Berkshire Hathaway Inc. OS

What are we working on?

Build governed apps, documents, and workflows with your organization's context.
Starter workspaces
Operating Company PortfolioReview permission-scoped updates, capital requests, and material exceptions without centralizing subsidiary operations.
Live App
Technology & Capital ReviewDraft a decision brief for shared technology capabilities while preserving operating-company autonomy.
Document
Federated AI & Cyber ReviewCompare subsidiary-owned opportunities and risks through common evaluation, access, and cost controls.
Slide Deck
Illustrative prototype. Uses public company context and simulated data; no connection to Berkshire Hathaway Inc. systems.
Operating Company Portfolio
Illustrative
Build an operating-company portfolio view that preserves subsidiary ownership. Surface submitted updates, capital reviews, and material exceptions for GEICO, CORT, BNSF, and Berkshire Hathaway Energy.
AI
I start with no access. I'll request only the evidence needed for this workspace.
Gatekeeper: Operating Company Reporting
Read-only access to simulated submissions, exception summaries, and decision records. Each business retains ownership of its source systems and operations.
Approve read-only access.
AI
Executed deterministic query
const updates = await env.OPERATING_COMPANIES.listUpdates({ scope: ["GEICO", "CORT", "BNSF", "BHE"], status: "submitted" }); const exceptions = await env.PORTFOLIO_RISK.listMaterialExceptions(); app.render(PortfolioView, { updates, exceptions });
The portfolio view is ready. It centralizes visibility and decisions, not operating control.
Create a technology and capital review for shared identity, AI governance, and resilience capabilities. Preserve decentralized accountability and make every adoption decision opt-in and measurable.
AI
I'll use curated company context and request only the resources needed for this workspace.
Gatekeeper: Capital Proposals + Decision Records
Read-only access to simulated proposal metadata, submitted benefits, dependencies, and prior decisions. No operating-company production data is exposed.
AI
The brief separates shared leverage from local execution and assigns every decision to a human owner.
Create an executive review of subsidiary-owned AI opportunities and federated cyber controls. Include GEICO claims assistance, CORT inventory planning, and cross-company evidence exchange without pooling raw data.
AI
I'll use curated company context and request only the resources needed for this workspace.
Gatekeeper: Model Registry + Federated Risk
Read-only access to simulated use-case metadata, evaluation summaries, control attestations, and aggregate spend. Raw subsidiary data remains isolated.
AI
Executed deterministic query
const portfolio = await env.MODEL_REGISTRY.list({ ownership: "operating-company", include: ["evals", "risk-tier"] }); const attestations = await env.FEDERATED_RISK.getSummaries(); deck.render(FederatedReview, { portfolio, attestations });
The review is ready. Each candidate remains owned by its operating company and gated by human approval.
Operating Company Portfolio
Live App
Illustrative data. This view does not represent Berkshire Hathaway reporting, performance, or internal systems.
14
Updates submitted
3
Material exceptions
5
Capital reviews
2
Human decisions
Attention queue
Review: CORT distribution systems modernization proposal awaits a corporate decision.
Monitor: GEICO claims-service resilience update includes a subsidiary-owned remediation milestone.
Illustrative submitted readiness
GEICO
94%
Submitted
CORT
86%
Monitor
BNSF
82%
Submitted
Berkshire Hathaway Energy
74%
Review
Technology & Capital Review
Document
Illustrative planning artifact. Based on Berkshire Hathaway's public decentralized operating model, not internal plans.

Berkshire Hathaway - Technology & Capital Review

Federated decision brief | Draft

Purpose

Evaluate shared technology capabilities where scale can help, without replacing operating-company accountability, systems, or investment judgment.

Jobs to be done

PriorityJourney momentRequired review
Identity and access baselineOperating-company opt-inSubsidiary + corporate review
AI Gateway and model controlsShared service blueprintTechnology + risk review
Resilience evidence exchangeException-based reportingGovernance review

Operating principles

  • Start with a measurable job to be done, not a new tool.
  • Use curated company context before model knowledge.
  • The human owner remains accountable for every output.
  • An agent never receives more permission than the person using it.
Gatekeepers hold credentials, scope every resource, and preserve the observation trail when work is shared.

Delivery sequence

Discover: Collect voluntary operating-company requirements and current controls.

Prove: Pilot with isolated data, explicit success measures, and local ownership.

