Gather, future architects. Beginners very welcome.Summer Data Platform Q&A Clinic · 2026

What "the AI has grown" really means

Three paths for giving generative AI context,
and the role of the data platform

Takeshi HiratsukaExecutive Vice President, Leave a Nest Knowledge Co., Ltd.

話し手についてSpeaker

Takeshi Hiratsuka / EVP, Leave a Nest Knowledge Co., Ltd.

平塚 武

Data & AI Advisor and Salesforce Architect

1Speaks at 20+ conferences a year in Japan and abroad, including Salesforce events in the US
2Active in Salesforce data platform communities, in Japan and globally
3Focus areas: Salesforce Data 360, Tableau, Slack, Snowflake

X: @marreta27_jp · marreta27.me

今日話すことAgenda

A 30-minute session, so you can explain
"the AI has grown" in your own words

1What "grown" really meansIs it really the AI that grew?
2Three paths for passing contextMemory / skills / data pipelines
3The data foundation underneathThe "how" on Salesforce Platform: Data 360

Written for beginners: technical terms are unpacked as they come up.

業務の現場でよく聞く声Field Voices

"Our AI has really grown, hasn't it?"

It gets what I mean now without me spelling it outSales team
The answers feel sharper than they used toPlanning team
The AI has learned about our companyIT department

That feeling is not wrong. But it is not accurate either.

しかしBut

Generative AI does not get smarter on its own
just because you keep using it

Time in use Perceived "intelligence" Model parameters = fixed Nothing you type retrains the model. It does not learn on its own.

The model's internals stay fixed. So what actually changed?

？
本日の問いThe Question

What does it actually mean
to say "the AI has grown"?

結論Answer

What grows is not the model.

Model (fixed) Surrounding systems Mechanisms that pass context: memory, skills, data The people using it Skill to instruct, verify and delegate well The organisation embedding it Processes rebuilt on the assumption of AI

What grows is everything around the model. Today we focus on how context gets passed.

前提のはなしKnowledge Cutoff

Every model has a knowledge cutoff

Generative AI model Weights (knowledge) are fixed Cutoff: the knowledge deadline Learned everything up to here Beyond this, the model knows nothing Yesterday's notes This morning's ticket This week's new rule The other wall: internal data was never public, so it was never trained on, whatever the date.

"Now" and "our company" can only be passed in from outside. That is what the three paths do.

「成長」を分解するThree Paths

Three paths for passing context to AI

① Memory Past exchanges and user details ② Skills / prompts Reusing procedures and criteria ③ Data pipelines Prepare and deliver company data Personal context Task context Company context Generative AI (fixed) Useful answers for that company

Same model. The context you pass changes the answer.

経路 ①Memory
Personal context

Memory: carrying "you" over

Preferred name and tone Work in progress, past decisions Past exchanges carried over Generative AI (fixed) "Let's pick up where we left off" You "Your proposal for Company A. 　You'd got as far as the quote." AI (with memory)

"From where we left off" just works. That is why it feels like a dedicated partner.

経路 ②Skills & Prompts
Task context

Skills and prompts: carrying the "how" over

Skill: summarise meeting notes Step 1 Decisions first Step 2 Owner and due date per action Step 3 Under 300 words, polite tone If in doubt, decisions come first Reuse it many times Today's meeting → same quality ✓ Client meeting → same quality ✓ Another team → same quality ✓

Don't let a good instruction be a one-off. Make the method a reusable asset.

経路 ③Data Pipelines
Company context

Data pipelines: delivering context to AI

Data pipelines: collect, prepare, deliverCompany dataCustomers / deals / casesPrepare & updateUnify / check / refreshSearch & retrieveRelevant, permitted dataGenerative AIUse data as contextQ: How has the return rate changed over the last three months?Retrieve the relevant data → answer using it as evidence

Collect, prepare and deliver up-to-date company data to AI.

