Control plane synced
Nick D.
nick@hiday.ai
Administrator · AIRB Chair

This is the agent view: the current state of the control plane rendered as plain Markdown, the way an AI agent would read it. The numbers are identical to the human view. Demo data; real metrics are defined per engagement.

Loading tenant state...

Executive view · EnterpriseOS Layer 01 · 2026

Your enterprise is an operating system. This is its control plane.

Every model, agent, workflow, and dollar of AI, governed in one place. Demo tenant: a 15,000-employee global enterprise. Metrics are illustrative and defined per engagement.

EnterpriseOS · seven layers · eOS v2.0 Hover a layer for live status · click to open

Value against risk, per production system

Top 12 by value · bubble = adoption

Portfolio pipeline

Intake to impact
68% pilot-to-production graduation Vertical baseline near 10%

Compliance posture

Frameworks

Risk overview

Register
Action required
Contract Intelligence: unclassified repository
Data readiness gap · Legal · due Aug 14
High
3 unvetted MCP servers in use
Tool gateway escape · IT Security
Critical

Adoption pulse

Detail
2,890
Active users, 28 day
86%
Satisfaction
Perception gap

Leaders estimate 4% of staff are heavy AI users. Measured: 13%. Your people are further along than you think.

Recent activity

Decision log excerpt

Framework pulse

1 / 8
Full library

Governance · Architecture & Intelligence Review Board

One board. Every AI decision. Governance enables speed.

The AIRB unifies what most enterprises still run as four separate tracks: technology review, architecture review, AI ethics, and change advisory. One agenda, one decision log, one SLA.

In review
8
AIRB queue today
Median cycle
11d
Target 14d
Decisions QTD
18
12 approved · 4 conditional · 2 declined
Auto-decided
31%
Low-risk consent agenda
Active policies
24
92% asset coverage

AIRB queue

Aging vs 14-day SLA

The unification

Why one board
a
Before: four queues
Technology Review Board, Architecture Review Board, AI Ethics Committee, Change Advisory Board. Four agendas, four backlogs, months of elapsed time.
b
After: one AIRB
A single AI-forward body with decision rights over architecture, risk, and change. Low-risk intakes auto-decide by policy; humans review what deserves humans.
c
The result
11-day median decisions, a defensible audit trail, and a portfolio that ships. Governance is the steering wheel, not the brakes.

Agent autonomy board

31 agents · Gartner tiers

Tool gateway

MCP control plane
18
Approved MCP servers
3
Unvetted, detected
214
Injection blocks / mo
37
Redactions / wk

Model registry

23 approved · allow-list

Platform-agnostic by design: Anthropic Claude, OpenAI & Microsoft Copilot, Google Gemini, AWS Bedrock. The model is not the moat; the architecture around it is.

Identity Ledger

Attribution coverage
99.6%
1.4M
Actions signed / mo
57
Non-human identities governed
12
Orphaned actions flagged
90d
Key rotation, all vaulted

Every invocation ties to a human, an agent, or a non-human identity, least-privilege by design. Orphans page the governance desk and their keys expire in 48 hours.

Evidence Vault

Lineage completeness
97%
4h
Median audit-pack turnaround
6
Evidence packs QTD
2.3M
Lineage records
100%
Retention per policy

Who ran what, with which model, on which data, and why. Two regulator requests answered this year from the vault alone; the rest were internal audit.

Decision log

Immutable · audit-ready

Recent policy changes

Change timeline

Policy library

24 active · click a policy for versions

Use case catalog · 73 governed systems

Every use case, from first idea to retired gracefully.

Intake in minutes, feasibility scored by the engine, routed by risk. The catalog is the single registry behind every number in this control plane.

Risk · identification and containment

Find the risk before the auditor, the attacker, or the headline does.

AI is now the number two global business risk, up from ten, the largest one-year jump in survey history. This register holds all 47 open items with named owners.

Open risks
47
Register below
Mitigation coverage
100%
All critical and high owned
Shadow AI found
17
This quarter · 9 sanctioned
Unsanctioned use
22%
Benchmark 45%
Uncontained incidents
0
2 contained YTD

Residual risk heatmap

Click a cell to filter
Likelihood (vertical) x impact (horizontal) · count of open risks

EU AI Act exposure

69 active systems classified
Wave tracker

Penalty bands: EUR 35M or 7% of turnover (prohibited), EUR 15M or 3% (obligations), EUR 7.5M or 1% (misinformation).

Boundary Sentinel

Unexpected data access
6
Unapproved model swap
3
Unauthorized tool addition
5
1.2M
Boundary checks / day
40s
Median containment

Live behavior is compared to the approved system card, continuously. 14 violations this quarter, all auto-contained; 2 escalated to incident response YTD; 0 uncontained.

Drift Radar

Baselines current
31/31
9
Drift events QTD
6 min
Mean detection
Actions taken
5 auto-rolled back 3 paged a human 1 quarantined

Every governed agent runs against a behavioral baseline from its system card. Deviation acts first and reports always: rollback, page, or quarantine.

Risk register

All 47 open risks

Risk taxonomy

Mapped to OWASP · NIST · MIT

Shadow AI funnel

Detection layers
Network analysis Browser DLP OAuth monitoring Extension audit Expense sweep
Posture

41 outstanding OAuth grants to AI apps under review. Sanction fast, block rarely: the funnel exists to say yes safely.

