Comparison
Reticle vs Playwright Test
These do different jobs. Playwright Test is your cross-browser release gate, written by people. Reticle is the agent's inner verify loop while it builds the app.
Playwright Test is the hand-authored end-to-end framework, you write .spec.ts files with locators and assertions and run them in CI across browsers as a release gate.
Where Reticle wins
- No hand-authored suite: Reticle records a flow once and replays it, you don't write or maintain .spec.ts files or fight selector churn.
- Determinism by construction: model-free replay is 0% flake over 8 identical runs; selector-based E2E suites are the textbook source of flake.
- Program-state, console-cardinality, render and blast-radius oracles the DOM never shows, a success end-condition compiles to signal | state | net{count} | console predicates, checked after settle.
- Built for the agent's inner loop: ~100 tokens per check, refuses destructive clicks, and returns the component + file:line so the agent edits the right file.
Where Playwright Test wins
- The release-gate E2E suite of record: a mature, hand-authored, reviewable test corpus is its whole job. Reticle is an inner-loop verifier, not your CI E2E gate.
- Cross-browser: the WebKit / Firefox / Chromium matrix.
- Any site, staging or prod, zero app cooperation. Reticle is dev-only, localhost-only, and instruments an app you own.
- Trusted native input, true pixels, network mocking and multi-tab, full browser-level scope.
- A huge ecosystem: fixtures, reporters, sharding, the trace viewer and deep CI integrations.
| Dimension | Reticle | Playwright Test |
|---|---|---|
| Authoring model | record / replay | hand-written .spec.ts |
| Who drives it | agent inner loop | humans + CI gate |
| Determinism | 0% flake (model-free) | selector-based |
| Re-run cost per flow | ~175 tok | full browser run |
| Program-state / signal oracle | ||
| Source localization (file:line) | ||
| Cross-browser engines | ||
| Runs on prod / 3rd-party site | ||
| Trusted input / pixels / mocking | ||
| Ecosystem (reporters, sharding) | minimal |
✓ yes · – no · text = how
The honest take
Playwright Test is your cross-browser release gate, written and owned by people. Reticle is the agent's cheap, deterministic verify-and-regress loop while it builds the app, complementary, not competitors.
Token figures use an o200k BPE proxy (within ~20% of Anthropic text tokens) except the live agent loop, which uses authoritative gpt-4o usage. That loop is one model, one turn budget, five scenarios, treat accuracy ordering as directional. The harness is committed; reproduce with `pnpm bench`.
Open source
See it on your own app.
The Perception SDK is open source. Paste one line into your agent and it sets itself up.