// verdict · checked 2026-08-11 · Comet Assistant (Nov 2025 rebuild)
Comet
Perplexity
A browser with initiative. Sometimes the initiative is buying the wrong thing.
- CAPABILITY
- 4/5 AGENT1 of the five could not be decided either way
- PROOF
- UNVERIFIEDWe could not confirm this either way. Not a claim.
- TRACE ACCESS
- OPERATOR-OWNEDWhoever runs it holds the record; nothing is published
- LEASH
- Not measuredno execution record scored — measure it yourself
| LOOPSruns until doneyes | Vendor says it "can work longer on more complex jobs" and keeps "not giving up easily" on tasks needing "many steps right, and all in the right order" — step count varies per task, not a fixed script. |
| CHOOSESpicks its own toolsyes | It picks each next click/scroll from live page state and which sites/connectors to use at runtime; vendor shows step-by-step reasoning and it "automatically browses when it detects the chance to be useful". |
| ACTSchanges the worldyes | It drives the real browser — "clicks, types, submits and autofills" — and enterprise admins must gate exactly those click/navigate/fill-form powers, so state changes outside the chat. |
| RECOVERSfixes its own errorsuncertain | Vendor says only that it "works harder, not giving up easily" on many-step tasks; no re-plan-on-failure mechanism is named, and hands-on traces show it reporting failure rather than rerouting. |
| UNSUPERVISEDnobody watchingyes | Enterprise "Always Allow" lets Comet act "without having to confirm", and users can let it browse automatically; approval is once per task, with pauses only for a narrow sensitive set. |
Capability decisions combine documented behaviour and execution evidence. The proof label on each row shows which of the two decided it, and Leash is reported separately, only when it was measured from an execution record. Primary source: www.perplexity.ai. Tested against: Comet Assistant (Nov 2025 rebuild). Our confidence in this dossier: 6/10.
Can you see it work?
the weakest of the five grades aboveUNVERIFIED — We could not confirm this either way. Not a claim.. Trace access: OPERATOR-OWNED — Whoever runs it holds the record; nothing is published.
Steps are readable live in the Assistant sidecar; agent steps also land in Enterprise audit logs.
Where we looked: Perplexity docs/blog (403, read via search snippets), TestingCatalog + Seraphic/Lifehacker hands-on reviews, Zenity Labs reverse-engineering of Comet's SSE/WebSocket step events · see the evidence → · re-checked by a second pass built to overturn it
- 2026-08-113/5 WRAPPER WITH AMBITION → 4/5 AGENTPre-launch adjudication: every criterion the verdict depends on was re-checked by hand against the published rules.editorial correction — no case was filed
Badges
earned, never for sale
These render from the live dataset, so a badge cannot outlive the verdict it claims. If the verdict changes the badge changes with it, and an unearned one returns 409 rather than an image. No sponsor can buy one, at any price.
Think this is wrong?
bring evidence, not opinionName the criterion and link a trace, a doc, or a recorded run that shows the behaviour. A verdict changed by evidence is the best thing that can happen to this site, and the change goes on the public record with your reason attached. Vendors are welcome, and a vendor's own filing is labelled as such rather than buried.