Applied AI Venture Studio — Investment Thesis
Four categories survive the shift and compound: infrastructure that frontier AI runs on, agent-callable surfaces and trust layers, specialised creation surfaces, and regulated / sovereign verticals frontier AI cannot reach. ntelio invests exclusively in those four.
The frontier-model capability jump (Claude Mythos / Opus 4.7 class systems, mature agent frameworks like OpenClaw) is not killing software — the interface has become the agent. The GUI-as-primary-interface for routine digital tasks is collapsing into agents. Four categories survive and compound: infrastructure (databases, payment rails, auth, ingestion pipelines), agent-callable surfaces (MCP servers, tool APIs, trust/verification layers), specialised creation surfaces where the UI is the capability, and regulated / sovereign workloads that frontier cloud models cannot serve. ntelio invests exclusively in those four. We build vertical AI products — some rooted in our own core experience, others co-created with enterprise partners — with a bias toward those that generalise into our platform and engineering practice. We ship only what survives the layer-split — never the dashboard-and-form class the agent era is absorbing.
The companies who own the system of record win.
The companies who sold the view of it lose.
We hold a clear, falsifiable view on which software categories survive frontier AI and which do not. Capital allocation flows from this view.
Anything that is essentially database + form + report: horizontal SaaS (CRMs, project management, HR, help desks, generic dashboards), bespoke internal tooling, most workflow apps. The UI was a prosthetic for the real task; agents remove the prosthetic.
Databases, payments, auth, storage, ingestion pipelines, observability. Every agent task is N backend calls; the substrate gets more valuable.
Software built specifically to be called by agents: MCP servers, agent-addressable APIs with semantics / permissions / auditability, verification systems that enforce scope and rollback when agents hallucinate. A category being born — barely anyone is building it.
Figma-category, not CRM-category. The UI is the capability — phone camera + barcode scanner + on-device AI for a merchant standing in their shop, a curator’s voice and avatar in a journalist production pipeline, a gallery’s curatorial framing of a work. Agents help inside these surfaces; they do not replace them.
On-prem, sovereign-cloud, audit-ready, repeatable, compliance-bound. Frontier models are cloud-only, probabilistic, and unverifiable. Government, regulated finance, healthcare, defence, public administration. Frontier AI cannot enter these workloads on the terms required.
Not “will people ask Claude for digital tasks” — they will. The question is:
What is the minimum software footprint a business needs when its employees and its customers mostly work through an agent?
The system of record. The policies. The payment and fulfilment rails. The regulated workflows. The specialised creation tools. Much less of the dashboard-and-form wrapper.
We invest in the answer.
Every ntelio product and every partner engagement we accept must sit in at least one of four buckets. Most of our products sit in two or three.
System-of-record, ingestion, payment / fulfilment rails, multi-channel operational moat, persistent claim graph — the substrate agents call into.
MCP servers exposing each product as an agent-addressable endpoint; corroboration and credibility scoring as a verification primitive of the agent era.
Figma-category UIs where the capability is the interaction. Agents help inside; they do not replace.
Workloads where frontier cloud models are structurally disqualified by privacy, compliance, audit, repeatability or sovereignty requirements.
We will not invest in a product that does not sit in at least one of these buckets, regardless of revenue or near-term traction. We preferentially invest in products that sit in two or more — these are the products that compound.
The same five questions decide whether a new engagement becomes a new ntelio product, an extension of an existing one, or a paid engagement we deliver and walk away from. A clean yes through all five is what triggers a new investment.
A budget line item with an internal owner. Not a hypothesis. Not a horizontal SaaS itch.
A — infrastructure. B — agent-callable / trust. C — specialised creation. D — regulated / sovereign.
Will patterns generalise back into the substrate? Will the next engagement be cheaper from an engineering, operational complexity and costs perspectives because we did this one?
Is the engagement also a market signal for a portfolio addition, beyond the partner who scoped it?
Substrate, surface, sovereignty, or curatorial relationship. Not a system prompt, not a UX wrapper.
A no at any gate either narrows the engagement to delivery only, or sends the partner back to PRISM for a different scope. We have made four investments through this engine. The fifth will be born the same way.
We’ll send the long-form thesis, an updated portfolio brief, and an introduction to the founder best placed to answer your specific questions.
investors@ntelio.ai → partners@ntelio.ai →