Slides: The Production AI Playbook: Deploying Agents at Enterprise Scale — Sandipan Bhaumik, Databricks
Source Video
The Production AI Playbook: Deploying Agents at Enterprise Scale — Sandipan Bhaumik, Databricks
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:
ary Engineer The Producti
a aeRerea ee
| AI Playbook,
7 J Learnings from deploying agents in enterprises
!
r ore
La
OCR text:
THEPROBLEM
Thepatternyou alreadyknow
AIE
Weeks1-4
Weeks4-8
Weeks8-12
Week14
Month6
Pickmodels.
Build features.
Demotoleaders.
"WhyisAl
SSin projects
Looksgreat.
Sign-off.Ship.
bs-ing us?"
failed lastyear.
Soundfamiliar?You'renotherebecauseyouhaven'tseen this.You'reherebecauseyouwanttostopit.
AlEngin
Moen
AlEnginee
EngineeringthefutureofAl
OCR text:
e
The AT is the easy part
gk ¥ You can't debug what you can't see. 01
* ~ ol Saat ate er
mae
* ad
oar ied tee U
You can't improve what you can't measure. 02
PTT toe td
OECD Ma te here xee Manan ete ny
Governonce gap
AlEngin
eee | PVE Sales |
’ aU cela
OCR text:
The Five Pillars of Production AI
as rep ey 0303 oye O05
mn . Evaluation fel estuarleliiiag Data Foundation Orchestration Governance
Each pillor enadles the next. Skip one and the whole thing breaks in production.
[sean
»
Lael Engineering the future of Al
OCR text:
PILLAR 02 &
o Define success with numbers
Not ‘accurate -- BUL 87%. en fer disputes, «2%, false positives, 60%, detection
Evaluation date Numbers make model selection obvious
ca
ate hws
oa rd
Build test cases from real data
ae] B
* i
a ra . ‘ Pull 200+ real conversations from: production fogs Ancnymise them Every case
* You reva | Tein kea by Reeds pass/faicrtena — nat locks good
your specification.
Without it, you're
guessing. B Wire automated grading
Evaluston pipeline that runs on every code change Dashboard showing metrics
You missure success before you tive anything fo messure
ele aace tia)
7
t
ae ,
ra | Al Engineer |
noe SUL tela
OCR text:
s
Three layers of evaluation
Layer 1 — Deterministic td
Pons
* eo ES Cea Ee red
im ‘+ en en ere
=
Layer 2 — Semantic
Laver 3 — Behavioural
Prete ae SAMPLE: LLM-as-a-judge prompt
=, e
eaaal Al Engineer
: EUROPE
OCR text:
Databricks Data Intelligence Platform
100% serverless
ae fee Me
_ a a
Lakehouse
ie: Unity Catalog
a
[_nttoae A peta Lake ICEBERG @ Parquet
D
7 |
Se | Al Engineer |
sb) te] a4
OCR text:
Voy databricks . aa eae
rd 7 aera
Multi-Agent BE
Orchestration a
aka ew o~ ae DO Dae em Dim -
»
ee Engineering the future of Al
OCR text:
— CUS ah c) > Al/Bl coe @ Custom Apps
eta eae carts a ve ewe * BA Shout data ond AL apes :
te gence
Pein
cd Od «1 Agent Platform
a *
Contextual
a Sd
* . Developer Platform Al Governance
Reasoning Agents P
Knowledge Assistant Agent Orchestration Agent/Skill/MCP Registry
Supervisor Agent Runtime thm, Al Gateway
Documents Agent Memory Agent Obdservability
Al Functions Model by Managed OAuth Apps
Capacity OO Gein fo aN 8
Alingin
ie | AlEngineer |
: se) Kola
OCR text:
CASE STUDY
Same project.
ate Same problem.
vay DUSCOUSIE eer toe
THE PROBLEM THE KEY DIFFERENCE
a We didn't pick a model
en ee TTP OLeO mG Wels) alee
aac Sete ea Pe CC CECA
»
liters | AlEngineer |
7 ae) e) a3
OCR text:
4 7 CASE STU
Six Ween, post-launch
I y Or
| AlEngineer 625
EUROPE
Estee Det
on J ee ee Ree Ler
. See Tot eae tet ts oot te eh ee
CE eee Bet kn Cie ee. i
ee Mit tides te Bei eae Re a oeeee
jee a ae
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.