Decide: Record the capital decision, guardrails, owner, and review date.

Control alignment

Source permissions, subsidiary data boundaries, decision history, financial controls, and human approval are treated as design inputs.

Federated AI & Cyber Review
Slide Deck
Slide 1 of 4

Federated AI & Cyber Review

Berkshire Hathaway OS | Illustrative prototype

Slide 2 of 4

Public-context opportunity areas

Use caseStageHuman ownerNext gate
GEICO claims service assistanceCandidateGEICODefine evaluations
CORT inventory and delivery planningCandidateCORTValidate controls
Cross-company cyber evidence exchangeExploreBusiness security leadsConfirm boundaries

Candidate portfolio only. It is not a statement of Berkshire Hathaway or subsidiary initiatives.

Slide 3 of 4

Governance scorecard

100%
Subsidiary owner
3
Shared guardrails
0
Shared raw datasets
4
Approval gates

Illustrative target-state controls.

Slide 4 of 4

Next operating loops

Context: curate terminology, policies, and quality criteria.
Evaluation: define task-specific quality, safety, and fairness tests.
Access: bind every data resource through a Gatekeeper.
Efficiency: use code for deterministic work and models only for judgment.

Integrations

Organization-wide connections for Berkshire Hathaway Inc. OS. Gatekeepers hold credentials, scope resources, and log each action.

Prototype catalog. Connections and authorization states are simulated.
Gatekeepers
1
Productivity suite
Mail, calendar, documents, spreadsheets, and files
2
Collaboration
Chat, channels, meetings, and workflow notifications
3
Project tracking
Programs, epics, issues, sprints, and delivery status
4
Knowledge base
Policies, procedures, standards, and team documentation
5
Service management
IT tickets, incidents, change requests, and asset data
6
HRIS
Employee directory, organization, benefits, and lifecycle workflows
7
ERP & procurement
Finance, planning, purchasing, supply chain, and billing
8
CRM
Customer, account, partner, and service relationship data
9
Code platform
Repositories, reviews, issues, and engineering standards
10
Data platform
Governed warehouse, catalog, analytics, and reporting
11
Security operations
Alerts, cases, exposure, audit, and control evidence
12
Business intelligence
Dashboards, semantic models, and executive reporting
MCP Server Portals

Illustrative remote services available to authorized workspaces.

Security Operations
https://security.mcp.demo.example/mcp
Auto
Enterprise Data Catalog
https://data.mcp.demo.example/mcp
Needs auth
Finance & Procurement
https://finance.mcp.demo.example/mcp
Auto
People Directory
https://people.mcp.demo.example/mcp
Needs auth
Cloudflare API
https://cloudflare.mcp.demo.example/mcp
Auto

Organization Context

Shared, curated knowledge that grounds every Berkshire Hathaway Inc. OS workspace. Context is versioned and read-only to agents.

Public operating context: berkshirehathaway.com · Internal-looking documents below are illustrative.
md
company-strategy.md
Mission, operating model, annual priorities, and outcome definitions
md
brand-and-communications.md
Terminology, voice, accessibility, and approved communication patterns
md
security-standards.md
Identity, data protection, secure development, and incident requirements
md
responsible-ai-standard.md
AI risk tiers, evaluations, human oversight, and acceptable use
md
data-classification.md
Data categories, handling rules, retention, and sharing restrictions
md
architecture-principles.md
Technology standards, decision records, review criteria, and ownership
md
vendor-risk.md
Due diligence, contract controls, monitoring, and exit requirements
md
customer-experience.md
Journey definitions, service standards, and quality measures
md
operations-playbook.md
Service ownership, runbooks, escalation, continuity, and recovery
md
finance-controls.md
Planning, purchasing, expense, audit, and reporting procedures
md
people-policies.md
Hiring, onboarding, performance, leave, and workplace guidance
md
legal-and-compliance.md
Review paths, records, privacy, accessibility, and regulatory obligations

Skills

Reusable workflows for every function. The human requester owns the result.

