The file transfer category looked finished a decade ago. WeTransfer-style products had captured the casual human send; Box and Dropbox had captured the team-shared folder; enterprise had its own deck. The product shapes settled. The interface conventions calcified. New entrants stopped being interesting to look at.
Then something quieter started happening, and it’s still mostly invisible from the outside. The hand on the trigger started being software — more often than the year before, less often than the year after.
Where the displacement actually shows up
A research analyst’s IDE assistant ships a generated chart to a colleague. A finance bot exports a monthly-close PDF to a recipient list. A coding agent ships a build artifact to a reviewer. Five years ago every one of those was a human dragging a file across a browser. Today many of them are a tool call, a JSON body, and a fire-and-forget.
The file still moves. The recipient still receives the same envelope. But somewhere upstream, a decision that used to be "open the browser, drag the file, type the email" became "call this endpoint, hand it a path, get a link back".
The product shape that quietly ate the browser
The browser-shaped file-transfer product was optimised for one caller: a person. Drop zone, copy-link, recipient field, click send. The UI is the contract. Beautiful for humans, opaque to automation.
The shape that’s slowly absorbing the category ships the same primitive with a different contract surface: an OpenAPI document, a stdio MCP binary, an A2A AgentCard. The browser flow stays — for the humans that still drag and drop — but it stops being the only entry point. It becomes one of several, and the others compound differently.
Why this is "slowly, then fast"
The browser flow doesn’t shrink. It just stops growing.
The agent flow does grow — and it grows by compounding. Every new MCP-aware host (Claude Desktop, Cursor, Cline, Continue, and whatever ships next quarter) makes "send a file" a one-line request from an AI assistant. Each host that adds the tool exposes a new audience overnight. The browser flow doesn’t add million-user audiences by absorbing a new platform; the agent flow does, every six months.
That asymmetry is the whole story. The category isn’t being eaten because the new shape is better — it’s being eaten because the new shape compounds and the old one doesn’t.
The hand on the trigger is software more often than yesterday and less often than tomorrow.
What this means for anyone shipping a "send" surface
If you’re building anything that touches "send something somewhere" — file transfer, document delivery, notification, escalation, archive — the agent surface is no longer optional. The contract surface (OpenAPI + a stdio MCP binary + an AgentCard) is becoming what an SDK used to be: the cost of entry, not a differentiator.
Pretty UIs still matter. They’re just no longer the only way in. The products that ship only a UI are quietly accumulating an integration gap they’ll need to close — and the longer they wait, the more it looks like the same gap mobile-only apps had when they discovered their best customers had switched to web.
Where we sit
ReTransfer started with the human flow and added the agent surfaces as the question kept being asked. We didn’t predict this shift; we noticed it. We described the surfaces we ended up shipping in our post on MCP, A2A, and REST. What we didn’t say there is why we shipped all three. The reason is this post.