Slides: Browser Agents Don't Need Better Models. They Need Better Eyes. - Kushan Raj, ARK
Source Video
Browser Agents Don't Need Better Models. They Need Better Eyes. - Kushan Raj, ARK
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
- No individual scheduled session mapping has been assigned yet; treat this as an event livestream deck.
Extracted Slides
OCR text:
—=s s ;
BC RR OME ies ta ogee ieee aS ol
Cookie Consent
Th ta pon Te EIR ting deats! °
dives newetsennntree sccast oo De
= ae ;
ose
3 wr pet woerud eerseaten stance
Secton 2 Unports Alert! 3
Thy va pope; MetApe The coe tert
_. ee ie ,
wee E : 3 *. _
id fab
won ED
® i gS ee la
*This Claude with Chrome
trying to navigate a browser challenge
OCR text:
demo-video
Recents
shared
YaiusaTmovie
Claude-browser chalenge
caude-treking
Todayat3:14PM
388.7B QTmove
Today 342PM
457N8 OTmovle
A.Aoplcations
D
feedyck-action-og.md
1on-
Tody337PM
kMre
D
leedack-bcker-recoverymd
Today336PM
-Mordo_letfe
Downloads
D
feedback-mutations.md
Tody336PM
Todayat329PM
Desktop
csour-agent
odel
cerebras/gpt-oss-12eb
-headed
2273 OTmovie
otrect
Todey95PM
. qTmove
D
unoog
avigatin
Todryar403PM
197.8MBQTmov
Today12:32AM
618KB PNGimage
D
paword.md
Today335PM
icloud Dr
D
tarido..textfile
D
v6-sep1-overays-blocking.md
2.KE
rkdo_etfle
kushanj
D
v5-step9-deep-blockinged5173/
Y337PM
1K8
arkdo_et fle
rkdo_tetfle
usans Mac（TING）
Dv5-p-d23
mtntcaptt
EKB
266Rs
2K8
arido...ete
MacintoshHO
4msfinalize=e.3ms
TIMING)
j5_collect=22.0m
python
MacintosHO
Extracted:4containers,
Loom0.3410-TATION-0B5ERVER]v21nsta1led(maxA1d:5)
1controls,0fors
Ardrop
[FAST_CLICK aid-5] d1spatchat（64#,456)
[TIMING]dom_stability=555nshtml_capture=4s
auto-click1laiu-5->ciickedaidsaid-5
Network
extractic
TAL-66985
Bin
[TDMING]
j_collect=B6.8mspYbon_bu1Ld12.6ms
n=9.1esfinalsze=1.lms
btocl
Auto-clickedLtrivialactions
[AuTD-NAv]->http://1ocalhost:5173/step17vers1on-1
important
ITIRINGarkdoW
ing rav executer(modelrcerebras/gpt-oss-12b）.
render=lns
OFINACCT
Omkt2
AI Tags.
OCR text:
. ; ;
re ae ; PvE ear eneE
. a, an a Section 43
ia ote oa mua
. ee a
; . Oo! se :
¢. | | | | ; 7 as
] . — a Fs wp, aE Deawress Hl, . . San
a , ns , Section 81
|
80OB9 y,ao » @O6§@8@--4~467 « Q@rOQ@Q @Qe.
OCR text:
Peete eed . a a ei a “2 O« Go + < 2 4 Hy a
ee. 8 on
ers = : :
omen Pe
:
re a © i tea
o_—_2 cs
= =
Ge
Protecting my
info lol 90
4
_ 2 S100 Cee a
3OC8 yao » O84G: -4¢67 Q@rOG ®_”
OCR text:
Pee tet a 4 cr eens a Of OC a. co Py : “
en)
| Overeant Sevtreer Biba) Cc
area
1, ai ctaaeo cited: Sod Cbd) 06 WIRE MecrertOod mEdcerteb ead Dona gin Seth vet &
Peet of rere AeTieer
2. OR UmerBAd nuts, Dery wols, mA Uo mar St am ou Mchabd ay Dobe! Huot oA page
&
undood os} Sabah, abab3
© -
Het ree Od bee bce i Ost dont Odie hota advore bone ngt Ore 3 owed ey upto th days Sher kath
eer paane Ly
ad, aenciencs states Rano Awproe tere
Bd cock, to . pmdirectt, evo tnd» tt eae a Sa
rs
etd, oe
WO upload por
Rderetity Pod: ard
Address PA:
ts The pared
pet we, Cae can Con ME WLAN Om the waraoerore
aervet rina Yas weg COA LE yr mats
Z a = | Sess
oe Be ag ee
a Ey ia a at
ee + 2 ie
4
(Rolo k: Maa ee ee? ; ee © OC
OCR text:
1I1EOTmoe
3667me
1Tmoe
AaaMsreso.edfle
Google AI Studio
3x3Markdo_et fle
Untited prompt/
1,813toke
2271QTmole
20S OTmoe
Playground
AOrmo
ExploreGeminimodels
13Gmge
XKR
Markdot I
IKIL
Morkdo_et fle
Mekd_tet fe
Featured
Code,Reasoning.and Chat
EMrkdo_tecfle
Markdotent fle
3Fai
png
sst
Docu
aid-47z435[ar
ed][biockedbyaid-75.aid-58.aid-43.aid-49]
"Buttant"
aid-53z439ar
ted][blockedby ad-75,aid-S8]
"Try Ths!"
aid-85[an
RSerh
"Moving!"
0What'snew
GetAPI key
BTools
0URLCX
Run
OCR text:
es an This browser challenge is a
Fairly simple webpage
But generally across webpage
The markdown has
~20x token compression w.r.t DOM
So a
OCR text:
3. Dismiss Hit vs Miss — Fake Close Buttons
4. Unblocked Signal — What Becomes Accessible
5. Failed Action > Diagnosis > Recovery
» ry a
-
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.