Last-Mile Data Entry for AI
David Mlcoch, co-founder & CEO of Asteroid, on browser automation and the last mile problem of AI
The key constraint on vertical AI is not generating an answer, it is getting that answer into the old system that actually runs the business. In healthcare scheduling, insurance quoting, and supply chain portals, the system of record is often a browser based tool with no usable API, so a human still has to click through screens and type data. Browser automation turns that final handoff into software, which is why it becomes the execution layer for voice agents and other AI front ends.
-
This is why voice AI alone stalls in traditional industries. A voice bot can take the call and collect the facts, but the appointment, claim, or order is not complete until someone logs into the old portal and updates the record. Asteroid is built around that exact gap, especially in insurance and healthcare.
-
The product split in browser automation is concrete. Browserbase gives developers hosted browsers and tooling to run automations in the cloud, while Asteroid is packaging the same underlying browser interaction layer into a workflow tool that non technical operations teams can supervise and reuse for repetitive back office work.
-
The incumbent benchmark is still large. UiPath finished fiscal 2026 with $1.611B in revenue and $1.853B in ARR, which shows how much enterprises already pay to automate brittle screen based work. The opening is that LLM guided agents can handle changing pages and popups far better than older rule based scripts, especially on messy legacy portals.
The next step is from single task automations to full agent workforces that handle intake, browser entry, spreadsheets, and exceptions in one flow. As long as core industries keep running on old web software, the winning AI stack will pair reasoning models with a reliable browser execution layer that can operate those systems at enterprise scale.