Launch report
State of the Agent Web Report
CrawlDex's live evidence report on where AI agents can complete website tasks, where they need a human nearby, and which blockers appear most often.
Publication status
Human reviewed- 8,401
- measured rows
- 2,190
- websites
- 21%
- need a person nearby
- 68
- mean AES
Last verified Jul 8, 2026. Based on 9,590 observations.
Thesis
The agent web is already useful, but not yet autonomous.
CrawlDex measures whether agents can complete real website tasks, not whether a site sounds automation-friendly. The current evidence says agents can navigate many public flows, but the most valuable tasks still depend on authentication, confirmation, payment approval, or support handoff.
The practical launch headline is simple: 21% of measured site-task rows need a person nearby. That is not a failure of agents; it is a map of where user-present checkpoints, clearer confirmations, and better task APIs would make delegated browsing safer.
Category league table
Where agents work best today
Categories are sorted by mean AES from the same public board used by task, ranking, and site pages.
| Category | Mean AES | Rows | Need a person nearby | Top blocker |
|---|---|---|---|---|
| Government And Civic | 82 | 107 | 29% | public url verified |
| Jobs And Careers | 82 | 98 | 58% | login walls |
| Developer, SaaS, And API | 81 | 108 | 70% | login walls |
| Education | 81 | 59 | 56% | public url verified |
| Creator, Social, And Media | 79 | 98 | 83% | login walls |
| Business Operations | 78 | 101 | 84% | login walls |
| Healthcare Admin | 78 | 101 | 59% | login walls |
| Future Agent Tasks | 76 | 99 | 70% | login walls |
| Legal And Admin | 76 | 50 | 54% | public url verified |
| Customer Support | 75 | 60 | 100% | required human handoff |
| Travel | 71 | 58 | 95% | login walls |
| Real Estate And Moving | 68 | 79 | 90% | public url verified |
| Finance And Bills | 66 | 34 | 100% | login walls |
| Insurance | 65 | 33 | 100% | required human handoff |
| Subscriptions | 63 | 104 | 52% | public url verified |
| Food And Local Delivery | 59 | 63 | 100% | login walls |
| Commerce | 56 | 60 | 62% | public url verified |
Blocker prevalence
What stops agents most often
Blocker counts are affected site-task rows, so one row can contribute to more than one blocker.
public_url_verified
public url verified
The task repeatedly runs into public url verified, which changes the expected agent path or prevents a clear final state.
final_action_requires_human
final action requires human
The task repeatedly runs into final action requires human, which changes the expected agent path or prevents a clear final state.
login_required_for_final_action
login required for final action
The task repeatedly runs into login required for final action, which changes the expected agent path or prevents a clear final state.
payment_or_fee_context
payment or fee context
The task repeatedly runs into payment or fee context, which changes the expected agent path or prevents a clear final state.
login_required
login walls
The site requires account authentication before the task can be measured or completed.
site_access_denied
site access denied
The task repeatedly runs into site access denied, which changes the expected agent path or prevents a clear final state.
http_403
http 403
The task repeatedly runs into http 403, which changes the expected agent path or prevents a clear final state.
sparse_public_text
sparse public text
The task repeatedly runs into sparse public text, which changes the expected agent path or prevents a clear final state.
Evidence tables
The current extremes
These tables link to route-level evidence pages so every claim is inspectable.
Most agent-friendly rows
Hardest rows for agents
Most-checked tasks
Priority site-task evidence to inspect next
These pages are selected from evidence quality, consumer intent, and Search Console signals when available.