Slides: Deterministic Infra for Non-Deterministic AI Agents - Nishant Gupta, Meta Superintelligence Labs
Source Video
Deterministic Infra for Non-Deterministic AI Agents - Nishant Gupta, Meta Superintelligence Labs
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:
BUILDING DETERMINISTIC
INFRASTRUCTURE FOR
NON-DETERMINISTIC
AI AGENTS
THE EMERGING CONTROL PLANE
FOR AUTONOMOUS AI SYSTEMS
DETERMINISTIC BOUNDARY [DB-SOO]
ENTROPY REDUCTION: 99.9%
PREDICTABILITY INDEX: 1.0
SYSTEM VARIANCE: <0.01%
CONTROL ENCLOSURE [CE-99.99%]
SIGNAL
SYNCHRONIZATION
SIGNAL
RECTIFICATION
STATE
SYNCHRONIZATION
RIGID FRAMEWORK [RF-AIZ]
CONTROL ENCLOSURE [CE-99.99%]
Nishant Gupta
Tech Lead @ Meta
OCR text:
Infrastructure was built for 2
predictable microservices.
THE GREAT MISMATCH
Traditional sexcrin ous x Autonomous
Microservices MASSES APE ene Al Agents
Stateless | —| —_, Stateful
_ (STATE: 0] ome {STATE: PERSIST]
Deterministic | a a a Probabilistic
| [PATH: FIXED] | °° [PATH: DYNAMIC]
- Request-response ———|> Multi-step workflows
[FLOW: SYNC] eh (FLOW: ASYNC]
, (TIME: <1¢0MS) 2 Pee [TIME: >MIN/HR]
OCR text:
Demos optimize for capability. | ad |
Production demands reliability.
EE (STRUCTURAL INTEGRITY CRITICAL]
eo
a —“» SE _|| Capabilities
CS ©
= ____ ge —
Reliability | Jai? ai
Core - a ee .
Citi il cali
: P eT : ~~ : | Reliability Core
iat ae an “
ae <> <—ae
in, oes
eae j Ca
OCR text:
s a = e f
Real production failures originat
the infrastructure, not the model.
Diagnostic Failure Tree <7 to .
. - reasoning loops deadlocks
re fo
Model Output . ‘ ““! hallucinations ““) storms “*“I explosions
, , “" drift poisoning
OCR text:
Ihe anatomy ofan > adil
y Exponential GPU — g - gases
t t t Compute Spike
agent retry storm.
nf SE
SSR
Recursive aL oston
reasoning loop ,
.. 1 oo. 10 og we . locks up. we
ito a : Step 3: Agent SE = |
Ron erat et eee a attempts to fix vooe | WA Step 5:
aOe ie eseas| : my a urror but RECURSION IN” oo IMvALio The
eras caren CEnar COGe. haltucinates new fe PARAMS. g Consequence
invalid parameter. ie =
: : g
Step 1: Agent Step 3: Agent Step 4: : é
hallucinates Step 2 attempts to fix Recursive 5 Step 5:
invalid API Tool returns error but reasoning loop x The
parameters. error code. hallucinates new locks up. Time (ms) " Consequence
invalid parameter. .
7 foe * DATA SPIKE EVENT
(CRITICAL FAILURE PATH) [TIMELINE AXIS
OCR text:
The model merelyproposes. Theplatformdecides.
Validate Validate →
ndno Aligned
Stochastic Core Execution Safe
System Compliance
Deterministic Wrapper Filter Route
[ANALYSIS:STRUCTURAL INTEGRITY] [FLOW:DETERMINISTIC CONTROL
OCR text:
requiremultidimensional observability. Logsaredead.Autonomousworkflows
AgentTraceTimeline [TelemetrySidebar]
T+0.1s T+0.2s T+0.1s T+0.5s T+1.0s T+1.2s
LLMDecisions Track1: Call Tool A Decision. [OY] Call ToolA Decision: Query Memory Decision: [FLON:R[ASONING] Query Memory Decision: Call Tooi A Decision:
Pians Track2: Orchestration [OATA:ACTIVE]
Execution Track3: Tool Calls& [OATA:ACTIVE] Tool A Executing API Response Received [FLON:DRXTING]
Track4: MemoryAccess Context BlockX Read [FLOW:REASONING] New StateY Write: [DATA:ACTIVE]
[NISY01] [OATA:ACTIVE] [04TA:ACTIVE]
Track5: State Transitions Slate Updated State
T+9.1s T+0.1s event T+8.5s T+1.0s T+1.0s
[ANALYSIS:FLOWTRACING] [OATA:CONTIMUOUSSTATE]
OCR text:
Shared memory coordinates 7
chaotic multi-agent consistency
- Shared Memory Architecture 3 .
“| State Store :
va SS = ee a ~ :
~*~ DT ~
we ge FOr
Context drift _/ “ See F sratememory
Ud ~ <S ;
Conflicting facts a . * a!
OCR text:
Human oversight is an escalation path, A: .
not an operational bottleneck.
Workflow Routing Diagram
; eee OME
| ; Automated Agent Tasks “KG a ‘7 Automated Agent Tasks ; “
a ae i a —
=. 7 a Vou
Human Approval Node ~t
(ANALYSIS: ESCALATION PATHWAYS]
OCR text:
Inference at scale fundamentally a.
becomes a cluster scheduling problem.
Elastic Scaling Graph
‘ - 4 See Long-cunning
3 200 5 Fi
5 mi Lt EE 2 | Seaview
| LL Oo
~ = Loe nH \a/ A J Bursty Demand
AeA —|]
FRC
° “— TIME (sec) “e “
OCR text:
r we
The Reliability Rosetta Stone. mdi
. Distributed Systems Pattern Agent Equivalent
Circuit Breakers Tool Isolation .
1 {PATTERN TSOLATION] ry t [PATTERN ISOLATION) i
Rate Limiting Agent Limits
; ‘ TUENtt Une cme tee Tf iy C toate £59 ORC OREM) | 1
Retries : Controlled Recovery
1 (MECOWERT LOGIC] iD ' (necovtmy coCreh x
Quotas Cost Governance
: Lae SOU e CAPS] a q _ __ _ Ce suet CAPS) k
Observability Agent Tracing 5
va (IMACEAMILATY system) | + . Cimacea@icity system) « #
[ ANALYSIS: RELIABILITY MAPPING ]
OCR text:
Competitive advantage has shifted fron a |
prompt engineering to systems engineering.
THE PARADIGM SHIFT
PROMPTS compound ocean .
- . . Trajectory L in
, ea . ir “y “A W
“ “MODE J .
7 on LS . Commoditization * 4 ;
J Th oe g L
f me ae ° , Commoctiation
f Oe oe INFRASTRUCTURE
- — We
OCR text:
Al agents are distributed systems. meee
Treat them accordingly. The future
. ‘ of Al won't
| ' | be won by
CORE TAKEAWAYS better
genio Pane * Firastructue must Besircty. prompts.
ee poblemtisenmastucure. ff | Itwillbe
" Mtidmensonlobeenabity s JF won by
v8 “"-“s Agent contro! planes are the : patter
: ms required runtime layer for systems. |
a _ Production Al. 4
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.