DEV Community: Ankur Bansal

The Day I Automated Web Check-in: A Tale of Wonder and Worry

Ankur Bansal — Sun, 05 Oct 2025 16:50:05 +0000

When AI Makes Life Too Easy
Last week, I did something that felt like pure magic — I let Perplexity’s browser assistant (Comet) automate my entire web check-in process. You know that annoying ritual of opening your airline’s website, entering your booking reference, selecting seats, and downloading boarding passes? Gone. Just… poof… automated.

And honestly? I’m both thrilled and terrified.

Why Check-in Has Always Been Stupid
Let me be blunt: I’ve always thought web check-in is stupid.

Think about it. I’ve already booked my flight. I’ve already paid for it. The airline knows I’m coming. So why do I need to log back in 24 hours before the flight to click a button that says “Yes, I’m still coming”?

It’s busywork. Digital busywork dressed up as “convenience.”

For a 5-year-old: It’s like if you told mama you want pizza for dinner, and then she makes you come back an hour later to tell her again that you still want pizza. Why? She already knows!

For the CS grad: From a systems design perspective, check-in is legacy architectural debt. With real-time inventory management and sophisticated no-show prediction algorithms, requiring manual check-in is an outdated process masquerading as user engagement.

The whole system is backwards. If I don’t want to fly, I should cancel. Otherwise, assume I’m showing up. But airlines insist on this redundant confirmation ritual and the ridiculous seat selection scramble.

It’s stupid. It’s always been stupid.

So when I realized Perplexity’s Comet browser could make this stupidity disappear? That felt like justice.

The WOW Moment
Let me paint the picture. It’s 24 hours before my flight. Normally, I’d be scrambling — juggling kids, wrapping up work, desperately trying to check in before the good seats disappear. But this time? I just told Comet what I needed.

I watched as the browser opened itself, navigated to the airline website, filled in my details, selected my preferred aisle seat (14C, if you’re curious), and downloaded my boarding pass.

All. By. Itself.

Wow. That deeper, almost spiritual amazement when technology crosses from “cool” to “is this even real?”

For a 5-year-old: Imagine if you had a magic friend who could see your computer screen and do all the boring stuff while you play. That’s what Comet did!

For the CS grad: We’re talking about an AI browser assistant with autonomous navigation capabilities that can interpret natural language instructions, navigate dynamic web forms, and complete multi-step workflows with contextual understanding — not hardcoded paths.

The Creeping Doubt
But here’s where my engineer brain kicked in, and the magic started feeling… murky.

Who’s responsible if something goes wrong? If Comet books the wrong flight or accidentally cancels my booking — is that on me? On Perplexity? On the airline?

The Legal Compliance Black Hole
Most airline websites have Terms of Service that explicitly prohibit automated access. By using Comet to automate my check-in, I’m operating in ethically murky waters. I might be violating their terms of service.

The irony? I used an AI assistant to automate a stupid process that shouldn’t exist, and now there’s this nagging feeling that maybe I’m doing something… wrong.

For a 5-year-old: The airline website has rules, and even though the rule seems silly, using my magic computer friend might be breaking those rules.

For the CS grad: Legal frameworks haven’t caught up with AI browser assistants. Most Terms of Service don’t distinguish between personal automation tools and malicious bots — and that’s the problem.

Why This Matters (And What Should Actually Change)
If AI browser assistants can automate check-ins, they can automate tax filings, insurance claims, job applications, and financial transactions. Multiply that by millions of users, and systems designed for human-paced interaction get hammered. Fraud detection systems can’t distinguish between helpful AI assistants and malicious scrapers.

But here’s my take: Maybe some of these processes deserve to be automated away.

What Should Actually Change
Airlines should eliminate stupid processes:

Auto-confirm passengers 24 hours before flight by default
Let passengers opt-out if plans change (reverse the model)
Auto-assign seats based on booking preferences
Only require interaction if there’s an actual change needed
Terms of Service need modernization: Current ToS are written for a world where “bots = bad” was universally true. But personal AI assistants are legitimate tools. ToS should:

Distinguish between personal automation and commercial scraping at scale
Explicitly allow AI browser assistants for personal, non-commercial purposes
Focus on protecting against actual harm (server overload, fraud) rather than blanket banning automation
We need a new digital social contract that recognizes personal automation as a right while restricting commercial-scale abuse.

