Markdown source

Slides: The Missing Layer After Launch - Raphael Kalandadze, Wandero AI

Source Video

The Missing Layer After Launch - Raphael Kalandadze, Wandero AI

Relationship To World's Fair 2026

These slides are extracted from a public AI Engineer YouTube video connected to World's Fair 2026. Speaker-matched clips are supporting context unless later confirmed as exact session recordings; official livestream recordings are day-level/event-level source material.

Related Scheduled Sessions

Extracted Slides

slide-001.jpg

Slide text:

The Missing Layer After Launch

Everything important starts after you ship.

slide-002.jpg

Slide text:

We shipped it in 3 weeks. 3 wks TO FIRST PRODUCT ~300K LINES ~$35K SPENT

slide-003.jpg

Slide text:

Monitor it? Understand it? Improve it? Find the holes?

slide-004.jpg

Slide text:

02: InviSible failure

The failure hides itself.

Claude marked features "complete" - without checking they actualty worked. the opposite. Agents reported success while the system state said

*EFfeCtIvE HaANESSeS FOr lOng-RUnnInG AGentS.' AnthROpIC NortheaStern ReO-teaM

Nothing crashes. Nothing turns red. Nothing knows.

slide-005.jpg

Slide text:

One task.

Hundreds of tools.

It writes code, runs the terminal, calls other companies' services.

Every tool is a new, quiet way to fail.

slide-006.jpg

Slide text:

The ciosed toop

Production Sessions traces Loys: & Trajectary. code pa.th sis

Next'fix pommoa revev.+ merye: ·Humon PR rever Pull cequest (lognontor)

terate wtl reay

The agents watch and draft. The human decides.

slide-007.jpg

Slide text:

Four operating agents.

01 log-monitor — detect & fix fast (reactive)

02 PR-review — gate the fix (reactive)

03 session-analyzer — feel the health (reactive)

04 QA / computer-use — go test it (proactive • roadmap)

slide-008.jpg

Slide text:

AGENT O1·LOG-MONITOR

what t con read rnonitor agent

sioy wenpod DETICT Rhte os yorebas

hpet njecires codelase ASALYZE corerepo red cole Pull rervest

Dntains redu) Troves DecrDe does thevserend stuet? deureton eilente meroa He labes

FEX intemto penRe

slackalert

seventy

slide-009.jpg

Slide text:

LOG-MONITOR-SLACK

Ourbug~itopensaPR andtellsme

Outside billing issue -just a heads-up, no PR

LogMoanitorfotlow

Ervronrreet Period 210206UC)

PToviderblnghw-RfCURRt D.mot roohrd:LPxv402(ededmmm(redts'101:4sUTC

utt rraf ar resolvedT a (an bta- thut n rrw contas tn Th /5orgsover48h

Next strpvvoy Deturubsru 1uruouf pun top cp xtiusy

It alerts only when it matters - and knows the difference.

slide-010.jpg

Slide text:

PA-Aeview - the root.cause fix

Verdict: ready - root cause, with a test that reproduces it.

slide-011.jpg

Slide text:

AGent o3 · session-analyzer

zer

tess'on s.lc'! Sessions. malyse EvERY tert'on orcoe mb cost: ssucs Fcel the pulse

Kest'os s-st?e 77 / 100 jealtr score

tels'os x-la02 tic sane issue aorors say Ctons = Ose pattem

Rrior hit ~95% on the model, then got to 99% by building this kind of loop over production.

DPHER LOVEJOY: ANTERIOR

slide-012.jpg

Slide text:

GB AGENT O3 · SESSION.ANALYIEA

sesstor!

settons · walyte eveRy sets'on mtcoeraty coit· 'ver fec tre plre

rerre rutle 2·callepre sto ptters 77 / 100 kealth rare Tll, wcter·'cort trerd

tcrton s-?a0x rerroin,orE ptcol

Anterior hit -95% on the model, then got to 99% by building thls kind of loop over production.

ChAISToPhER LOVEJOY - ANTeAIOA

slide-013.jpg

Slide text:

SainnsAnibes 口 PoiringeScnp

$216.67 7.7 84.5% 15.5% 0% 1

BooreTand SaisensperDeyog.e Cost jerDey

30

AmiyMs

OegTng Atons

noboopt Agetaerpnded erBecgPttis

OrgariSMiOA TeEd Cod

67 ACostbAC N2.8r

Norwng Tri Ch 18

WrieVeey Fan

Wlerr Jxeroen 3.8

y

slide-014.jpg

Slide text:

Seheygant lyons RT

Crpanitao

Orginiation TolaiCom

74 AivaCtC 7

24

HepinguitorTrn

Wia hgiiny

QoigaeuteTund

Celirs Shy Touns

CppeAnan FDestiNMe 5.4 M 4.00 $5008 MsChu e Maint Cver Swtimes O

rotgil Cueta Bocw

Atooifen

3,320 8% 73

slide-015.jpg

Slide text:

177m 2

59ve9 (12

Session s-096

OVKORG S00CXSS·5:/0

Ceet 500

Whatthecustocrer wanted

What happnid

Issuesfours

MOMC2

Tepls used

slide-016.jpg

Slide text:

Go looking for what's about to break.

Roadmap / rebuilding — drives the live product like a user.

A validator that drives the live app through computer use.

slide-017.jpg

Slide text:

AGENT 04 • QA / COMPUTER-USE — LIVE

It drives the real product like a user — and finds the bug before a customer does.

slide-018.jpg

Slide text:

THE ENABLER

Don't give them a keyhole. Give them the whole room.

Give them all of it — or they're guessing.

slide-019.jpg

Slide text:

THE BIGGER PATTERN

The harness that watches itself is the moat.

The model is the part you swap. The loop is the part you own.

Classification audit: raw/sources/slide-ai-classification/slides/kZsf_Sfm7RU/audit.json

Slide-Derived Subjects To Review

Subject extraction uses video title, related session titles/descriptions, transcript context, and OCR text when available. OCR is best-effort and should be reviewed against the embedded slide images.