Task intent
01Locate a team plan and API setup guide
starting goal AGENT JOURNEY RELIABILITY / IN DEVELOPMENT
Make every agent journey visible.
Know where agents get through, where they get stuck, and what your next release changed. AgentLens is being built to turn agent activity into evidence your team can act on.
Clear activity. Verifiable outcomes. No invented agent thoughts.
···
✳ NORTHSTARProduct Pricing DocsN
WORKSPACE PLANS
Good work starts
with a clear view.
Tools for teams that move with intention.
TEAM$24 / seat
✓STUDIO$58 / seat
→✳Co-workexample profile
OBSERVED ACTIVITY REPLAY
Web search
02Found the pricing and API guide paths
search resultsFetch + crawl
03Read route hints, then opened key pages
/pricing · /docsVerify
04Found the API quickstart link
evidence captured✓Goal reached4 steps · 8.4s · example trace↗
Task verifiedwith page evidence✓
Release diffjourney still passes+1
01 / SAMPLE JOURNEYExplore the interface
BUILT FOR TEAMS SHIPPING THE AGENTIC WEBMAKE THE INVISIBLE ACTIONABLE ✳
DESIGNED AROUND
THE AGENTS PEOPLE USE
THE AGENTS PEOPLE USE
✳ THE AGENT EXPERIENCE LAYER
Not another score.
A better view of the work.
A site can look ready on paper and still leave an agent circling the wrong page. AgentLens is being shaped around real tasks: what an agent tried, what the website returned, and whether the intended result was verified.
See how a run reads01⌕
intentrouteanswer
Start with the task
Write the job a person actually needs done. Track the path toward that outcome across pages and tools.
TASK-BASED02✣
URL /pricing/teamCHECK plan details foundTRACE 4 observable steps
Show the evidence
Put observed actions beside the URLs, responses, and checks that support the outcome.
SOURCE-LINKED03↗
Catch the regression
Compare runs after a change and pinpoint the step where a previously successful journey diverged.
RELEASE-AWARE✳ INSIDE THE LENS
Follow the trail.
Find the turning point.
Replay a sample journey and see how task intent, page actions, and verification fit together. All values are illustrative—this demo does not browse a live site.
✳Journey playback/northstar.example
DEMO DATA
MISSION READ-ONLY EXAMPLE
“Find the team plan, then locate the API quickstart.”
JOURNEY MAP6 / 6 STEPS
01IntentGoal defined
02Web searchPaths found
03FetchRoutes read
04CrawlLinks mapped
05BrowsePage opened
06VerifyEvidence ready
ObservedCurrent stepNot in this sample
RUN TRACE
6 EVENTSObserved activity
Intent received
GOALFind the team plan and API quickstart.
Goal · read-only taskWeb search
SEARCHSearch results point to pricing and docs.
Search · 2 relevant pathsFetch route hints
FETCHRobots and sitemap paths were read.
GET /robots.txt · 200 OKCrawl key pages
CRAWLPricing and API guide links were discovered.
/pricing · /docs · 3 linksBrowse team plan
BROWSEPlan details match the stated team size.
/plans/team · page evidenceOutcome verified
PROOFAPI quickstart link found on the docs page.
Assertion · destination presenti Activity summaries describe visible actions and evidence, not private chain-of-thought.
06OBSERVED STEPS
8.4sDURATION
2,184TOKENS sample value
VerifiedSAMPLE OUTCOME
SIMULATED EVENT STREAMNO REAL REQUESTS · NO CUSTOMER DATA
✳
When connected to an agent runner, this is the kind of trail AgentLens aims to make inspectable.
✳ A CLEARER KIND OF OBSERVABILITY
Show what happened.
Be honest about what didn’t.
Agent behavior is hard to interpret. AgentLens should make the evidence legible without presenting guesses as facts.
OBSERVABLE
What the run can prove
- ✓ Pages and links visited
- ✓ Tool calls and returned status
- ✓ Checks that passed or failed
- ✓ Timing and provider-reported usage
CLEAR LIMITS
What stays a boundary
- ↗ No invented private thoughts
- ↗ No “success” without a check
- ↗ No live claims from demo data
- ↗ No credentials in this prototype
✳
“The goal isn’t to watch a cursor move. It’s to know whether the work got done—and why.”
THE AGENTLENS PRINCIPLE✳ THE ROAD AHEAD
From first scan
to release confidence.
AgentLens is an early product direction. The first version is focused on making a small number of valuable, read-only journeys easy to inspect and repeat.
⌕domain→signals
Choose a journey
Describe a task and the outcome a successful run should prove.
PLANNED✳agent→evidence
Observe the path
Replay the observable steps through the site and linked tools.
PLANNED↗release A→release B
Compare releases
Spot the change that shifted a journey, with a report your team can use.
PLANNED YOUR SITE. THEIR NEXT MOVE.Make room for
Make room for
the agentic web.
Help shape an observability tool that makes agent journeys easier to understand, test, and improve.