Last month, a friend who'd been using my Chase credit card for a year asked "so how much do I owe you?" In the old world, this would've been a nightmare — digging through 12 months of PDF statements, cross-referencing Venmo payments and wire transfers, building a spreadsheet, arguing over line items.
Instead, I told my AI agent: "Parse all Chase statements from 2025, identify Rahul's charges, subtract his payments, build a breakdown page." Twenty minutes later, there was a URL. Every transaction, every payment, every balance — traceable to source documents. I sent the link. No ambiguity, no arguments, no spreadsheet tennis. Done.
This keeps happening. Every decision I make now produces a webpage. And it's changed how I think about decisions entirely.
The Pattern
Here's what my decision-making looks like now:
- I describe what I need to my AI agent
- It researches — web searches, PDF parsing, data analysis, API calls
- It builds a comprehensive HTML page on my server
- I review, iterate, ask for changes
- The page becomes a permanent, living reference
The Key Insight
The document IS the research. It's not a summary of research I did — it's the direct output of AI doing the research. The page doesn't describe findings; it contains them.
For solo decisions, this is already transformative. But it gets really interesting when other people are involved.
Group Decisions: The Shared Page
I'm flying home next month for a friend's wedding. Planning involved flights, a week-long itinerary, restaurant recommendations, outfit coordination for four wedding events, and transport logistics. The kind of thing that usually drowns a WhatsApp group in 300 messages over two weeks.
Instead: I told my agent what we needed. It scraped Google Flights for options, researched restaurants with Google Maps links, built a day-by-day schedule, and published it as a webpage. I dropped the link in our family WhatsApp group: Trip Plan — Feb 2026 (redacted demo — the real one has actual flight details and phone numbers).
Mom wants to add a temple visit on Day 3? She says so in the group. I update the page. (Soon, a sandboxed AI agent in the group will do this automatically — more on that below.) The page is always the source of truth. No more "scroll up to find what we decided."
Six Decisions, Six URLs
Here's what this looks like in practice. These are all real — pages I've built and use.
The Ledger: "How much do you owe me?"
A friend used my credit card for about a year. The AI parsed 12 months of Chase PDF statements, identified every transaction that was his, cross-referenced Venmo payments and wire transfers he'd sent me, and built a public ledger page. Every single line item links back to its source — which statement, which page, which date. The running balance updates automatically. I sent him the URL. He checked a few entries, confirmed the math, and paid. A conversation that could've been awkward became a 30-second link share.
The Itinerary: Family trip planning
Flying home for a friend's wedding. The AI built a complete trip page — flight options with prices, a day-by-day schedule, restaurant recommendations (with Maps links), outfit planning for each wedding event (haldi, sangeet, wedding, reception), and local transport logistics. Published it, dropped it in the family group. Everyone can see the plan. One URL replaces forty messages.
The Wedding Outfit Guide: Shopping decisions without the stress
I had a friend's wedding coming up and needed to figure out what to wear. Instead of wandering through malls hoping for the best, I had the AI research it — 5 different outfit styles compared (bandhgala, sherwani, indo-western, western suit, tuxedo), with pros and cons for each at an Indian wedding. Then it found the best tailors in my city, compared bespoke vs ready-to-wear pricing, mapped out timelines for custom tailoring, and built a decision matrix. The resulting page had everything — shop addresses with click-to-call phone numbers, price ranges, Google Maps links, and four different action plans depending on how much time I had. One page replaced what would've been a dozen Google searches for tailors.
The Investment Plan: Where to put idle savings
I had some savings sitting idle and wanted a deployment plan. The AI researched 50+ mutual funds, narrowed to 4 based on my risk profile and goals, built a deployment schedule, ran 10-year projection scenarios, and created individual analysis pages for each recommended fund. This one's behind authentication — financial data isn't public. But it's a living document. When market conditions shift or my income changes, the projections update.
The Stock Report: Brokerage allocation
Similar pattern for US stocks. AI screened for high-growth candidates, deep-dived 12 stocks across 4 sectors, built a styled research report with bull/bear cases for each, risk matrices, correlation analysis, and specific allocation recommendations. Again, behind auth. But it exists as a page I can revisit, not a ChatGPT conversation I'd never find again.
The Tax Strategy: NRI transition
Moving from NRI to resident Indian tax status with assets in two countries. The tax implications are genuinely complex — DTAA treaties, FBAR reporting, capital gains treatment differences, timing of residency status changes. The AI researched all of it, built a comprehensive report with specific action items and deadlines, and set calendar reminders for each filing date. As tax laws change (and they do, every budget season), the document updates.
The Security Model
When you're giving an AI agent access to build and publish web pages, security matters. Here's how I think about it:
My main AI agent has broad access — it can read my files, run scripts, access my server. That's necessary for it to be useful. But when a page is shared in a WhatsApp group and other people interact with it, I don't want the group-facing agent to have that same access.
The solution: sandboxed agents. A group-facing agent gets access to exactly one file (the page it manages) and the internet (for fresh research). Nothing else. Can't read my other pages. Can't access my financial data. Can't run arbitrary commands. It's like giving someone a key to one room, not the whole house.
The Evolution of Design Docs
We've all written design docs. The ritual is always the same: spend days researching, synthesize into a Google Doc, share it, people leave comments, you manually incorporate feedback, the doc slowly rots as reality diverges from the plan.
The new version inverts the whole thing:
Old way: Human researches → human writes doc → humans comment → human updates → doc dies
New way: Human describes need → AI researches → AI builds living page → humans give feedback → AI updates → page stays current
The document doesn't summarize research — it is the research. Every claim is backed by data the AI actually gathered. And because the AI can re-run that research, the document can stay fresh in ways a static Google Doc never could.
There's a reason this works so well now: models have been heavily optimized for generating clean UI. Years of RLHF and fine-tuning on code generation means modern LLMs produce genuinely good HTML, CSS, and structured layouts out of the box. The car comparison table, the trip itinerary, the outfit decision matrix — these aren't ugly text dumps. They're well-designed, responsive pages that look like someone spent hours in Figma. The AI builds better UI for a one-off research page than most people would bother to build for a production app. That's what makes "every decision gets a URL" practical — the output is actually worth sharing.
Why This Matters
Every decision now has a URL. That sounds trivial. It's not.
When I need to remember which tailor had the fastest turnaround — there's a page. When someone asks "what time is the flight?", I don't have to dig through emails — there's a page. When I need to review my investment strategy, I don't have to reconstruct my reasoning from scratch — there's a page.
Decisions stop being ephemeral. They become artifacts. Referenceable, shareable, updatable artifacts. Your brain stops being the bottleneck for holding context because the context lives somewhere durable.
This is what decision-making looks like in the AI era. Not "ask ChatGPT and copy-paste the answer into a note." Build a living document that evolves with the decision. Share it with the people involved. Let it update as circumstances change. Go back to it in six months and it's still useful — maybe more useful, because it's been updated.
Every big decision I make now follows this pattern. The pages accumulate. They cross-reference each other. They form a personal knowledge base that grows with every choice I make.
The Pattern
Describe what you need → AI researches → living webpage → iterate → share. Every decision becomes a permanent, referenceable artifact. Your brain stops being the bottleneck for holding context.
The AI isn't replacing my judgment. It's giving my judgment a permanent address.
This blog post — and the platform serving it — was built and deployed by another AI agent system I run. That one handles personal life infrastructure instead of code: finances, memory, email processing, web publishing. I wrote about it here.