Slides: Building an Agentic Platform — Ben Kus, CTO Box
Source Video
Building an Agentic Platform — Ben Kus, CTO Box
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:
aWS
ee)
@®Graphite W Windsurf MongobB
Mdaily £3augment code WorkOS
OCR text:
MCE LAinylam sees OO SB see
6 Wi % Canon e If ( eurostar Fidelity (Hosinel) RK
0
Fortune 500 “HOVIA Osu FP GBH KOMATSE 8 es hereof
‘ ‘ Le) K Seen MICHAEL KORS — MorganStanley ri Nationwide nature Ge NEC
oe met Ts onaa tr Son 6, Seman ®© Brie SBR Seotlabonk
fexororase REMY COINTREAU Ye neo Motus =
L' ans gS Wie USAA" 4 . SS Ox
OCR text:
. ae é I Ph . yy 1.
Industry-leading Ini: anagement platform
Box as 1,500+ integrations Custom 2 geue
Web : APIs&SDKs = ;
_ Mowat ton, a! Slack DCopilot Zoom -
Mobile — = rd * P Integrations
a DT co gewcrnace nw GBD okta crowosrmme fF uicemens _¥ ”
a. F§ - ; = 9:
oa 58 @openAl Georvni aNTHROP\C watsonx Sy :
: "g *
Multi-doc quenes Transparency Unkmited end user Al
B Al eta ST ov lt mAs Ne] Mode!- agnostic Perera m ole lol 1g
. eH Files and fokders Collaboration E-signature average]
Ce aL) I CIeTe RG) Workflow Analytcs No-code apps
rae tecti Threat detection Ransomweare protection Classifications Pll scanning
d. bebiaiaie sad ton Data residency Audit trails TLIO LSS Encryption
Ure) compliance Centificaton Retention Neate
oo A Unimited storage Data ingestion Ove RU Liat)
and Global infrastructure High scalability High speed High availability
OCR text:
Box has fully integrated generative Al starting in 2023
Q&A across Data Al-powered
documents extraction workflows
g How can we speed up our product roadmap? 172 Contracts Loan Approved: “eee
1 Above $5000 90 vein 90 Days
Beemeiatentoris ceutean oh tesbrdgse Neweeronvonses @
ee sein Dy semana
Smmrnoenn’ O
OCR text:
Box has fully integrated generative Al starting in 2023
Data
2>xtraction
Cia i :
- J a Microsoft @yry{?
OCR text:
Metadata is the important structured data contained in
unstructured data
ao
peels ig
Ths batual Mat-Dieciowure Agreemend (this “Agreement”! 1s made and enteeed into 3s of the date last executed by
the Parties below ("Effective Oate’| by and betaees Bor. Inc and its athhiates (“Baw”) with oth es 2¢ 900 Jeter son
Ave. Redwood City. CA 94063 sn QE Participant’) with oft cs at QE SERIES
Bee any GRR erty the Partin individual a Party) Vitereas the Partaipant wishes tu explore a potential business
opportursty to purchase or hcente services from Bar and In connection with the oppartunity each Party may
Grctose to the othe+ cov 2: Qi rot ne ooscloung Pxty desires tne
recorving Pacty 10 treat at contider tial Purpose’) The Parties han agreed to du 40 wtyect Lo the terns and
CLASSIFICATION Cind tans as vet forth below
ED ~ 5-1-0 ero ots
by a Party Disclosing Pasty’: 10 the otter Party ( Receiving Party”) cer or after the Elective Date of Une
Agreement that are lin Laag-bie form and Labeled “conPiential’ or the lite (61-4 div coved or ity are wunmarired
BA CORA ened on ner tong 10 Dee CONG] a wsthen a fe atonudie bene from [he it ted Gat honure OF) Atormatica.
