Announcing Viable Systems

Today we are announcing Viable Systems, an Applied AI research lab building the missing infrastructure that organisations need to actually put AI to work.

Our thesis is simple: the gap between what AI can do and what it delivers inside real organisations isn't a model problem, but rather an infrastructure problem. The data, structures and primitives that AI needs to operate inside of your company don't exist yet. We are building them.

The problem

AI capabilities have improved dramatically, but organisational productivity has not. Between $30 and $40 billion has been poured into enterprise AI pilots^{1MIT NANDA — 95% of AI pilots deliver no measurable P&L impact}, yet 95% deliver no measurable P&L impact^{1MIT NANDA — 95% of AI pilots deliver no measurable P&L impact} and only 6% of organisations report meaningful bottom-line results^{2McKinsey — Only 6% qualify as AI high performers}. Benchmarks on complex reasoning, coding and knowledge work improve quarter over quarter. Reported gains inside actual companies remain close to zero.

This pattern is not new. Every general purpose technology; steam, electricity, computing, went through a prolonged phase where the technology existed but the infrastructure required to exploit it did not. Economists call this the Productivity J-curve^{3Brynjolfsson et al. — The Productivity J-Curve}: heavy investment, flat returns, then an inflection once the complementary infrastructure catches up. There is recent evidence the curve is starting to bend, Brynjolfsson estimates US productivity jumped roughly 2.7% in 2025, nearly double the prior decade's average, and attributes the shift to a small cohort who built the surrounding systems, not just deployed the models^{4Brynjolfsson — US productivity jumped ~2.7% in 2025}. The implication is clear: infrastructure is what accelerates the transition, and that is what we are focused on.

The common take is that AI is not yet ready. We believe the opposite. AI is ready, organisations are not.

What's missing

Every tool in a company's stack captures state; what happened, what exists, what changed. We think of this as the state clock. Your CRM knows you lost a deal, your project tool knows a deadline has slipped.

What none of these systems capture is the why. What was this decision made? What context shaped it? What precedent led to it? This is the event clock, the reasoning behind the state and it barely exists anywhere in structured form. It lives in people's heads, in Slack threads or in meetings that leave no trace.

We call this the invisible organisation. It's the layer of context, reasoning and institutional memory that drives every meaningful decision but is never captured and therefore completely inaccessible to machines.

This is the actual bottleneck to useful AI systems, not model capability, but missing data^{5NVIDIA — 48% cite data issues as #1 challenge to scaling AI}.

What we are building

Two things, at the infrastructure level.

First: the data layer. We're building systems to capture decision traces, structured records of why things happened, not just what. These traces accumulate into a context graph that represents how people, decisions, and knowledge relate to each other. Over time, this becomes an organisational world model, a living representation of how a company works that both humans and AI can query and act on.

Second: the substrate. We believe organisations are heading toward what we call the programmable organisation, where processes, roles, and workflows are defined in data and code rather than scattered across disconnected tools and undocumented habits. If an organisation isn't machine-readable, it can't be machine-augmented.

We're not building another AI feature. We're building the layer underneath, the one that makes AI actually useful inside the messy reality of how people work via systems that evolve and compound in value over time.

Why this, why now

The buy-vs-build calculus for software has flipped. A large share of business software is mostly records, forms, workflow, permissions and dashboards — useful, but much easier to rebuild than it used to be. Satya Nadella put it plainly: "SaaS applications are essentially CRUD databases with a bunch of business logic. The business logic is all going to AI agents"^{6Nadella — SaaS will collapse in the agent era}. Klarna proved the point by consolidating 1,200 SaaS tools into a single knowledge graph^{7Klarna — Consolidated 1,200 SaaS tools into a Neo4j knowledge graph}, but it took thousands of engineers.

The data locked inside those platforms, your decisions, your context, your institutional knowledge, is more valuable than the software housing it. Organisations are going to want control of that data, and the tools to make it legible to intelligent systems.

We're early, and we're honest about that. We're working with design partners, running multiple experiments, and converging on the right product surface. What we're certain about is the thesis: closing the gap between AI capability and AI impact requires new infrastructure, not just better models.

Dennis Yurkevich
Founder & CTO, Viable Systems LTD