Andrew Warner's REA episode is a tour of other creators' experiments, not one controlled benchmark. He watches agents investigate QuickTime Player, TickTick, Wispr Flow, a CapCut-inspired editor and Loom, then pauses on a different use: finding bugs in an app the tester owns. The useful question is not whether an AI can produce a convincing screenshot. It is which behavior the new tool really reproduces, what remains missing, and whether the inspection is authorized.
REA can lower the effort of inspecting how some applications behave. These demos show promising partial implementations and one reported QA investigation, not exact clones or proof that every app can be reproduced in an hour. The clearest near-term use is an evidence-led migration or test of software you own or have permission to examine.
Watch Andrew's REA Use-Case Tour
Credit: The reactions and selections are Andrew Warner's. The underlying builds belong to the creators shown in his episode. The observations below come from Andrew's supplied transcript and the three original videos linked later; no project files or independent test logs were provided for the other demos.
What REA Does, and What It Does Not
Reverse Engineer Anything exposes investigation workflows to coding agents through a CLI and MCP server. The project's setup guide currently shows npx rea-agents@latest setup; review the proposed configuration changes and the software it will run before approving them. Depending on the target, analysis may involve established tools such as Ghidra, Hopper or IDA, along with other ways to inspect web and desktop applications (03:27).
REA is not a new decompiler that magically returns an app's original source. It helps an agent collect evidence, infer behavior and write a separate implementation. Andrew says it has made “every software open source” while reacting to the QuickTime demo (01:39). That is a metaphor, not a change to any application's source license or the owner's rights. Even a working reconstruction needs feature, security, accessibility and data-integrity tests.
QuickTime and TickTick: Useful Features, Uneven Fidelity
QuickTime to Tempo: playback speed with more control
In Mark Kashef's original demo, the new Mac player, Tempo, adds a more granular speed control, waveform, pitch adjustment and export options to a QuickTime-like workflow. Mark estimates a first version in roughly 20–30 minutes and says most of the build came from one pass; those are his estimates, not timed results from Andrew's test (01:39). This is a plausible narrow product: a player designed around a particular review task. It does not establish parity with QuickTime's full media handling, codecs or system integration.
TickTick to a task manager: the missing details matter
The task-manager creator shows lists, completed items, tags and a calendar-like view. Andrew is impressed that a calendar appears in the new app even though the creator's free TickTick account shows an upgrade prompt (06:54). That is an observation about this demo, not a recommendation to circumvent a subscription or access control. The same walkthrough shows that multiline notes fail and overdue colors differ from TickTick (07:39). The creator's “about 80%” characterization is an impression without a published feature checklist; it should not be read as a measured coverage score.
Dictation and Bug Testing: Two Different Jobs
Wispr Flow-style dictation
Eric Tech's source video shows a small voice-to-text desktop app with a shortcut, transcript history and English/Chinese input. Eric says transcription runs locally and Andrew highlights the possible privacy and latency benefit (09:54). The screen recording demonstrates selected interactions; it does not independently verify that audio never leaves the machine. Anyone making that privacy claim for a client should inspect network activity, model dependencies and storage behavior before publishing it.
Read-only bug investigation on your own product
The AI LABS segment switches from cloning to testing: the agent is asked to explore the team's website without reading its source or changing it. The creator reports a mismatch between displayed and charged prices for two accounts, plus negative-seat and plan-limit issues (13:39). These are reported findings from that video; Andrew does not reproduce them independently. The lesson is to compare visible product promises with real transaction outcomes. For a web app, ordinary black-box browser tests may be enough; REA is worth adding only if deeper inspection answers a specific unanswered question. Use test accounts and non-production payment flows when probing billing.
A Narrow Video Editor and a Local Recorder
CapCut-inspired, not CapCut-complete
The video-editor creator reduces the request to a web app with import, timeline, splitting, auto captions and export. The clip shows a split and caption generation; Andrew points out that the result looks like a generic editor rather than a close CapCut replica (16:03). That narrower scope makes the prototype easier to understand, but the episode does not test long projects, codec coverage, export fidelity or the wider editing suite.
Loom-like recording, with local files
Nate Herk's source demo shows a recorder with a countdown and a movable, resizable camera bubble. Nate emphasizes that the interface is different and that recordings stay in a local folder rather than Loom's cloud service (19:48). That is a meaningful product distinction, not an exact replacement for hosted sharing, collaboration, playback analytics or administration.
Watch the Original Creator Demos
Andrew reacts to excerpts. These three verified originals give more context for the QuickTime, dictation and recorder examples; the TickTick, bug-testing and editor clips are available at their timestamps in Andrew's episode.
Inspiration Is Not Permission
Andrew asks where inspiration becomes copying (19:21). Changing a few features or visual details does not, by itself, settle that question. REA's own responsible-use statement says the tool is for lawful analysis and rejects DRM, license or access-control circumvention, unauthorized access, and copying protected code or assets. Local analysis does not create authorization. Review the target's terms, your permission and the intended use before inspecting or shipping anything based on it.
For an authorized project, write down the exact behavior to preserve, collect evidence with its source and uncertainty, build the smallest implementation, and run acceptance tests on real inputs. Distinguish a working prototype from a secure, maintainable replacement. If the goal is just to check a website, first ask whether standard browser automation can answer the question with less access and fewer dependencies.
Jump to a Segment
| Time | Topic |
|---|---|
| 00:00 | What REA claims to do |
| 01:12 | Setup and agent integration |
| 01:39 | QuickTime-inspired Tempo |
| 03:27 | Tools behind REA |
| 04:30 | TickTick task-manager demo |
| 06:54 | Calendar and paid-feature question |
| 07:39 | Missing notes and UI details |
| 09:54 | Wispr Flow-style dictation |
| 13:39 | Testing your own app for bugs |
| 16:03 | Narrow CapCut-inspired editor |
| 19:21 | Inspiration versus copying |
| 19:48 | Loom-style local recorder |
| 21:00 | Final take |
A Better Offer Than “We Clone Any App”
Offer an owner a fixed-scope behavior and migration audit: document one critical workflow, identify failures and produce a tested prototype only when authorized. The output is a decision the buyer can use even if they choose not to rebuild. This prompt starts with permission and evidence, not a decompiler.
Test an owned workflow
Scope, permission and acceptance tests first.
Sources and Limits
Primary episode: Andrew Warner's REA use-case tour and the user-supplied transcript. Project: official REA repository, setup and showcases, and responsible-use statement. Original demos verified by video title and creator: Mark Kashef, Eric Tech and Nate Herk. The remaining examples are discussed from Andrew's episode; timings, feature coverage, billing defects and privacy claims are not independently benchmarked here. This is practical analysis, not legal advice.