Say goodbye to pain of trip planning and deal hunting.
Let me show you how I'm using #ChatGPT Plugins like Kayak and Expedia to plan my vacations.
What's your biggest struggle when planning a trip?
1. Kayak is a game changer.
I can chat with it like a personal trip planner, setting my budget, dates, car rentals, even specific dining needs!
Here's how I got it to plan a trip to Austin, Texas for under $1,000 with #keto friendly options. Seen anything like it?
Obviously this trip is over $1,000. No worries, just ask ChatGPT for the damage report, and you get a nice breakdown.
How many of you will find this helpful for your budgeting?
2. Let's talk about Expedia plugin. It gets into details and ask more specific questions, which can slow things down a bit. I prefer the speedy and straightforward responses from Kayak. Here's how Expedia handled the same Austin trip.
However, Expedia plugin didn't catch all my requests at once - it missed the flights, car rentals, and dining options. Had to remind it!
So, the cost breakdown from Expedia plugin is roughly same as Kayak. That's because that's just ChatGPT doing its thing!
That's a wrap!
It's clear that these plugins will completely transform your trip planning! What's your take?
Don't forget to follow me @minchoi for more AI contents and Retweet/Like first tweet below👇
Google AI Studio just got an insane upgrade, full-stack vibe coding 🤯
Now anyone can build production-ready apps with auth, databases, and real API integrations from one prompt
5 wild examples + prompts (🧵👇)
1. 3D multiplayer claw machine
PROMPT: "Create a 3D multiplayer claw machine game using Three.js where players compete for control to grab prizes. Players join a shared arcade claw machine, take turns controlling the claw, and compete to grab the highest-value items before others. Core gameplay: - Shared claw machine with one claw, multiple players watching/waiting - Turn-based: ~15 second timer per player to position and drop - Prize pit filled with primitive objects (such as cubes, spheres, etc.) with varying point values based on size - Grabbed prizes add to your score, missed grabs end your turn - Physics-based claw grip (items can slip!) Multiplayer: - Real-time shared state via WebSocket - Player queue system with live turn indicator - Spectators watch current player's attempt in real-time - Live scoreboard showing all players' totals - Drop-in/drop-out, auto-skip AFK players Visuals (arcade cabinet aesthetic): - Glass-walled machine with metallic frame and neon accent lighting - Shiny collectibles as primitive objects with subtle bounce - Claw with animated grip mechanism and cable - Dramatic lighting when claw descends - Bloom and reflections on glass - Camera: spectator view default, shifts to top-down during your turn Controls: arrow keys/WASD or drag to move claw, spacebar/tap to drop. Make joining instant and easy. Add a button in the UI to open in new tab for easy multiplayer. Ensure UI doesn't block canvas pointer events"
2. Build a first person laser tag game
PROMPT: "Build a first-person laser tag game in an indoor arena. Core WASD movement + mouse look (pointer lock) Click to fire laser (raycast hitscan) Simple box/corridor arena with obstacles for cover Visuals: Indoor arena lighting (ambient + point lights) Laser beam: thin cylinder or line that flashes briefly on fire Hit feedback: flash on target, particle burst Gameplay: 3-5 AI opponents (simple patrol + chase behavior) Vest hit = 3 second disable (target flashes, can't shoot) Score counter, maybe a timer Respawn after getting hit Stack: React Three Fiber @react-three/drei (PointerLockControls, useKeyboardcontrols)"
"7 things I wish someone told me before I almost gave up on OpenClaw"
Especially #1 and #4 are the unlock.
And how I set up mine.
Bookmark this. 🧵👇
1. "Don't run everything through your best model"
The post says: cheap model as primary, strong as fallback.
I do the opposite.
My setup:
→ Opus 4.6 as my MAIN brain (strategy, decisions)
→ Haiku for heartbeats ($0.001 per check)
→ Sonnet for sub-agents (research, drafting, browser)
→ Codex GPT-5.3 for all coding (for now)
Don't default to cheap and escalate to smart.
Default to smart where it matters and use cheap where it doesn't.
My heartbeat costs dropped 70% on Haiku. But my main brain is Opus 4.6 because that's where every important decision happens.
2. "Your agent needs rules. A lot of them."
Most underrated tip in the post.
My AGENTS .md is 4 pages:
→ Identity, personality, boundaries
→ 12 anti-patterns it must NEVER do
→ Model routing table
→ Memory rules (when to write, when to read)
→ Platform-specific formatting (Signal = plain text, Discord = no tables)
"You are a conductor, so conduct."
Best line in the whole post. Your agent is only as good as the rules you write for it.