Gold — admissible / authority Blue — review required Red — refused / contained Controlled review · patent pending
Execution Control Infrastructure

AI can act.
What decides whether it should?

ThePraesidium.ai builds the runtime layer between AI capability and real-world consequence — the category we call Execution Control Infrastructure. DynamicDesk is the product that puts it to work: see what AI proposes, decide whether it’s admissible, and keep a replayable proof.

DECISION VIEW Example ↗
AI proposes: issue a supplier payment.
Cash impact exceeds the approval threshold; a vendor's bank details changed within 48 hours.
ApproveEscalateContainRefuse

DECISION  CONTAIN
REASON  POLICY_LIMIT_EXCEEDED
▶ Run it live
Where we are Controlled review stage Open to design partners — defense · fintech · regulated ops Founder-built · patent pending Demonstrated end-to-end in controlled mode
01 / THESIS

Monitoring is not control. Approval is not admissibility.

Dashboards tell you what happened. Logs tell you after the fact. Neither stands at the moment an AI action becomes consequence. Execution Control Infrastructure is the layer that decides — before the action commits — whether it is allowed to become real.

Capability is no longer the bottleneck. The bottleneck is consequence: the moment an autonomous system can move money, send records, change state, or take an action no one reviewed. ECI governs that moment.

"The moment capability becomes consequence is the moment that needs an authority."
02 / WHY A DEDICATED LAYER

Why not guardrails, IAM, or observability?

Each adjacent tool solves part of the problem. None stands at the commit boundary, across vendors, with a reviewable proof. That gap is the category.

AdjacentWhat it doesWhat it doesn't
Model guardrailsFilter prompts and outputs inside one model.Don't govern the action across tools — or prove it.
Cloud IAMStatic identity and permissions for infrastructure.Don't reason about a specific AI action's admissibility at runtime.
Observability & evalTell you what happened, after the fact.Don't decide what is allowed before consequence.
OrchestrationRun the steps an agent takes.Don't hold the commit boundary or preserve proof.

Execution Control Infrastructure is cross-vendor, action-level, and proof-first: it sits at the moment of action, resolves admissibility against authority and policy, and leaves a replayable record. That is the layer none of the above own.

03 / SHIFT

Every platform era needed a control layer. AI is no exception.

EraThe control layer that made it safe to operate
DatabasesTransactions & ACIDCorrectness, consistency, and recoverability of state — by construction.
InternetTLS & certificate authorityA trust layer so parties could transact without trusting the wire.
CloudIAM & policyIdentity and permission boundaries for elastic, shared infrastructure.
AIExecution Control InfrastructureAdmissibility, authority, and proof at the moment AI acts. The missing layer.

Each era's control layer was invisible until it was indispensable. AI is at that threshold now.

04 / THE DECISION

Four outcomes at the commit boundary — each with a reason.

APPROVEWithin authority, reversible, and evidenced — the action proceeds with a record.
ESCALATEValid but beyond standing authority — routed to the right human, not guessed.
CONTAINPartially admissible — the safe part proceeds; the risky part is held at the boundary.
REFUSEFails a hard constraint — refused, with the reason preserved for review.

Every outcome carries a structured reason code that makes the decision defensible — and replayable later. See it decide across ten operating worlds ↗

05 / THE STACK

One control layer. Start where the risk is visible.

SurfaceWhat it does
Agent Flight RecorderRecords & replays AI-mediated actions · START HEREThe wedge: capture what AI proposed and did, with a replayable record — before you change anything.
Approval GatewayDecides allow / escalate / contain / refuseThe commit boundary where admissibility is resolved against authority, policy, and state.
DynamicDeskThe operating surface · MVPWhere operators and executives see proposals, evidence, replay, and decisions — live.
Proof SurfaceCreates reviewable receiptsEvidence packs, integrity checks, and replay — so any decision can be reconstructed.
06 / PROOF CHAIN

Every action flows through one accountable chain.

Proposed actionAdmissibilityApproval GatewayGoverned executionEvidenceReplayProof

So you can always answer four questions: what was proposed, whether it was allowed, what evidence exists, and how to replay the decision later.

07 / FOUNDATIONS

The category, the substrate, and the platform model.

Execution Control Infrastructure

The category: the runtime layer that resolves authority, policy, state, and proof before AI acts. What is ECI ↗

Enterprise Memory

Governed, durable memory as the substrate decisions run on — deployed, and the hard part to copy. See Enterprise Memory ↗

Intelligent SaaS (iSaaS)

The platform model that runs on ECI: software that remembers, adapts, and acts under control. See iSaaS ↗

08 / WHY NOW

A new control layer, at the moment the category forms.

Regulation is tightening, autonomous systems are reaching real operations, and there is no accountable layer at the moment of action. That gap is the opening for Execution Control Infrastructure — and the wedge is already in hand: the Agent Flight Recorder, sold through a Recorder Review.

  • Wedge-led GTM: Recorder Review → Approval Gateway → DynamicDesk.
  • A category with a clear primitive: admissibility at the commit boundary.
  • Honest maturity: controlled review, proof shown in controlled mode, patent pending.
PROOF

Proof you can see — not just claims.

Every controlled run leaves a structured evidence receipt: a decision, a reason code, an integrity check, and a replayable record. A sample from controlled mode:

EVIDENCE RECEIPTSAMPLE · CONTROLLED MODE
DECISIONCONTAIN — PARTIAL
REASON CODEPOLICY_LIMIT_EXCEEDED
EVIDENCE PACKev_fin_4029
REPLAY IDrp_fin_4029
DIVERGENCEnone
INTEGRITYPASS · REDACTION ACTIVE · REPLAY AVAILABLE
09 / PROOF OF BUILD

Current controlled runtime proof.

Built and merged. Each surface runs draft-only, with proof — nothing is sent, deployed, or executed autonomously.

Founder Command Surface
Unified cockpit over rooms, approvals, evidence, memory candidates, Atlas changes, F3 actions, and blocked risks.
Proof + Approval Spine
Normalizes governed workflows into decision object, authority route, approval decay, evidence binding, replay pack, truth label, and next human action.
Verified Room
A real counterparty correspondence — bound, redacted, hydrated, drafted, locally approved, and preserved with evidence and replay.
Controlled Surfaces
F3 Growth Desk · Customer-One Memory Lane · Atlas Real-Time · Coder Room L3 · Social Proposal Room · Local Draft Harness.

Boundary — controlled mode only. Not production. No live connector execution. No autonomous action. No sending.

From demonstration to controlled review

See it decide. Then bring your own decision.