Compiled Replay vs. Computer-Use Agent

One clinical task, two ways to automate it, one success check. The wedge in one line: for a task you have already demonstrated, a compiled replay is model-free, ~$0 per run, and faster on these measured trials. Each arm's success is reported separately. Generated straight from the repo's real benchmark results.json files; no number here is hand-typed.

Same task, measured success (19/20 vs 10/10) — compiled replay costs $0 and runs 1.8× faster than the claude-sonnet-5 agent

What this is. The economics of the 500th run of a known task: once a workflow has been demonstrated once, replaying it should not cost a model call.

What this isn't. A capability claim. Compiled replay is the wrong tool for a task nobody has automated yet — exploring an unfamiliar screen is exactly the agent's job. The agent explores; the compiled bundle exploits a demonstration.

Both arms drive the same vision-only interface against the same app; the only difference is whether a model is in the per-action loop. Reproduce with python -m benchmark.comparison_artifact.generate.

LEADfield result — real third-party app

OpenEMR public demo — a real third-party EMR

OpenEMR public demo: log in, search the demo patient, open the chart, scroll the dashboard to the Messages card, open Patient Messages, add a parameterized note, save · 18 compiled steps

measured success
19/20 · 10/10
compiled · agent — same arm-independent screen contract
model cost / run
$0 vs $0.5522
agent arm total $5.52 over 10 runs
latency p50
39.2 s vs 70.4 s
compiled is 1.8× faster at the median
Model cost per run
$0$0.2000$0.4000$0.6000USD / run (list price)compiled replay$0computer-use agent$0.5522
In this measured sample, compiled replay made zero model calls and incurred $0 in model API charges.
Latency (wall-clock)
0255075100secondscompiled p5039.2 scompiled p9541.0 sagent p5070.4 sagent p9582.6 s
Per-run wall-clock around the replay / agent loop only.
Illustrative repeat-run model cost (measured model $/run × N; not new runs)
modeled runscompiled model APIagent model API
1$0$0.5522
100$0$55.22
1,000$0$552
10,000$0$5,522

Agent values multiply the measured per-run model charge ($0.5522) at list price; an introductory $2/$10 rate applies through 2026-08-31. The compiled column projects the sample's zero model calls only. It excludes authoring, maintenance, compute, storage, review, and exception-handling cost and assumes the measured task and rates remain unchanged. The one-time demonstration was about a minute of human demonstration.

Both arms drive the same vision-only interface (screenshots in; pixel clicks / typed text / scrolls out) against the same target; the agent is claude-sonnet-5 with the computer-use tool, prompted with user intent, not steps. Success is one arm-independent OCR check applied to both arms — neither arm's self-report is trusted. Not CI-reproducible: a single shared public instance that every internet visitor mutates and that resets daily. Treat as a field result. figures: benchmark/openemr/results.json

One compiled run (#20) self-flagged expected-screen drift at step_017 and aborted. The corrected saved-row OCR check also rejects it: the retained final frame shows the note in the unsaved entry form, not in a saved message row. It counts as a failure.

CI-reproducible anchor

MockMed — the bundled demo clinic (CI-reproducible anchor)

MockMed triage: login, open first referral, create a Triage encounter, enter the note, save

measured success
100/100 · 20/20
compiled · agent — same arm-independent screen contract
model cost / run
$0 vs $0.2716
agent arm total $5.43 over 20 runs
latency p50
4.9 s vs 37.5 s
compiled is 7.6× faster at the median
Model cost per run
$0$0.1000$0.2000$0.3000USD / run (list price)compiled replay$0computer-use agent$0.2716
In this measured sample, compiled replay made zero model calls and incurred $0 in model API charges.
Latency (wall-clock)
0204060secondscompiled p504.9 scompiled p955.1 sagent p5037.5 sagent p9543.4 s
Per-run wall-clock around the replay / agent loop only.
Illustrative repeat-run model cost (measured model $/run × N; not new runs)
modeled runscompiled model APIagent model API
1$0$0.2716
100$0$27.16
1,000$0$272
10,000$0$2,716

Agent values multiply the measured per-run model charge ($0.2716) at list price; an introductory $2/$10 rate applies through 2026-08-31. The compiled column projects the sample's zero model calls only. It excludes authoring, maintenance, compute, storage, review, and exception-handling cost and assumes the measured task and rates remain unchanged. The one-time demonstration was about a minute of human demonstration.

Both arms drive the same vision-only interface (screenshots in; pixel clicks / typed text / scrolls out) against the same target; the agent is claude-sonnet-5 with the computer-use tool, prompted with user intent, not steps. Success is one arm-independent OCR check applied to both arms — neither arm's self-report is trusted. Anyone can rerun this deterministically — same orchestrator, same agent harness, same style of OCR success check, on a local app. figures: benchmark/results.json

Under a hostile theme swap (?drift=theme, which invalidates every recorded template crop), compiled still succeeded in 9.7 s with 8 self-heals; the agent succeeded in 87.4 s at $0.6319. This drift row is n=1 per arm — an existence result, not a rate.

Read before quoting these numbers

This comparison would be dishonest without the caveats it does not wave away. Pulled straight from the two BENCHMARK.md methodology sections:

Bottom line: on these measured trials, the compiled path removes the model from the loop — $0 and 1.8× faster on the real EMR, and a tighter, reproducible version of the same gap on MockMed. That is a cost/latency result on known tasks, disclosed with its limits — not a general capability claim.