Initiative Studio

From intent to a verification program in one conversation

Say what must be true — “no USB exfiltration,” “every admin behind MFA,” “billing stays recoverable.” The Architect decomposes it into conditions, signals, probes, and owners — scoring from the first poll.

Try Studio See how it works

Panaptico Studio

USB Exfiltration Prevention

Composing · 12 min ago

Okta Admin Assurance

Yesterday

Japan Office Zero Trust

2 days ago

Backup & Restore Proof

Mar 18

Exec Protection Coverage

Mar 15

The Architect

USB EXFILTRATION PREVENTION

Create Initiative from Chat

We need to prevent USB exfiltration across the fleet. Device control exists in CrowdStrike, DLP lives in Purview — but nobody can prove a write actually gets blocked.

Querying the live model · Resolving device populations · Checking control coverage

I found 3 enforcement points that claim to cover removable media. Here's what the live model shows:

Enforcement Points

Grounded

CrowdStrike FalconDevice Control policy

13,588 endpoints

Active

Microsoft PurviewEndpoint DLP

11,904 endpoints

Partial

JamfmacOS media control

2,412 endpoints

Unverified

412 endpoints carry no device-control policy at all — and no probe has ever attempted a blocked write

Before I compose the program, one fork: deny-by-default with owned exceptions, or allow-with-audit? Deny-by-default is provable — the probe attempts the write and watches it fail. Allow-with-audit only gives you evidence after the exfiltration.

Declare what must be true...

Programs fail in the gap between declared and actual

Policies live in documents nobody checks against reality

Controls get attested annually and drift daily

The people who declare the intent aren’t the ones watching the systems

Studio collapses intent-to-running-verification from weeks to minutes.

How It Works

The Architect

OKTA → ENTRA MIGRATION

We need to migrate 4,000 users from Okta to Entra ID. Hybrid AD with on-prem DCs in 3 regions, about 120 SAML/OIDC apps.

Evaluating identity federation patterns · Checking hybrid AD implications

Before I scope this, a critical decision. Three migration approaches — each with real tradeoffs:

Approach	Risk	Timeline	Downtime
Big-bang cutover	Critical	2–3 weeks	4–8 hrs
Phased by app groupRecommended	Low	6–8 weeks	< 30 min / wave
Parallel run + cutover	Medium	4–5 weeks	~2 hrs

Grounded in:

Okta

Entra

Runbooks

With 120 apps and hybrid AD in 3 regions, big-bang is asking for a P1 at 2am. I'd recommend phased — each wave verified against its baseline before the next one moves.

Describe the intent. The Architect shapes the program.

Open-ended conversation with the Architect. Best when you know what must be true but not how to prove it. It asks the highest-signal clarifying questions, surfaces the forks that decide whether a control is provable, and progressively shapes the answer into a verification program.

Opinionated — pushes back on unprovable designs like a senior engineer would

Grounded in your live model, not generic best practice

Renders inline diagrams, tables, populations, and coverage maps

Streams token-by-token with visible chain of thought

The Architect

An opinionated verification partner, not a chatbot

The Architect is grounded in your live model, prior initiatives, vendor docs, and compliance requirements. It pushes back on unprovable designs, forces the forks that decide what can be verified, and refuses to invent state it hasn't observed.

Architecture — Target State

SA generated

WIF eliminates long-lived credentials. Services authenticate via OIDC → STS AssumeRole → Secrets Manager. All access logged to CloudTrail with 90-day retention.

render_component

Inline UI blocks — checklists, timelines, charts, stat rows — so structured info renders as interactive UI instead of prose.

sketch_diagram

Architecture diagrams generated on-demand. When a visual communicates the idea faster than words, the Architect draws it — including on the first turn.

query_model

Retrieval against your live model and knowledge base so answers are grounded in observed state, not generic best practice.

finalize_initiative

Converts the agreed shape into a running initiative — conditions, signals, and probes scoring against the live estate from the first poll.

The Output

A running program, not a doc

Every Architect-authored item carries provenance — why this condition, which observed state grounded it, and a confidence level that shapes downstream auto-acceptance vs. human-review gates.

Intent

Declared Claims

Populations

Success Thresholds

Non-Goals

Model

Entities in Scope

Identity Links

Enforcement Points

Data Sources

Verification

Signals

Probes

Cadences

Primitives

Guardrails

Exceptions & Expiry

Owners

Approvals

Escalation

Provenance

Rationale

Evidence References

Confidence

Open Questions

InitiativeContinuous Verification

The initiative goes live the moment it's composed — scored on every poll, with gaps, exceptions, and evidence in one place.

Who it's for

Built for everyone who has to say “it works”

Platform Engineers & SREs

The people who declare intent and consume verification daily — deployment gates, drift, readiness, and the proof that closes the ticket.

VPs of Engineering & Heads of Platform

Deployment verification and change safety with ROI provable in weeks — deploy time, change-failure rate, MTTR.

CISOs & CIOs

Initiatives translate engineering reality into governance language — compliance programs, risk initiatives, board-level posture.

What makes it different

Not another AI chatbot

Grounded, not generic

The Architect is tied to your live model and refuses to hallucinate vendor behavior or invent state it hasn't observed.

Opinionated

Pushes back on unprovable designs the way a senior engineer would, instead of dutifully answering whatever you ask.

Structured, not prose

The output is a machine-readable verification program with provenance — not a doc that needs human re-interpretation.

Live from day one

Initiatives start scoring the moment they're composed. The intent-to-proof seam is the product.

Declare it once. Verify it forever.

Describe what must be true. Get a running verification program. Watch the proof arrive.

Try Studio Book a Demo