AI spend · FinOps for AI

Cheaper tokens, bigger bills. Spend needs a steering wheel too.

73% of enterprises overran their AI budget this year. This tenant is at +10.7% with every dollar tagged to an owner, a use case, and an outcome.

FY budget
$8.4M
Approved AI envelope
Projected actual
$9.3M
+10.7% variance
Run rate
$780K
Per month
Tagged spend
87%
Benchmark: 22% achieve this
Waste flagged
$61K
Idle GPU / month
Savings redirected
$1.2M
Self-funding YTD

Monthly spend

FY 2026 · hover for values

Where it goes

$780K monthly run rate

Seats include Microsoft 365 Copilot at 2,400 x $30, ChatGPT Enterprise at 900 x $60, GitHub Copilot at 800 x $19, Claude Enterprise at 500 seats, and embedded vendor AI SKUs.

Token economics

1.8B
Tokens / day
$6.07
Blended $ / 1M tokens
41%
Agentic share of tokens
34%
Cache hit rate

Blended token cost fell from $18.40 to $6.07 per million this year. Consumption tripled. Unit price is not a budget.

Infrastructure

Utilization
31%
Target band
60-70%
64
H100 equivalents
70/30
Reserved vs on-demand

$61K per month of idle GPU flagged for rightsizing. Industry average utilization is near 5%.

Unit economics

Cost per outcome
OutcomeAIBaseline
Resolved support ticket$0.85$9.40
Contract review$3.20$65.00
Code review, per PR$0.42$14.00
Call summarization$0.11$3.10
Merged PR, assisted$6.80n/a

Showback by business unit

Monthly · 13% unallocated

Spend Governor

Spend under enforced budget
93%
$86K
Overruns prevented QTD
42
Workflow budgets enforced
38 / 7 / 2
Alerts / rate limits / hard stops QTD
23/23
Models under rate limits

Budgets attach to workflows, owners, and models, and they enforce: alert, then rate-limit, then stop. Cost is a control, not a report.

Anomalies

Median 4h to root cause

Compliance · frameworks and evidence

Auditors ask for evidence. This page is the evidence.

One control set, mapped to every framework the board will ever ask about. Select a framework to see how the same controls answer it.

NIST AI RMF
85%
61 of 72 subcategories
ISO 42001
26/38
8 in progress · target Q2 2027
EU AI Act readiness
71%
7 high-risk systems
System cards current
83%
38 of 46 production
Guardrail blocks
214
Prompt injections / month

Framework crosswalk

Same controls, five lenses

KPMG Trusted AI pillars

Control coverage · 46 systems

NIST AI RMF functions

Govern wraps the cycle
Trustworthiness characteristics

Average 5.8 of 7 characteristics measured per production system: valid and reliable, safe, secure, accountable, explainable, privacy-enhanced, fair.

Adoption · Copilot, Claude, Cowork, and the humans

Licenses are bought. Adoption is earned.

4,600 paid seats across four platforms. 63% active against a 36% industry median. The gap between the two is training, champions, and workflow redesign.

Paid seats
4,600
Four platforms
Active users
2,890
63% utilization
Assisted hours
14.2K
Per month
Assisted value
$1.02M
Per month at $72/hr
AI literacy
56%
8,400 of 15,000 · EU Art. 4
Champions
46
Across 21 departments

Platform adoption board

28-day active vs licensed

Copilot usage depth

Power user: 15+ actions per week in 9 of the last 12 weeks. The 37% inactive block is the single largest recoverable spend item in the program.

Engineering

38%
Merged PRs AI-assisted
27%
Suggestion accept rate
+14%
PRs per dev-day
$6.80
Cost per merged PR

Accept rate sits inside the healthy 20-30% band. Top contributors feed the champions network.

Agentic work

130
Weekly agentic users
1,240
Cowork sessions / mo
92%
Background-task completion
8.4%
Escalation to human

Supervision ratio today: 1 human to 6 agents. The operating model is built to scale that safely, not to race it.

Enablement funnel

14 office-hour sessions QTD840 attendees

Training programs

Role-based paths

Knowledge center · frameworks, templates, answers

The library that keeps the whole program honest.

Every framework this control plane answers to, every template a team needs to ship, and every question we keep getting asked.

Framework library

What each one demands

The Hiday position

EnterpriseOS canon
01
EnterpriseOS
Seven layers, one operating model. Layer 01, Governance and Security, is the control plane. Layer 07, AI, is the next upgrade.
02
AIRB
One Architecture and Intelligence Review Board instead of four committees. Decision rights, SLAs, and an immutable log.
03
Agentic Harness
Six components every production agent must map to: Orchestration, Memory and Context, Tool Interface, Guardrails, Telemetry and Evals, Governance.

Insight feed

Attributed research behind every module

How to submit a use case

1
Describe it
Problem, value hypothesis, data involved, and the first named user with the date they touch it. No user, no build.
2
Score it
The feasibility engine scores seven dimensions and pre-classifies the EU AI Act tier while you type.
3
Route it
75 and above fast-tracks to the AIRB consent agenda. Lower scores get a standard review or a conditioned path with named gaps.
4
Track it
Approved cases appear in the catalog with owner, spend, value, and risk telemetry attached from day one.

Templates

Download and go

AI tool selection guide

Approved platforms

Questions we keep answering

FAQ

Need a human?

The control plane has operators
Governance desk

ai-governance@hiday.ai

Incidents

incidents@hiday.ai

Slack

#ai-command-center