Slides: Claude Agent SDK [Full Workshop] — Thariq Shihipar, Anthropic
Source Video
Claude Agent SDK (Full Workshop) — Thariq Shihipar, Anthropic
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:
/ ANTHROP\C
Claude Agent SDK
OCR text:
Agenda
e What is the Claude Agent SDK?
e Why use it?
e How do you design an agent?
e Example: prototyping an agent
ANTHROPAC CONHDENTAL 2
OCR text:
Ta
ce
7 N
y : ~
| ; ‘ v a .
ve ok . : | .
ee ae — = E ; .
L | |
7) 7 ;
OCR text:
The evolution of agents
. “8 @
y aa 3 a
Single-LLM Features JER WOIKIOWS am: Uae c
Summarization, LLMs orchestrated LLMs deciding their Agency %
classification, by code own trajectories Capability +
extraction ...
Flexibility t
ANTHROP\C COME IGEN TIAL 4
OCR text:
Claude Code
Claude Code is our an agent that Oe
lets developers automate software
development tasks with Claude. : ee
This is the first digital worker
doing hours of work.
a
ANTHROP\C CONFIDENTIAL 8
OCR text:
Building effective
agents
1/ Models
High-performance
reasoning models
that understand
complex instructions
and execute multi
step workflows.
Models
mm Fastest with near-frontier intelligence Bee T SC Tee ley efor ad bis ta eee geen
ANTHROP\C COSEIDEN TIAL
OCR text:
Building effective
agents
5/ File System bce ee seen tect cn cin en 2
Give Claude a
persistent workspace Tools Prompts File System
to pass files, write
just-in-time code, “we com Neen Prosens ies
and learn from Custom Custom mess
example outputs File System Workllows Ex Output
Claude Haiku Claude Sonnet Pelt Ce tT]
Dee em csul iad RT a teva e evel tg beet Ere Bi gn
ANTHROP\C CONHPIBENTIAL
OCR text:
Building effective Your Application
Pp :
agents
Claude Agent SDK
9/ Application Fe = a a RT 9 TN Re TE =
. SKILLS Enhancements.
Focus on your unique Subagents
value - build tne user Tools Prompts File System Web Search
experience an Research Mode
domain workflows “we Som Neen Proce is Auto Compacting
that differentiate your Coton “sen ures Hooks
product File System Workllows Ex Output Memory
- [ é Models
oT CS Seat POUT CRS Tita Claude Opus
DE mesure Smartest for complex agents /coding be se Eire BC mete ag
ANTHROP\C CONEIDEN TIAL
OCR text:
Users are already building cutting-edge agents
with the CC SDK
e All software tasks
SRE agents
Security agents
- Incident triage agents
Automated bug fixing
e Site & Dashboard builders
e MS Office agents
e Legal, Finance, Healthcare, HR
e Custom agent instructions + any integration
Cloud logs, CI/CD, ete.
ANTHROP\C CONEIDEN TIAL w
OCR text:
Why use the Claude Agent SDK?
- We realized people were using Claude Code for non-coding tasks.
- We build our agents on top of our Agent SDK.
- Itis built on top of lessons we've learned deploying Claude Code to
millions of users.
- We have strong opinions on the best way to build agents.
- One of our biggest learnings: the Bash tool is the most powerful agent
tool.
OCR text:
The Agent SDK is the best way to build agents in the
“Anthropic Way”
- Unix Primitives: Every agent should use bash [1] &a file system [2] (e.g.
skills & memory)
- Agents > Workflows: Agents build their own context + decide what to do
with tools.
- Code Generation for non-coding: We use code gen to generate docs,
query the web, etc.
- Every agent has a container.
OCR text:
Bash is all you Need
Bash is what makes Claude Code so good.
Bash is a generic version for “code mode” or programmatic tool calling.
The bash tool allows you to:
- Store the results of tool calls to files, so you can search them.
- Store memory in files.
- Dynamically generate scripts and call them.
- Compose functionality, use unix primitives like tail, grep, cat, etc.
antundSe existing powerful software like ffmpeg, libreoffice, etc.