The Honest Truth
What Comet did is impressive, convenient, and potentially wrong. Just because something is technically possible doesn’t make it ethically clear. Just because a process is stupid doesn’t give us the right to violate terms we agreed to.

But we’re in a transition period. The rules haven’t caught up with the technology. Your discomfort is appropriate because the situation is genuinely ambiguous.

The future belongs to people who can navigate ambiguity with integrity — who can say “this rule is stupid, AND I’m still going to be thoughtful about how I handle it.”

What You Should Do
Immediate actions:

Use Comet for personal tasks, but don’t build a business around it
Recognize you’re in a gray area — don’t pretend it’s perfectly legal
Don’t abuse it (don’t hammer servers, don’t scale it commercially)
If you get warned, stop and comply
Shape the future:

Write about your experiences. Public discourse is how norms evolve.
Tell airlines their processes are outdated and their ToS are hostile to legitimate personal automation
If you believe check-in should be automated, build the system that does it the right way — with airline partnerships, proper APIs, and legal clarity
This is how you get rich, famous, and successful: not by hiding in gray areas forever, but by being early to a legitimate trend and helping formalize it.

Final Thoughts
The wow moment is real. AI browser assistants like Comet are genuinely magical. But we’re in a transition period where the rules haven’t caught up with the technology.

I automated away a stupid process. It felt amazing. It also felt… questionable. And maybe that’s exactly the conversation we need to be having.

Use the tool. Question the system. Advocate for change. The discomfort you feel? That’s not weakness — that’s wisdom trying to catch up with capability.

And in a world where AI can do almost anything, that gap between what we can do and what we should do might be the most important space to navigate carefully.

ChatGPT Prompt Engineering for Developers

Ankur Bansal — Thu, 02 Oct 2025 16:52:15 +0000

Course: ChatGPT Prompt Engineering for Developers
Source: DeepLearning.AI Course
Instructors: Andrew Ng (DeepLearning.AI) & Isa Fulford (OpenAI)

Why Should Developers Care About Prompt Engineering?
AI is changing how we code, build products, and even debug. If you’re an engineer looking to level up your LLM application game—or just tired of getting inconsistent results from ChatGPT—learning prompt engineering is essential.

This post condenses the comprehensive DeepLearning.AI course into actionable, developer-ready best practices you can apply right away, with real-world coding and product examples.

Table of Contents
Understanding LLM Types

Core Prompting Philosophy

Best Practices Checklist

Iterative Prompt Development

Summarization Techniques

Inferring Insights

Text Transformation

Content Expansion

Building Chatbots

Your Next Steps

Understanding LLM Types
Base LLMs:
Predict next-word sequences; may not follow instructions.

Instruction-tuned LLMs:
Specifically trained to follow instructions. Always prefer these for practical use—think gpt-3.5-turbo, gpt-4, etc.

Core Prompt Engineering Philosophy
Treat LLMs as highly skilled collaborators who lack your context—be explicit, be clear, and always check your instructions.

Best Practices (Checklist for Devs!)
Write clear, specific instructions

Use delimiters (e.g. triple backticks) to separate user content

Request structured output (JSON/HTML)

Instruct the model to check preconditions

Provide examples (few-shot prompting)

Chain-of-thought reasoning for complex tasks

Specify output format/order

Ask for reasoning before answers

Validate critical outputs (watch for hallucinations)

Ask for citations/quotes in summaries

More Developer-Focused Sample Section
Iterative Prompt Development: The Dev Way
Start with a simple prompt.

Analyze the output.

Refine requirements: add structure, word limits, or technical requirements.

Automate testing: For production, validate against real data and edge cases.

Reminder: “Perfect” prompts rarely exist—iteration is success.

Practical Prompts: Real Use Cases
json
// Ask for structured, machine-readable output
{
"sentiment": "positive",
"emotion": "joy",
"main_topic": "product feedback"
}
text

Example: Summarize Product Reviews

Prompt:
Extract the main feature request and overall sentiment from the following review:

Output as JSON.

Your Next Steps as a Developer
Experiment: Try integrating LLMs into your projects.

Document: Keep a log of prompt outcomes (helps rapid improvements).

Iterate: Small tweaks, big results.

Share: Comment with your own prompt hacks below.

Keep Learning: New LLM capabilities land regularly—stay curious.

About This Guide:
Summarized from the DeepLearning.AI course by Andrew Ng and Isa Fulford. Highly recommend taking the course for direct hands-on practice.
Course Link: https://learn.deeplearning.ai/courses/chatgpt-prompt-eng/