Markdown source

Slides: User Signal Dies at the Retrieval Boundary - Sonam Pankaj, StarlightSearch

Source Video

User Signal Dies at the Retrieval Boundary - Sonam Pankaj, StarlightSearch

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:

User Signal dies at the Retrieval Boundary.

Sonam Pankaj

CEO & Co-Founder StarlightSearch

slide-002.jpg

Slide text:

What is an agent?

An agent is an LLM with agency: it reasons, invokes TOOLS to interact with the world, RETRIEVES from memory, and loops until the TASK is complete.

slide-003.jpg

Slide text:

What is an agent?

An agent is an LLM with agency: it reasons, invokes TOOLS to interact with the world, RETRIEVES from memory, and loops until the TASK is complete.

slide-004.jpg

Slide text:

Retrieval is Static

Context Stuffing

slide-005.jpg

Slide text:

The missing layer between evals and action

slide-006.jpg

Slide text:

The GAP

It has no access to why yesterday's runs passed or failed. The eval signal dies in a dashboard.

This is the missing layer: a system that consumes traces, absorbs eval outcomes, and converts both into retrievable guidance for future runs.

slide-007.jpg

Slide text:

In practice, most agent memory frameworks have focused on user continuity: preferences, profile facts, conversation history, and long-lived personalization.

chat experiences is not self-improving learning system for production agents. *

toddy What h storts Rttrlrvsl signa! L+hr ffom outcoemts?

Raw ctut history Recency To

Etacted lktr, protroco(mbodang imirity Ho

Ertity reutiorahips ovtr dme Graph traveryd · recency No

A14r Yerbu sell refvectiors Shmlarty to currert tast Partisty. refietlons capture lesons, but rerieval k not

atemtarx ttitty scores Iall·linhrd roftctomy with Sinilarity weighted by outcomalderred uirty

slide-008.jpg

Slide text:

Memory as Reasoning

slide-009.jpg

Slide text:

reflect

米

t²-bench

Wie compre beg mode on themirinrs tak inT-bench

erhancedperfarrsaroe.

GLMnenthinking OPTSA

AertRTX

slide-010.jpg

Slide text:

Fettures Hewiworts nvgwtiors Doca Dog Detymlec

GRTSA

ncA ang Loiong-08

SAD

42. M.3.

3.Y

EnowlesgeFnontir H CodirCan Uielony-ce CNTetk

slide-011.jpg

Slide text:

ref/ect Howitwors ntogrtienn Gors

Refecton ebenehand Utieng-ce

RAO

63.3

P3

Lieng-Cs ceTeua

AentRTX

slide-012.jpg

Slide text:

Limitations

Cold Start

Utility Drift

Review quality

Lambda

slide-015.jpg

Slide text:

4. Agentbench

AgentRTX 61.3

Mem0 58

RAG 47

Baseline 35.7

AgentRTX 98

DSPy 82

Baseline 57

slide-019.jpg

Slide text:

reffect Memories 0/pg

PROJIC! Feomceviewedtraces.Sortbytimeora

Oveniw

Prcpects ）

cel

Revievs LEANNEETASE eerou fauebeupug

project Scops

iing CUeage sJafes4s-8fue-Siea-a97f-e29leSoo8rde 68/2026.4251FM FAILED

RETLEOTIGN

SCMUARY

The userreguestedagaming moueDut theagentfalod toretrone anyresits

bndoeppouppu broodersearchterm

COROTACTOM

agentRTx The sgent shcud hse triedabeosde seych_groouctstooltofindreie.antitems.ratherthnghinguparee

Hidden Non-Slide Evidence

Classification audit: raw/sources/slide-ai-classification/slides/Jx4ZFEAq6bY/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.