The Reference Behind the Decision.

EPAI · Embodied Physical Agentic Intelligence · B2AGI’s proposed framework
Capital · Security · Global Cooperation

Embodied AI, physical AI and agentic AI have arrived on the same desk.

Korea · China · United States reference setsEPAI 2027 · Edition v1.6
How a reference is made

Bring a question. Bring a draft. B2AGI adds evidence, review and a way forward.

01
Start with a question

The concept begins with the decision at hand, then brings relevant evidence, options and open items into view.

02
Or bring a draft

A review can mark what to keep, change or add, with the reasoning and supporting references beside it.

03
The reply

Judgment, options and a way forward. For drafts, what to keep, what to change, and why. Your principal’s questions, with basis and open items.

04
Review statement · concept format

The proposed format records the scope, references and limits of a review. Issuance is not yet available in this concept.

Enter the briefing room  →

SPECIMEN · NO RECORD NUMBER
REVIEW STATEMENT
Specimen · not issued · illustrative case
Subject
Ministerial briefing draft: AI options for a distribution operation
Result
2 changed · 1 kept with an addition · 1 added
Not verified
Supplier prices, other customers’ peak schedules and the platform outage plan (not provided to B2AGI)

Open the full specimen →

Concept preview · your question stays in this browser tab; it is not sent or analyzed. The walkthrough uses a fixed warehouse case.

Briefing room

What do you need to walk out with today?

Start with a question, or bring the draft you already have.

01

Ask a question

This concept uses a scripted example to show how a question could lead to evidence, options and a way forward.

02

Bring your AI draft

This screen shows a sample review of an AI-generated draft, including what to keep, change or add.

Sample draft · 12 pages
Shown for this walkthrough · upload is disabled
Review the sample draft→
View a sample reply  →
B2AGI · Briefing roomThe Reference Behind the Decision
Sample conversation · warehouse caseConcept preview · scripted
Illustrative caseThe distributor and its three proposals are a constructed example, following EPAI 2027, Chapters 1 and 3.
Sample reply · question

“We received proposals from an embodied AI, a physical AI and an agentic AI company. Are they competitors, or do we need all three for the same warehouse operation?”

Asked by: policy adviser, preparing a ministerial briefing · Market: Korea

01 · Judgment

Treat them as parts of one operation. Compare them against the work the distributor must do: orders, stock and delivery promises handled together.

02 · Follow the work

Each proposal uses a different AI vocabulary. Put the customer’s task in the middle and mark where each one takes part.

StepAgentic AI proposalPhysical AI proposalEmbodied AI proposalExisting staff
Order received or changedProcesses the order——Handles exceptions
Stock checked, work scheduledAllocates work———
Goods moved and handled—Handles goodsHandles goodsResolves what machines cannot
Delivered as promisedUpdates the promise——Customer contact

Overlap: two proposals claim the same handling work. Settle this before comparing prices.

03 · Where they must connect
  • When the customer changes an order, what tells the warehouse?
  • If a package cannot be handled, where does the request go?
  • If the platform is unavailable, what work continues?
04 · Options and a way forward
AOne lead supplier made accountable for the whole result. Simple to contract; check whether any one of them can keep the full promise.
BA local operator keeps the service working, with the suppliers providing tools. Often better placed to own the handovers.
CA shared-capacity service, if other customers’ busy periods fall at different times. Depends on their schedules and the cost of moving capacity.

Next step: write each supplier’s responsibility boundary, and who takes over, into the purchase negotiation.

05 · Your principal’s questions

“Why not choose the cheapest single proposal?”

No single proposal covers the whole operation. The cheapest one may leave the handovers between steps without an owner, and those are where the delivery promise fails.

Basis
EPAI 2027, Ch. 3, “Put the work in the middle”
Open
Price of operating the handovers has not been quoted by any supplier.

“Who is accountable if the system stops?”

Nobody yet. Put it in the contract: where each supplier’s responsibility ends, who takes over, and what work continues during an outage.

Basis
EPAI 2027, Ch. 3; Ch. 14, “When the Robot Stopped”
Open
Service terms and outage plans not provided.
Illustrative caseA constructed draft on the same distributor example, written as a general-purpose AI assistant might.
Sample review · draft as submitted

“Ministerial briefing: AI options for the national distribution pilot”

Submitted: 12 pages · Prepared with a general-purpose AI assistant

p. 3 · ¶2“The three vendors offer competing solutions; the embodied AI proposal is the most advanced and should be selected.”
Change
The comparison is between labels. The three proposals touch different steps of one operation; two overlap on goods handling. Compare them against the work, then decide.
p. 5 · ¶1“Each vendor provides AI for warehouse automation with proven deployments.”
Keep, with one addition
Accurate as a summary. Add which step each deployment covered, so the reader can see what has been proven and where.
—Not in the draft
Add
A section on who takes over when the platform is unavailable, and which supplier’s claim of completion holds if the customer refuses a delivery.
p. 9 · ¶4“Total cost: the sum of the three license fees.”
Change
The cost of operating the handovers is missing. Mark it as an open item until a supplier or operator quotes it.
SPECIMEN · NO RECORD NUMBER
REVIEW STATEMENT
Specimen · not issued · illustrative case
Subject
Ministerial briefing draft: AI options for a distribution operation
Material submitted
Client draft prepared with a general-purpose AI assistant, 12 pages
Scope of review
Factual claims · cited sources · terminology · anticipated questions
Reference sets
Korea · China · United States
Result
Reviewed with corrections: 2 changed, 1 kept with an addition, 1 added
Not verified
Supplier prices, other customers’ peak schedules and the platform outage plan (not provided to B2AGI)
Specimen · B2AGI seal
Concept specimen · no record issued
Concept specimen · no record issuedPage 1 / 1
Concept format · not issued

The seal records what was reviewed, and against what.

This is a specimen of a review statement under the B2AGI seal. The live review and issuance process is not connected.

It also names what B2AGI could not verify. That line is part of the statement, so the reader of your briefing knows where the review stopped.

← Back to the sample reply
Reference room · EPAI 2027, v1.6

Embodied Physical Agentic Intelligence

Embodied AI, physical AI and agentic AI have arrived on the same desk. EPAI brings intelligence, action, infrastructure, people, money and institutions onto a common map.

§3.1

EPAI

Embodied Physical Agentic Intelligence is B2AGI’s proposed common coordinate for understanding AI capabilities and their relationships with physical action, people, infrastructure, capital and institutions. It connects overlapping fields without declaring their terms interchangeable.

§3.2

Ecosystem

The EPAI ecosystem is the field of participants, capabilities, resources, dependencies and affected lives through which intelligence produces consequences. Its boundary is stated for the undertaking being examined and can be revised when an excluded relationship matters.

§3.3

Framework

The EPAI Framework supplies common concepts and questions for examining that ecosystem: purpose, capability, dependency, authority, evidence, economic arrangements and consequences. It supports comparison without substituting for domain-specific requirements.

§3.4

Prism

The EPAI Prism is a method of bringing accounts from different domains into contact so that evidence, constraints or objections in one can change the questions, options or judgments in another.

Cite as: Henry Chan Jung, EPAI 2027, v1.6, 27 October 2026, §3.1–3.4.