Agents Can Work Inside Documents Now — But Who Handles the Cross-App Repetitive Tasks?

Today's GitHub Trending (Sept 29, 2026) sent a clear signal: AI agents are moving from chatbots to actual office workers.

dream-num/univer redefined spreadsheets, docs, slides, and PDFs as "The Office Harness for AI Agents" — a unified runtime where agents can read, write, and manipulate office documents programmatically.

paperclipai/paperclip became "the open-source app everyone uses to manage agents at work" — a unified dashboard for tracking multiple agents.

debpalash/VoiceStudio brought voice cloning, dubbing, and transcription fully local — 646 languages, no cloud, no API costs.

The Gap: What Happens Outside the Document?

These projects are impressive. But they all operate within boundaries:

• univer gives agents a document runtime — but what about tasks that span multiple applications?
• paperclip manages agent state — but management ≠ execution.
• VoiceStudio handles voice locally — but voice is just one input channel.

Consider a realistic office scenario: every morning, an agent needs to check three email accounts, extract attachments, parse invoice data with OCR, enter records into a legacy CRM (no API, only web UI), generate a summary spreadsheet, and post it to a team chat.

This workflow touches email clients, a browser-based CRM, a spreadsheet app, and a chat tool. No single document runtime covers all of that. No management dashboard executes the steps.

What's Missing: The "Hands"

What's needed is an execution layer that can drive desktop and browser operations the way a human would — clicking buttons, filling forms, copying data between apps — but unattended, on schedule, with agent-level intelligence for decision-making.

This is exactly where iflytek/astron-rpa fits in. It's an Agent-ready RPA suite: the agent handles the "what" and "why" (decision-making, data understanding, exception handling), while the RPA layer handles the "how" (clicking, typing, navigating, data entry).

Think of it this way:

Layer	What it does	Today's trending example
Document runtime	Agent reads/writes within a document	univer
Agent management	Track and orchestrate multiple agents	paperclip
Cross-app execution	Agent operates across desktop & browser apps	astron-rpa

The Full Picture

For true unattended office automation, you need all three layers:

1. A runtime where agents can work with documents (like univer)
2. A management layer to track what agents are doing (like paperclip)
3. An execution layer to drive cross-app repetitive tasks (like astron-rpa)

And when the execution layer needs capabilities like voice transcription, OCR, or translation — that's where iflytek/iFly-Skills comes in: iFLYTEK's official skill collection (voice, OCR, translation, proofreading, multimodal) that can be plugged into agent workflows.

Takeaway

The trend is clear: agents are becoming office workers, not just chatbots. But an office worker who can only work inside one document isn't enough — they need to move between apps, handle repetitive processes, and do it all without someone watching over their shoulder.

That's the gap between "agent-assisted" and "agent-autonomous" — and RPA is the bridge.