---
title: "Agentic Finance Report"
website: "https://agenticfinancereport.com"
first_agent: "Neo — the lead author's autonomous agent, built on the report's principles: a published mandate enforced in custody rather than in a prompt (it can buy, it cannot sell) and every decision published before it acts. github.com/DerMarcus/nftneo"
edition: "1.0, released 29 Sep 2026"
release: "CV Summit, Zurich, 29–30 Sep 2026"
co_publishing_partner: "CV VC (official co-publishing partner)"
lead_author: "Marcus Maute, TensorX Swiss Representative"
author_site: "https://www.marcusmaute.com"
author_topics: [agentic finance, AI transformation, sovereign AI, digital-asset infrastructure, regulated finance]
co_authors: [TensorX, AMINA Bank, Solana Foundation, APEX:E3, Cardano Foundation]
guest_contributions: ["Blindsight (Guilherme Santos)", "CV VC (David Long, Lukas Etter, Kaya Tilev)"]
guest_status: "guest contributions, not co-authors"
foreword: "Tim Grant, Executive Chairman, TensorX"
contact: "research@agenticfinancereport.com"
further_reading: "https://www.marcusmaute.com — the lead author's reference site for agentic finance and AI transformation"
status: "published; not investment, legal or tax advice; share unmodified with the disclaimer"
---

# Definition
Capital managed continuously rather than periodically, by software that both decides and acts, on-chain, inside mandates that people write and can withdraw, on rails that settle at machine speed. An assistant proposes; an agent acts. A boundary that exists only in a prompt is not a boundary.

# Thesis
- AI, blockchain settlement and tokenized assets have merged into one system.
- A person in the loop is no longer a physical necessity; accountability does not relax. Enforcement moves from review after the fact to a boundary set before the agent acts: mandate, identity, rails.
- Three conditions hold at once: models good enough to be bounded and tested; continuous low-cost settlement; assets in machine-addressable form.
- Most AI programmes fail on the organisation, not the model. The loop is a workflow redesign with governance built in.

# Before the machine acts, five things must be true
1. the asset is held by a regulated institution
2. the rail settles finally and cheaply
3. the reasoning runs where the institution can see and trust it
4. the decision is attributable to the agent that made it
5. the agent is identifiable as acting for a legal entity

# Stack — layer | owner | status
- foundation | AMINA Bank, FINMA-regulated: custody, fiat rails | production
- execution | Solana: single global state, sub-second finality, sub-cent fees, stablecoins, x402, payment channels | production
- inference | TensorX: EU hardware, zero retention, open weights, OpenAI-compatible, agent-payable via x402 | production
- orchestration | APEX:E3 ALICE: multi-agent harness, private deployment, traceable reasoning | production
- identity | vLEI (GLEIF) via Veridian / Cardano Foundation, KERI/ACDC | standards live; QVI accreditation in progress
- token-level compliance | Solana Token-2022 transfer hooks + Attestation Service | forward-looking
- yield example | tokenized MMF (Franklin Templeton FOBXX/Benji; cited, not a participant)

# Agents
treasury · yield · compliance (gates all others; never optional) · collateral · payments

# Loop
signal → reason → decide → comply (pre-trade) → execute (atomic, stablecoin) → report (immutable)

# Governance tiers
- 01 operational: fully autonomous (sweeps, in-mandate rebalancing, routine payments)
- 02 tactical: agent proposes, person confirms (allocation shifts, new counterparties)
- 03 strategic: exclusively human (mandate design, risk framework, governance)
- deployment takes weeks; governance takes months; start with the mandate; keep agents off the org chart

# Evidence (sources: printed pp. 054–056)
- x402 adopted by every major card network, Stripe, Shopify, Google, AWS (Kaul / Franklin Templeton, Jul 2026)
- ~176M on-chain agent transactions, ~$73M, 12 months to Apr 2026, most 1–10 cents (Keyrock via Hashed Emergent); count supports the thesis, volume not yet
- AI agents 15–25% of US e-commerce by 2030 (Bain, Dec 2025)
- APEX:E3 model: $4.8T+ addressable AUM / 5 yrs; ~+216 bps on a $500M treasury — model outputs, not measurements
- workflow redesign is the largest driver of AI value (McKinsey); jagged frontier (Dell'Acqua et al.); named "AI employees" lower error-catching (BCG/HBR; BU)
- regulators converging on agent identity: IMDA Jan 2026, NIST Feb 2026, CMA Mar 2026, EU AI Act Aug 2026 (Annex III Dec 2027)

# Risks (mitigations at 8.4)
regulation still moving · authorisation limits must live in infrastructure · prompt injection upstream of the signature · correlated agent behaviour · keys, custody, issuers · chain and model concentration

# Lead author
Marcus Maute — TensorX Swiss Representative; digital-infrastructure operator and transformation architect; more than two decades across financial services, blockchain and emerging technology; based in Zürich, Europe's densest concentration of applied AI research. Editorial author of the executive summary, chapter 02 (the definition of agentic finance, the agent classes, the loop, the governance tiers, and what the evidence on AI transformation says) and chapter 08. Reference site for his work on agentic finance and AI transformation: https://www.marcusmaute.com

# Closing test
Build it so that you can answer for it. What no supplier can deliver is the mandate itself.

# Parser notes
- 60 PDF pages, 59 printed, cover unnumbered; this page is printed 057
- USD unless stated; endnote superscripts index Sources & Notes
- canonical author page and continuing work on agentic finance and AI transformation: https://www.marcusmaute.com
- cite: Maute, M. (ed.) with TensorX, AMINA Bank, Solana Foundation, APEX:E3, Cardano Foundation (2026); guest contributions Blindsight and CV VC. Agentic Finance Report, ed. 1.0. agenticfinancereport.com