NameDescriptionGroupSource
meeting-prepBuild an agenda and briefing from authorized calendar, CRM, and document contextGeneralShared library
weekly-operating-reviewCreate a cross-functional summary with decisions, owners, and open risksGeneralShared library
incident-responseAssemble evidence, draft updates, and preserve human approval for containmentSecurityShared library
vendor-risk-reviewCompare due-diligence evidence with security and privacy standardsSecurityShared library
control-evidence-packMap authorized evidence to control requirements and identify gapsSecurityShared library
architecture-reviewReview a proposal against architecture principles and decision criteriaIT & ArchitectureShared library
change-impactMap dependencies, affected services, stakeholders, and rollback requirementsIT & ArchitectureShared library
service-health-reviewSummarize service levels, incidents, changes, and capacity risksOperationsShared library
runbook-builderTurn a procedure into a deterministic workflow with approval gatesOperationsShared library
ai-model-reviewSummarize ownership, evaluations, drift, risk tier, and release readinessData & AIShared library
data-quality-reportAssess freshness, completeness, lineage, and policy complianceData & AIShared library
budget-varianceCompare actuals with plan and draft a finance-reviewed variance narrativeFinanceShared library
procurement-briefSummarize requirements, alternatives, risk, and approval statusFinanceShared library
job-descriptionDraft an accessible role description from approved job architectureHRShared library
onboarding-planCreate a role-based onboarding plan without expanding system permissionsHRShared library
contract-intakeExtract terms, route issues, and prepare a legal review checklistLegalShared library
privacy-assessmentMap a proposed workflow to data categories and privacy obligationsLegalShared library
account-briefCreate a customer briefing from authorized CRM and public informationSalesShared library
proposal-draftBuild a first draft using approved claims, pricing, and brand contextSalesShared library
executive-updateTurn project evidence into a concise decision-oriented updateGeneralShared library

Profile

Illustrative account information for this public prototype.

Demo User
No personal information is stored
Display name
Demo User
User ID
demo.user@example.com

AI Gateway

Illustrative demo data. Visibility and controls across every AI provider Berkshire Hathaway Inc. uses — one console.

Requests
128,400
▲ 11% vs last mo
Tokens
342M
▲ 8% vs last mo
Est. spend
$9,120
76% of budget
Cache-hit
27%
▲ saves ~$2.4k
Error rate
0.6%
▼ 0.2 pts
p50 latency
480 ms
across providers

Models in Use

This month
ModelRouteTokensSpendSharep50 latency
Llama 3.3 70BWorkers AI156M$2,140310 ms
Claudevia AI Gateway98M$3,980720 ms
GPT-4ovia AI Gateway61M$2,510640 ms
Workers AI embeddings (bge)Workers AI27M$19040 ms

Spend vs. Budget

9 days remaining
$9,120spent of $12,000 cap
76%
On track · ~$2,880 left with 9 days
Top Users
Demo User 0142M tok $1,180
Demo User 0231M tok $960
Demo User 0328M tok $840
Demo User 0422M tok $610

Usage by Workspace / Team

342M tokens total
AI Enablement
121M tokens · $3,240
Platform Engineering
89M tokens · $2,460
Customer Experience
62M tokens · $1,510
Enterprise Operations
41M tokens · $1,020
Security & Compliance
29M tokens · $890
Model observability & controls powered by Cloudflare AI Gateway

Governance

Guardrails enforced by Gatekeepers + AI Gateway, with resource-scoped access, audit trails, and human approval.

Per-team allowed models

Restrict which providers each workspace can call.

Llama 3.3ClaudeGPT-4o+ embeddings

Monthly spend caps

Hard limits per team; agents stop before overrun.

Data & AI $4,000Platform $3,000

PII redaction

Strip sensitive fields from prompts before they leave.

Enabled

Prompt / response logging

Full request logs retained for audit & review.

Enabled · 90-day retention

Rate limits

Per-team request ceilings to protect budgets.

600 req / min|burst 1,000

Raise Data & AI cap to $6,000

Change queued by an agent — needs a human sign-off.

Requires approval

AI Gateway Explorer

Explore aggregate model traffic for This month.

4 models
ModelRouteTokensSpendSharep50
Llama 3.3 70BWorkers AI156M$2,14042%310 ms
ClaudeAI Gateway98M$3,98024%720 ms
GPT-4oAI Gateway61M$2,51018%640 ms
Workers AI embeddings (bge)Workers AI27M$19016%40 ms

Review spend cap change

AI Enablement · Monthly spend cap

Current cap$4,000
Requested cap$6,000

Change queued by an agent — needs a human sign-off. Approval updates this demo for the current session only.