Using the delivered information as evidence for an answer is grounding. Here, the path includes data preparation through retrieval for AI.

整理Recap

The model is a fixed amplifier

Feed it clean context Model Clean, useful answers Feed it messy input Model ← same amplifier Noise comes out amplified too

What improved is not the model but input quality and operations. That is what "the AI has grown" really is.

では、どこから育てる？Where to Start

Start with a pipeline for company data

Memory Personal context The gain stays with each person
Skills / prompts Task context The gain stays within each task
Data pipelines Company context Prepared data becomes a shared, reusable asset

Personal and task context only pay off on top of correct company data.

The second half covers the data platform that supports collecting, preparing and delivering company data.

価値の出し方Gartner's 3 Business Cases

Generative AI value comes in three shapes

Best bet Defend Extend Upend Hold the line Grow the core Change the game Raise day-to-day productivityand stay competitive Strengthen existing processesand differentiate Create new products and marketsand reshape the industry Return: given back to employees Return: return on investment Return: recovered in the future Gartner puts the surest short-term financial return in Extend Source: Gartner, The 3 Business Cases of Generative AI Value (Nate Suda, Hung LeHong / Jan 6, 2025)

The best bet, Extend, is existing work times your own data. Foundation quality becomes value quality.

ただし、現実はReality

Company data is never in one place

Inside Salesforce alone, many systems

Sales CloudService Cloud Marketing CloudCommerce Cloud TableauSlack MuleSoftAgentforce …and many more

Across all internal systems (enterprise average)

957applications
27%: the average share that is actually connected

Data piles up separately, wherever it was born.

Source: Salesforce / MuleSoft "2026 Connectivity Benchmark Report" (survey of 1,050 IT leaders worldwide)

Salesforce Platform での HowData 360

Data 360: the foundation for unifying
scattered data and putting it to work

Scattered data CRM (customers) Core systems Web behaviour Email sends External DWH etc. Data 360 Salesforce's data platform (formerly Data Cloud) Collect ingest → Connect and align unify, identity resolution → Deliver sent back to each system Analytics & BI Marketing AI & agents

Make scattered data usable as one single fact.

設計思想ELT, not ETL

Not "transform first" but "load it all first"

Traditional ETL E Extract from source systems T Transform first decide the use, narrow it down L Load only what you kept Use case A only Data 360's ELT E Extract from source systems L Load it all first as is, into Data 360 T Transform on use per use case, each time Use A Use B Use C… Transform and "activate" into each system

Bring it in first with E and L. Then transform (T) and redistribute to the systems that need it.

全体像Data Hub

Data 360 becomes the data hub

Data 360 the data hub Sales Cloud Customers, deals Service Cloud Cases and support Marketing Cloud Personalised campaigns Core & external DWH connects to Snowflake too Tableau Analytics & viz Agentforce Grounding for generative AI

Data from one system starts working in every other system.

3つの経路、ふたたびBack to Data Pipelines

Delivering trusted company data to AI

Company dataCustomers / deals / historyData 360Unify / prepare / refreshSearch & retrieveBy question & accessGenerative AIEvidence as contextQ: What is our next step for Company A?A: Confirm last week’s two support tickets are resolved before proposing more.Grounding = using retrieved company data as evidence for the answer

The data platform supports the pipeline that delivers trusted data to AI.

むすびにClosing

"Growing the AI" means growing
your data and your operations.

What grows is not the model: it is your data assets, and your ability to use them.
Put a data platform underneath.

ありがとうございましたThank You

Bring your own data questions
to the open clinic session after this.

Source articleIs AI something you "grow"? note.com/marreta27 (Japanese)
X (Twitter)@marreta27_jp
Websitemarreta27.me: talks and activity
X: @marreta27_jp

Takeshi Hiratsuka / Executive Vice President, Leave a Nest Knowledge Co., Ltd.

← / → to move · F fullscreen · L language
Privacy