that 2 reasonable person knows oF should hive known to be contdential piven the ce cumslances surroundeng
Sixelowute The lollowiag formation shall be considered Cormidential ledor mation whether of nok muried oF
rent fied as wich any per wonaity BentiBable informahon Wxh as the names of either Party's customers, steatoen
mur heting plans 3nd product roadmaps, Source Code techie ab infrastructure secunty and complune
Gocumentaton hardwsee COnfiguralion deicounte and the terms of this Agreemen! Confidertea! lntormateon shad
at ime hate of Abt Cease Co mnt ludie at appt able inne matwin oe materials thet (a) were geareaty known to the:
a me public on the Effective Dale (bi become generally bryan to the pubia after the EHtectoe Date other than as a
~a result of the act of omnson of Ihe Rece ving Panty. 1c] were reghttully known to the Recewving Party preor torts
| ba receipt thareot from the Divckourg Party, d) are of were disc lated thy the Dru oveng Party gemerally without
oa revtrertion on disc lovee. [0] the Recesving Party lawfully recereed trom a third party without tht {cd party's
=
-)
OCR text:
Metadata extraction enables Al to understand
and process enterprise content effectively
Que wee e-- —— » eS
8 e Contract ate
eee einen NON DISCLOSURE AGREEMENT 108 2 ene °
cece wee vee Je
utomating wor! Vey cpaiicnngaennel , a a ~~ i °
os) fete eum nee cece a enero 5
understanding TUE aoe _ — °
ae eee
Access errrec Lesa irece tee maclerpesrrews ieee nenone Sopw aren Dete is
|
| :
1 ry
ed ]
aws
OCR text:
Before GenAl, extracting data was
too hard for most enterprises
Ny
ML-based Didn’t work well for Not responsive to
techniques too free form text data data format changes
expensive to develop
a : 1
| a Microsoft = rnal)®
al
OCR text:
Building generative Al data extraction
aaiech GNA =IecStO1|
Performed well
ae
- a nnn Flexible, based on reasoning
™ [ Cree iit wera We. Ges
Paroe ty LLM ee
; i caine a Eerie Rss rellive
wae “vee Al model
NN Sg . porn s
Seed a“ Extracted data
yeaa) ed Sete) Required ML-based pre-
. aa processing for OCR
ere zals
mu Microsoft =U
_
an
OCR text:
Building generative Al data extraction
Challenges
Started getting longer /
more complex docs
5
ae ae oe Eur
ae * aay | aa ; lose valuable info
aa ee a deren ee ee —*
ern ere et
a aes
LN O° ae
bad ill PY Extracted data
ad : id iahiaale Spertt rore srd more t Te ta
2. step aera nn he
1
PaaS CE eke rece
Deter aA TS
sence iain
rs
(ie .
a | mE corto ES UOU
OCR text:
Building generative Al data extraction
Challenges
feed Bal as Site Le ie
Lp feids ints groups
Ky
Started getting longer /
Needes a qua:ty ind cater morse complex docs
oe Poo Teil ar RO Eka Meet od
a P
- Tre OCR would sometimes
- r
ae a LLM a Lert E UK| sl Manto)
—_
de Selanne aT ACTRESS m TLE)
ype me Orcas amt aad
x C7 :
Pret a ia
jc MIDENSA Oats MAST ee ee eer
eed as: ae : se determine that sometimes it
: aaa be was wrong on its first pass
a]
ray PT OR Ce NE er acer ome a
PC rns]
eee
eo
eel = A
Po ;
A fai]
— a Microsoft
. a ie
OCR text:
Bringing agentic technology to the Box platform
Al model a as
e P Ye Objectives
Instructions a> , : -
er eee ae , Capabilities and tools
Secure access to your content
2a
7 : |
. “s oa ~ an
OCR text:
Al Agents driven by advanced agentic reasoning framework
kL) ;
al / comprehend ‘\
cn Pe) ee a ee)
and iterate execute
eae ast aie lee aoe ae ed
sdentts reqused foots, Jota Passe ratroctocs, constacte. 7 , ; , - .
Oe Rene ea) fener rang sed Bieta rra) en - ne
ped ata Bee eat eae Rea
eee eeeeeacnennnne
foekt st armed ete tora ted
nat oer Ce ed
a
; mV (ai)
ae | a Microsoft §= SUDO
. ,~ 3
OCR text:
Improving Box Q&A - Agentic RAG
Answer Critique
question answer
Question —(8} EEC —> Answer
. . :
As aia@edecig
[o[reselelany
-3%-
.
5 A >i)
- | a Microsoft §=GDCU
= ys i
OCR text:
Ich-scto] am ict- 1a aT cre
Build agentic architecture early!
Cleaner platform Ease to evolve Improved engineering
abstraction quickly PN Ror celae
Separation of high-scale Easily adapt as new Helps engineers think
infra work vs. improving techniques and Al agentically (state diagrams,
specific agent actions models become Tools, A2A, ...) as part of
available overall ecosystem
== = ,
co? | nicki ES 0000 Oh
7 dl
OCR text:
Thank you!
ee
a
bY —
a Microsoft @yry{?
OCR text:
Thank youl |
Ye
| World boyy
Ve 7
OCR text:
Thank you!
rd
Loved a
: 7
Worlds Fair
| y
S| a - Dead
ie —,_ -
oe
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.