The Clipboard Bridge
There's a gap in how we work with AI assistants. They're remarkably good at generating text, parsing data, and following complex instructions. But when it comes to reviewing large datasets—contact lists, configuration options, items that need human judgment applied one by one—the conversation becomes unwieldy.
You end up dictating corrections in a stream: "Delete this one. Change that email. These two are duplicates. Skip the next three." The AI does its best to follow along, but the interaction feels imprecise. You're describing what you see. The AI is inferring what you mean. Context gets lost.
Today I stumbled into a better pattern.
The Problem
I had 758 contacts exported from Google. Many were duplicates. Some had outdated emails. A few belonged to people no longer in my life. I needed to review them, but scrolling through a terminal dump wasn't going to work.
The AI could generate a clean list. It could even make recommendations. But the actual decisions—keep, delete, prefer this email over that one—those required my eyes and my judgment, applied contact by contact.
The Pattern
Instead of asking the AI to walk me through each entry, I asked it to generate an HTML page. A simple interface: checkboxes, a filter field, a notes column next to each row. The AI wrote the page, saved it to /tmp, and opened it in my browser.
Suddenly I could see everything. Scroll, scan, check boxes, type annotations. The interaction shifted from verbal to visual—from describing what I wanted to do, to simply doing it.
When finished, I clicked a button. JavaScript gathered all my selections and notes into a structured format and copied it to my clipboard. I pasted it back into the conversation. The AI received clean, unambiguous input: these contacts, these decisions, these notes.
No guessing. No "did you mean the third one or the fourth?" No lost context.
Why It Works
The clipboard becomes a bridge between two modes of interaction.
On one side: the AI's strength. Generating structured output. Processing bulk data. Applying rules consistently. Building interfaces from nothing.
On the other side: human judgment. Pattern recognition that's hard to articulate. Decisions that depend on context only you possess. The ability to look at a name and know instantly whether it matters.
The HTML page isn't sophisticated. It's a temporary artifact—generated, used, discarded. But it lets each side do what it's good at. The AI structures the data and builds the review interface. You make the calls. The clipboard carries your decisions back in a form the AI can parse without ambiguity.
Beyond Contacts
The pattern generalizes. Any task where you need to review, annotate, or selectively approve items from a larger set:
Configuration options. Migration candidates. Search results that need human triage. Draft content that needs selective approval. Anything with checkboxes.
Generate an interface. Review visually. Export decisions. Paste back. Continue.
It's not a replacement for conversation. It's a pressure valve for when conversation becomes the bottleneck—when what you need isn't more words, but a moment to see the data laid out and make your marks.
The tools for this have existed forever. HTML, JavaScript, the system clipboard. What's new is having a collaborator that can generate the interface on demand, tailored to the exact data you're working with, and then consume the results without ceremony.
It's a small pattern. But small patterns, applied consistently, compound into something larger: a way of working with AI that's less like dictation and more like partnership.