Mandate · Ledger · Scorecard · Lifecycle
The accountability plane for AI agents
ThetaLayer gives every AI agent in your enterprise a mandate, a ledger and a scorecard — across every vendor and every system of record. See what each agent did, what it cost, and whether it was worth it.
The gap
AI agents now spend real money. Almost nobody can say what they did.
Agents act across email, spreadsheets, CRM and systems of record. The ERP doesn't know an agent was involved; the agent runtime doesn't know what actually changed. You are paying for a digital workforce you can't yet manage.
of organizations see any EBIT impact from AI — flat on 2025
McKinsey, The state of AI in 2026
already limit AI use because of token and running costs
McKinsey, The state of AI in 2026
have a mature governance model for autonomous agents
Deloitte, State of AI in the Enterprise 2026
now have a Chief AI Officer — the owner of AI returns
IBM Institute for Business Value, CEO Study 2026
One plane · four pillars
Manage agents like a workforce — across every platform
A job description, a timesheet, a performance review and an HR record for every agent. One plane above all your agent vendors, anchored in the systems of record below them.
Pillar 01 · Mandate
A job description every agent can be held to
Define what each agent may do — tools, data, amount thresholds, spend caps, autonomy level and identity rules. Mandates are versioned, and every change is re-checked against the agent's full history, so you see the impact of a rule before anyone enforces it.
- Approval thresholds and sensitive-field rules
- Per-run and monthly spend caps
- Autonomy levels earned on track record
- 100% passAllowed tools9 tools
- 4 failedApproval above> €10,000
- 3 failedBank details need approvalbank_account
- 22 failedSpend cap per run€2.50
- passActs under own identityrequired
Pillar 02 · Ledger
What each agent actually did — on both sides of the record
Every run is reconstructed end to end: the trigger, each tool call, the approvals, the payment. Each action an agent claims is matched to independent evidence in the system of record, with an explicit link method and confidence. Gaps and contradictions become exceptions, not footnotes.
- Runtime claims vs. system-of-record changes
- Identity of the agent, the person and the credential used
- Late evidence — approvals, payments — linked automatically
- 07:49:23Logged invoice in AP tracker · AgentConfirmed
- 07:49:26Supplier email opened · Mailbox auditRecord
- 07:49:58Posted invoice to ERP · AgentContradicted
- 11:20:04Approved bank change · ApproverPerson
- Tue 06:00Invoice paid · ERP paymentsRecord
Pillar 03 · Scorecard
Cost per outcome, comparable across every vendor
Measure what each agent produced for what it cost — cost per completed workflow, straight-through and takeover rates, evidence coverage and mandate compliance — by agent, version, team and task type. Every metric is defined, versioned and traceable to its source records.
- Version regressions caught in days, not quarters
- Routing decisions per task type, not per vendor
- Allocated platform fees labeled as allocated
Pillar 04 · Lifecycle
An owner, a review date and a decision
Treat agents like a workforce. Each agent has a named owner, an autonomy level and a quarterly review backed by its scorecard and exception history — so renewal, re-scoping and retirement are decisions, not defaults.
- Named owners and review cadences
- Autonomy raised or lowered on evidence
- Retire agents that don't earn their cost
Two-sided evidence
An agent's log is a claim. The system of record is the evidence.
ThetaLayer matches every action an agent reports against what the ERP, CRM, HRIS, mailbox or approvals system actually recorded. When they disagree — a timeout that really posted, a success that never landed, a change nobody's agent made — you see it, with the evidence attached.
- Direct reference
- 95%
- The agent returned the record's ID
- Business key
- 85%
- Hashed vendor + invoice number
- Time window
- 70%
- Identity and timing only
Every link carries its method and confidence. Uncertainty is shown, never hidden.
What you can decide
Every agent decision, backed by cost per outcome
Built around the decisions a Chief AI Officer and CFO actually make — not around dashboards.
Spend
- Route each task to the cheapest agent that does it well
- Set spend caps and charge costs back to teams
Value
- Scale, renew or retire each agent at quarterly review
- Decide build vs. buy on measured cost per outcome
Control
- Raise or lower each agent's autonomy on its track record
- Export evidence for internal audit, DORA and the EU AI Act
Leaders already measure this: more than 60% of “future-built” companies rigorously track AI value, against 17% of stagnating ones. — BCG, The Widening AI Value Gap (Sep 2025)
Your data stays yours
Read-only. Minimized. Runs where your data lives.
Credentials and raw records never leave your network. Only minimized, schema-validated evidence reaches the control plane.
Security & trustRuns in your environment with read-only access to agent runtimes and audit logs. Your credentials stay with you.
IBANs, invoice numbers and message IDs are hashed; email bodies, subjects and cell values are dropped before export.
Frankfurt by default, with tenant isolation at every layer. Other regions on request.
Start read-only. Add enforcement only at the tool boundaries you choose, when you're ready.
Solutions
Start where agents already touch money and records
Invoice intake, vendor master and payment agents across email, spreadsheets and ERP.
Lead qualification, CRM updates and service agents, measured per resolved case.
Onboarding and HRIS agents handling personal data under strict mandates.
Coding agents measured by cost per merged change and revert rate.
Audit-ready agent evidence for DORA, the EU AI Act and internal audit.
If an agent writes to a system with an audit log, ThetaLayer can account for it.
Sits alongside, not instead of
Neutral by design
Vendor control planes govern agents best inside their own ecosystems. Security tools protect the perimeter. ThetaLayer owns the link between what an agent costs and what it produces — across all of them.
Microsoft Agent 365, ServiceNow AI Control Tower and platform consoles catalog and control agents inside their ecosystems.
Protect against prompt injection, data leakage and risky permissions — and sell to the CISO.
Trace latency, errors and token usage inside one application.
The pilot
Six weeks. Read-only. Success criteria agreed up front.
One workflow, one agent runtime, two or three systems of record. No changes to your agents.
Book a Demo
See your agents through one plane.
Book a 30-minute walkthrough. We'll reconstruct a real workflow end to end — mandates, evidence and cost per outcome — and scope a six-week, read-only pilot around your agents.