LAST MILE Where agents actually land

Scans pass. Agents fail.

Last Mile runs real browser agents through real customer journeys, then verifies the outcome with a deterministic harness. The Index ranks completed tasks, not readiness theater.

View the Index Inspect methodology
Measured airlines
7
Task completion
9%
Honest gates
3
Riyadh to London A restrained route arc with a small plane marking a supported Friction Airways route.
RUH Riyadh LHR London

Route 1 of 8: King Khalid International Airport to London Heathrow Airport

Three gates. One completed journey.

A site can be discovered and understood and still fail before the task is complete. Last Mile tests every gate with a real browser agent.

  1. Discoverability

    Can agents find you?

    Surfaced and reached

  2. Understanding

    Can agents read you?

    Structured path · legible inputs and errors

  3. Interaction

    Can agents operate you?

    Unpaid task completion · failure recovery

Every gate receives one verified outcome

  • Fails
  • Struggles
  • Completes

Ranked by completed tasks, with the static scan kept in view.

The public leaderboard preserves the inversion that matters: sites can present clean static signals and still fail the browser-agent run.

01 Riyadh Air 1 / 1 pass L— 02 airBaltic 0 / 1 fail L1 03 Delta 0 / 2 fail L0 04 easyJet 0 / 3 fail L—

Open the full Index →

A real agent drives. The harness decides.

Task Real customer journey

Lowest fare, check-in, policy extraction, or another measurable task with a known success check.

 Run Browser agent attempt

The model drives the browser while Tarmac records friction signals and evidence.

 Verify Deterministic outcome

The agent never grades itself. The harness returns pass, fail, or blocked access.

Blocked access walls are separated from failures so the Index stays defensible. Read the scoring spine.

Friction Airways proves the levers move the score.

Real airlines show where agents break. Friction Airways shows why: same airline, same tasks, same scorer, three readiness profiles. Flip the levers and the Index swings from zero to perfect.

Open the Playground
Hostile0%
Median71%
Ready100%

Every failing signal has a fix.

The recommendation ladder turns each friction signal into a tiered fix — the playbooks that move a site up the maturity model.

Browse the playbooks →

The report is generated from the same run records.

The public site, PDF report, and methodology pages all point back to the same task records and scorer. No parallel narrative, no self-assessment.

Every report in one place →

Get your site to the last mile.

Benchmarking, per-site diagnosis, and the fix roadmap — the engagement that takes a site from failing the agent run to completing it.

See what we offer