Why
Tracking where computer-use agents actually stand in production, not in demos. The a16z piece is built on interviews with real operators — a CPG data platform running 15-20M portal interactions a month, a systems integrator running 27 live workflows — not just benchmark numbers. A useful checkpoint for judging whether an "agent operates a browser or UI" idea is worth building now versus still a research toy.
The headline number is real and the argument is that it stopped mattering. OSWorld-Verified, a real-desktop benchmark, went from 42% a year ago to 85% today, above the roughly 72% human baseline. But once models cleared a good enough bar, buyers stopped comparing models and started buying infrastructure — verification, escalation, error handling. The model is no longer the product.
And the other end of the same question is a much smaller, much cheaper answer that this catalogue kept as a separate card for no good reason. Alibaba's page-agent is an in-page GUI agent: one script tag or npm install, and natural language operates that page's UI. No screenshots, no browser extension, no headless browser — it reads a text-based DOM, connects to any LLM you bring, has a beta MCP server, and is built on browser-use. The general capability question and the specific embedding decision are the same question at two scales, and holding them together is what makes either useful: the a16z piece says the bottleneck is now operations, and page-agent is the case where almost all of the operations disappear because the agent only ever touches one page you control.
The failure mode that matters most is the one neither half advertises. It is not the UI click failing — it is silent failure with no verifiable signal at run time. An insurance claim that "succeeded" on screen and stalls two days later on a phone call nobody logged. That is why verification, not accuracy, is what buyers pay for, and it is the same shape as the-harness-not-the-model: the interesting engineering is around the model, not in it.
How it works
Where the benchmark went, and why it stopped mattering
| A year ago | Today | |
|---|---|---|
| OSWorld-Verified | 42% | 85% |
| Human baseline | ~72% | ~72% |
| What buyers compared | Models | Infrastructure |
Once models cleared the bar, the purchase became verification, escalation and error handling. A common pattern: an agent runs a workflow once, caches it as deterministic code, and is only re-invoked when something breaks — so cost falls over a workflow's lifetime rather than staying flat.
The cost comparison that decides whether to build
| Option | Rate | Against agent inference |
|---|---|---|
| Agent inference | ~$6-8/hr ($3-15 range) | — |
| Offshore BPO | ~$10/hr | Roughly break-even today |
| Fully-loaded US labour | $30-45/hr | 70-80% margin |
Two scales of the same capability
| General computer use | In-page agent (page-agent) |
|
|---|---|---|
| Surface | Any application | One page you own |
| Input | Screenshots, accessibility trees | Text-based DOM |
| Needs | Extension or headless browser | A script tag |
| Operations burden | Verification, escalation, error handling | Almost none — the surface is yours |
| Fails at | Silent failures downstream | Anything the DOM does not express |
| Model | Whatever the vendor ships | BYO, any LLM; beta MCP server |
The right way to read the pair: the general case is where the operations layer is the product, and the embedded case is the one where you can skip most of it — because you control the surface, so a failure is visible in your own application rather than two days later on somebody else's phone call. A landing-page or SaaS copilot, or smart form-filling, is the natural first use.
The failure mode to design against
Not the click that fails — the one that appears to succeed. No verifiable signal exists at run time, so the workflow reports done and the real state diverges silently. Verification is not a feature bolted on afterwards; it is the reason the infrastructure layer exists, and it is why the honest first question for any browser-agent idea is how would I know this went wrong rather than can the model do it.
The authoring end of the same question
This card asks whether an agent can drive a computer and concludes that once it can, what you buy is operations. demonstration-is-not-a-specification arrives at the same conclusion from the other end — how a task gets stated — by way of the screen-recording interfaces that a startup and a large vendor shipped weeks apart in 2026. The short version: recording lowers the cost of authoring a task and leaves the cost of running it reliably exactly where it was.
Where it lands in Jayverse
- Rabbit: treat verification, not task success, as the product surface for agent-driven mandates. Session keys and mandates already let an agent act on a user's behalf; log a false-promise rate — the share of agent actions that reported success but whose downstream effect never landed — the same metric simulate-before-sign proposes for Wallet.
- Wallet: give any agent-driven flow through the embedded wallet its own escalation path. A mandate-authorized agent action that looks successful but whose effect never lands needs a defined "what happens next," not a silent stall two days later.
- Game: if the Unity street ever hosts an in-page agent, scope it as page-agent's narrow case. Keep it limited to the DOM Jayverse itself controls — its own boards and UI — rather than general computer-use, since that is what keeps the operations burden to almost nothing.
Key expressions
| Expression | 뜻 · 쓰이는 자리 |
|---|---|
| clear the bar | 기준을 넘어서다·통과하다 · 모델 성능이 충분한 수준에 도달했음을 말할 때. "once models cleared a good enough bar" |
| break-even | 손익분기(본전) · 두 비용을 비교해 이득도 손해도 없는 지점. "Roughly break-even today" |
| fully-loaded | (인건비 등이) 모든 부대비용 포함된 · 임금 외 복리후생 등 전체 비용을 말할 때. "Fully-loaded US labour" |
| BYO | bring-your-own의 줄임말, 직접 가져와 붙이는 방식 · 모델을 자유롭게 골라 연결할 수 있음을 표현. "BYO, any LLM; beta MCP server" |
| escalation | (문제를 상위·사람에게) 단계적으로 넘기는 절차 · 자동화가 실패했을 때 사람이 개입하는 과정. "verification, escalation and error handling" |
| verifiable signal | 검증 가능한 신호 · 작업이 실제로 성공했는지 확인할 근거가 있는지를 가리킴. "No verifiable signal exists at run time" |
| cache (as deterministic code) | 실행 결과를 고정된 코드로 캐싱해두다 · 반복 작업을 결정적 코드로 저장해 비용을 낮추는 패턴. "caches it as deterministic code, and is only" |
| smart form-filling | 지능형 자동 양식 입력 · 에이전트가 웹 폼을 대신 채워주는 활용 사례. "smart form-filling, is the natural first use" |
| the cost of authoring | 작업을 정의(기술)하는 데 드는 비용 · 작업을 시키는 절차 자체의 비용을 가리킴. "the cost of authoring a task and leaves" |
| OSWorld | 실제 데스크톱 환경에서 에이전트를 평가하는 벤치마크(a real-desktop agent benchmark) · 에이전트 성능 추이를 보여주는 기준점. "OSWorld jumped from 42% to 85% in a year" |
| CPG | 소비재(Consumer Packaged Goods) · 이 리포트가 인터뷰한 실제 운영 사례 업종. "a CPG data platform running 15-20M portal interactions" |
| BPO | 업무프로세스 아웃소싱(Business Process Outsourcing) · 해외 인력 위탁을 에이전트 비용과 비교하는 항목. "Offshore BPO" |
| a16z | 벤처캐피털 Andreessen Horowitz의 약칭 · 이 리포트를 작성한 기관. "The a16z piece is built on interviews with real operators" |
| browser-use | 브라우저를 조작하는 오픈소스 에이전트 라이브러리 · page-agent가 기반으로 삼은 도구. "and is built on browser-use" |
← All Knowledge Notes · Workspace Index · Top ↑ · Open on jaylabs.xyz →