Markdown source

Slides: The Prompt Is Still a Punch Card - Ted Johnson, JoinIn AI

Source Video

The Prompt Is Still a Punch Card - Ted Johnson, JoinIn AI

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-003.jpg

Slide text:

8088 8010 ；x86-64:bytes 0018 80001 6604 0068 0018 0011 000c push mov sub mov mov pop ret 868606868686 554889E54883 488B75F05DC3 EC20488B7DF8 0x01 rbp,rsp rbp rsp,8x26 rsi,[rbp-0x10] rdi,[rbp-0x8] rbp sFind Python files>18 sfind.-typef-namepy-size·k $grep-RInTO0O1wc-1 sSho top5by size s1s-Uhs（find.-name'.py） 5curl-shttps://api.exanple.com/\ sCount Tooos in those files 42 sFetch API andpretty-print -re-r--r-. -W-F--r-- -r-r---- -rw-r--r-. -r--r- （"id:2.name²:Linus}. （p.PT.） （"id":3.name²:Grace°） users?linit-31j0. sort-kS-hr|head-n5 8.0Kutils/i0.py 6.1K tests/test_api.py 18k Lib/parser.py 12xapi/elient.py Conditionals user. Variables timit100 Functions ifresp.status_code200: Loops for Item in data: Data (JSoN）Primitive itens-[] def fetch（url,timeout-5): else: "role":'adnin" "16°:123, "email:adaexanple.con* resp·http.get（url,tineouttineout） return resp.json() raise Error(resp.text) ifiten["active"]: itens.append（iten) (NOsr） son Response Format Endpoint Method Auth Token Timeout(s) 10 https://api.example.com/v1/users GET Bearer Token JSON 'actiw':troe, "iit:100, upe,:olo, What would you like to build?.Stream large files.Validate emails Add type hints and deduplicatesrowsby asaJSONarray. Use csv.DictReader Sort by updated_at desc Write aPython function that parses a CSVfile, keeps themostrecent record byupdated_at, and returns theresult Requirements: Include a count summary email (case-insensitive), docstring

Opcodes Shell Primitives Fields Open Language

slide-004.jpg

Slide text:

Tell me what matters.

JUST A STRAW INTO THE OCEAN.

slide-007.jpg

Slide text:

Because it shouldn't need us to anymore.

It can ask a follow-up. It can clarify mid-thought.

It can notice it's missing something and say so.

It should be human conversational

slide-008.jpg

Slide text:

Batch Carried Forward

Loom PunchCard Terminal Prompt

LOOm 2 PUNCH CARD TERMINAL 4PROMPT

2111161111135111366051113s1013151 0G Canno G OJOB payrOKL GSUBHIT Job 31245 subaitted Anatyze Q2 sakes by region S and sumnarize top drivers.: Mew Chst

encode pattem. run batch > encode program, submit batch enter command, wait package intent, submit!

ENCODE / PACKAGE SUBMIT WaIt / PROCESS Output / result.

New intelligence, old protocol.

slide-009.jpg

Slide text:

Expressive growth

hgh

the gap the thoss

Punoh:cirds low Cormmandline GuImouse Touch LUMS

Expressree ceding what you are nlowed to say

Channelhowhebitspbysicalymove

Inletaction protocol Who holds the tloor

slide-012.jpg

Slide text:

Frontier Voice

:tuguo 鲁

You 2:18 PM:

8 When's the next Timberwolves game?

: Fromtier Voice 2:18 PMi

The next Minnesota Tmberwolves game bs

I (Eabanos-

rdoez: no,

Hey. Ted, cane on'rt.

Frontlor Yoico 2ig pM

Sure, I'm here! Whet's on youu mnd?

Mossoge Frontier Voico-.

slide-013.jpg

Slide text:

TL;DR

- OpenAI is reportedly testing an unannounced bidirectional voice model called “GPT-Bidi-1.”

- Code references and early user tests show that the model can speak, hear, and listen simultaneously, handling mid-sentence interruptions naturally.

- The unannounced model has already started rolling out to a select group of app users, hinting at an official release window this week.

slide-014.jpg

Slide text:

Enriching PersonaPlex's output with non-verbal aspects creates an important qualitative

difference relative to systems without this dimension: PersonaPlex now recreates some of

NVIDIA. the same cues humans use to read Intent, emotions, or comprehension.

A Examples

NVIdIa ADlR The following examples showcase PersonaPlex's behavior across different scenarios. in

all audio files, you can hear the user speaking in the left channel and PersonaPlex in the

right channel (shown in green).