OCR text:
Without Bash With Bash
User: How much did I spend on ride sharing User: How much did I spend on ride sharing
this week? this week?
Tool Cali (Gmail Query Search) query: “uber Tool Call (Bash): ‘gmail_search.ts query
OR lyft” “uber OR lyft*
Thinking: I found the following emails that Tool Call (Bash):
mention Uber and Lyft, I need to find the
cost of each email. gmait_search.ts ~- a> :
wot sf
ret > : " IN
Claude: The total is... (hallucinated) wee : 1 \ a, /
OCR text:
e
ae i ee: a 7 an a 7 re re ye an keer ee ae ae ad / eT
ce or ee ne
Con Cams CE Co OY 6 FS 6 Tobiys apna 55) \
nL ae ee Ld OY
JQ atte te ia
sort . >» contacts this week.txt
Ch ee ce en ke ee to er ne ee eee
read ematl;
(10 as nt ee ce \
rT eee are Oe a CLT Cs LS ar 2k oa 01a: (en Oe (Ree Xel an
Sleep (il 8 tate ar
< contacts this week.txt
a cae ee ea ed
fem contact detatls.jsonl > final contacts.json
OCR text:
®
a) 7 ee 2 Ce eee Pia ee 8
a aes a |B Bue BO pee 4 / en ans
fimpeg earnings call.apd .: rn Lee aone Mr: TUrO ROC PRU LG
whisper audio.way mene @ ROLL a ee 2:1 0)) ee ese a Oh OED 01D
ee i ea ee ee ee
‘gee a PP tee transcript.json > moments. txt
ce ora oe ee nr re ee: ce
_ ag ee transcriupt.}son | \
read start end,
{fepeq earnings call.mp4 ren ae rn
Como) 0 EO
OCR text:
ANTHROP\C
Workflows & Agents
OCR text:
Workflows & Agents
We build both agents and workflows on the Claude agent SDK.
Agents are like Claude Code - You talk to them in natural language
and they take action.
Workflows are like our Github action - You define inputs and
outputs, e.g. take in a PR and give a code review.
OCR text:
Workflows & Agents
When building workflows, use our structured outputs:
https: //platform.claude.com/docs /en /agent-sdk/structured-o
utputs
OCR text:
lay
i :
A '
‘ ‘
x ; : MW
n 2
Pn -
oe
ma]
OCR text:
Claude Agent SDK Loop
Gather Take Verify
context action work
ANTHROP\C COMPIEENTIAL 29
OCR text:
Tools vs Bash vs Code Generation
Tools:
- Pros: Highly structured, highly reliable
- Cons: High context usage, not composable
Bash:
- Pros: Composable, static scripts, low context usage
- Cons: Longer discovery time, slightly lower call rate
Code Gen:
- Pros: Highly composable, dynamic scripts
- Cons: Needs linting & possibly compilation, careful API design
OCR text:
Tools
Use tools for: Atomic actions your agent mostly needs to execute
in sequence
For example:
- Writing a file
- Sending an email
OCR text:
e
Why Skills?
agent-skills / skills / public; (0
Skills let one agent accomplish longer, more complex = eter
7 * @ peteriai-ant adc ng better docx redir ng {B19b)
tasks without needing Subagents. The agent can use
many Skills and read them only when needed. Name
se
Nom fet roe cteate ibe Laced dovument th all the information fue gathered. and ben
eile an cxevuthe summary in Woed Be docx
eae meee gd ete MD pptx
In oe oe
Some tet me dorate The Wotd document with the execuxlve summary
a Law. QUAN aeegineRneie
New Pilteute the Ward document cveuthe sunimasy:
OCR text:
What are Skills?
An organized collection of files (instructions, nd onw
executable code, assets) containing everything docs.nd
Claude needs for a specific task. L apply..template. py
Skills give an agent:
General capabilities Claude isn’t good at out of the box (yet)
e.g. creating PDFs, Excel, & Powerpoint files
Knowledge of an organization’s workflows and best practices
e.g. Anthropic’s Brand Styling
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.