way down.

3. Ridiculously Fast Video Summarization

Open Comet Assistant on an hour-long YouTube video, click the recommended summarize prompt, and in under five seconds, you get a solid summary and key bullet points. Wildly useful for anyone trying to keep up with technical talk videos, conference sessions, or even product demos.

4. Voice Navigation That Actually Works

You can issue real, complex navigation commands by voice (not just "scroll down"). Honestly, this is the first time voice navigation hasn't felt like a gimmick to me.

Why This Matters

This isn't just "browser wars, round 23." It's a rethink of what browsers should even do, especially for anyone working in tech/product spaces:

The "AI-to-Web" layer isn't just about convenience. It's starting to change how we build, test, research, and ship.
Perplexity isn't just layering AI on top of a browser. They're making the browser itself a command interface for web automation.

Not All Sunshine and Rainbows

Perplexity's search engine, which is the default in Comet Browser, doesn't have a toggle to make AI summaries optional. It's always a full-on essay, but sometimes I just need the top result from an indexed search.

In contrast, Google's search engine AI summary is typically only a few sentences long, with an optional "expand" toggle for a deeper dive, and indexed search results right below the summary.

Personally, I think more "progressive design" could be considered. Giving users more agency and control over their workflows is essential, especially as AI is increasingly shifting power away from the end user.

Try It!

If you work in tech, AI, or digital products, I think it's worth checking out. My full demo (above) walks through the flows I found transformative, but I'm very curious what power users and skeptics think.

Where do you see AI-native browsers going? What can they change versus what still needs old-school tabs? Let's debate in the comments.

Tags: ai, productivity, discuss, tooling