Assistant

11 0:01 0:26

engaging way. Prompt: You are a wise and friendly teacher. Answer questions or provide advice in a clear and

In this example from FullDuplexBench's Interruption evaluation, PersonaPlex

demonstrates general knowledge, interrupterability, and natural turn taking.

Customer Service - Banking

0:00 stilif- 0:55

Information: The customer's transactlon for S1,200 at Home Depot was declined. Verify Prompt: You work for First Ncuron Bank which is a bank and your name is Ssnni Virtanen.

customer identity. The transaction was ffagged due to an unusual location (transaction attempted in Miami, FL: customer normally transacts in Seattle, WA).

slide-015.jpg

Slide text:

Conversation is more than turn-taking

Whatreal-time conversational Al has to understand

1.PARTICIPANTS 2.FLOW

mm-hm

Whoishere Who isspeaking Turn-taking Backchannels

Whoislistening 3.MEANING Who isbeing addressed repair+social understanding Natural conversation= timing+context+ Interruptions&repair 4.ACTION Overlap Timing

口

Attention Thread tracking

Grounding Shared context 800 intentions Beliefs/ commitments Decisions/ Whento speak vswait

slide-016.jpg

Slide text:

O Produet roqurements revlew 1 Pauta (URaply)

Listening

Szhuhold euea. Prodvs - Lesd. Sem Broretrg Jordan chooerg rot to sp

FT

TAternoon, Sam. H Jorden.

It joins like a teammate - and starts by listening.

slide-017.jpg

Slide text:

Productrequlrementsrevlewe IPsuse UReplzy

AI

Presenting

Sskeholoer Dana Product-Leod Sem Eogineering Jordan Suggested clarification

QSHAREDBYAI Finance Controls Policy v3 Expense approvals exceeding S5,oo require second-level authorization-the reporting manager and a Finance spprover. 54.2-Approvalthresbods doaypwatoay COALS Hod

O-wnichrequirement is this? REQ-142-expensacrvls

What kind of requests, though? Resove approval scope

Dara Expense approvalsfirst.Accessrequests eventually. asdwey

Al,hold that. Sam

Sam Actually-let'spause.Expense approvals,orageneral approvalworkfiow?

Dana Expense approvals.Flrstrelease.

Sam Access requests are future scope.

Jordin That changos the data model.Good toknow.

Al,pull that up for everyone. Sam

Oh-Idforgotten thatwasarule. Sorduas Pulling up the source -- so the room sees it too.

slide-018.jpg

Slide text:

Product requlrements revlew Il Pause UReply

CHAT

Suskeholoer Dana Product-Lead Sem Eogineening Jordan SAM What'sour currentmedanappxoval time? 23days median right now,the targetis same-day. A

Right-over the imit it routes to a second approver. Sam ogreg re.h'sfine DAV

Agreed.Under five,one tap's fine. Dana GOALS

Sam Worksforme. Okay -— agreed, Expense approvals, five-thousand threshold. Jerztsrs agree Expense spprovals-firstrelesse Resove opprovalsoope Qi-Wnichrequremene is thia? REQ-142pnp

Sam Al,capture that for us. Dana Actually-make the threshold ten thousand,not five. cAChive REQ-142 Answered-aised to$10.0o0[Samconfrmed) Q-ls theS5,oo0threshold stie curreet？ 000

Sam Yes. Want me toupdate the requirement to a ten-thousand threshold? A CUEIS frcanofication.Expens0sovers10,000rouAetosecond-leve approvalper Firstrelesse:managers can approve crreject pending erpense requests drectly requests are explicity outofscope. Approve /reject exponserequests from a notification

Al-is thisroom free after the meeting? Let me check-the room looks free after this,but- A Jordan ODxpnsos oevS1o,o00requiresecond-fnwlapproval (manigirFinance) EmployeeIsnodfied of the declslon Decision is recorded in audt history OManagerreceves anodfication forapendng expense OManager canapprove crejectfromtenotilcation ACCEPTANCECRITERIA

Sam Until three? It updates the spec -- live. Spts

slide-020.jpg

Slide text:

Loops, Prompts, Agents...

useful patterns — but still constraining the interface

Loops: predefined flow; hard-coded steps

Prompts: manual context; submit + repair

Agents: delegated tasks; still shaped by old affordances

Design for what is possible, not just what old interfaces allow.

Hidden Non-Slide Evidence

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