The wrong primitives
Enterprise software was built on three primitives: state (what's true now), identity (who's acting), and activity (what they're doing). Every major platform — ERP, CRM, ITSM, IAM, BPM — is a sophisticated arrangement of these three ideas.
The decision was never one of them.
Decisions happen constantly inside organizations — pricing, approvals, exceptions, escalations, access grants, commitments. But they've never been a first-class object in the systems that run the enterprise. There's no structure for why a decision was valid, no mechanism for reusing proven judgment, and no way to connect a decision to the action it authorizes.
That gap has always been there. For decades, it was survivable. It isn't anymore.
Three tracks, same wall
The software industry hasn't ignored the problem. It has tried to solve it three times, from three directions, with escalating sophistication. Each track made real progress. None arrived.
The data track
If we assemble enough context, will better decisions follow? Data warehouses gave way to data lakes, then data catalogs, then data mesh, then knowledge graphs. Each generation solved a real limitation of the last. But more context doesn't produce a decision — it produces a bigger input to a decision that still has no structure, no closure criteria, and no connection to what it authorizes.
The identity track
If we know precisely who's acting, can we control what they do? RBAC gave way to ABAC, then federated identity, then zero trust, then fine-grained authorization. Each generation got more precise about who. But identity answers "who logged in." It has never answered "what can this person bind the organization to, right now, under current policy, in this specific scope." Access is not authority.
The process track
If we automate the flow of work, will governance follow? BPM gave way to case management, then RPA, then workflow orchestration, then agent frameworks. Each generation handled more complexity. But automating the sequence of steps is not the same as governing the judgment at the commitment boundary. Activity is not decision.
The ontology trap
There's a fourth attempt worth naming, because it sounds like the right answer and isn't.
Global ontologies — the idea that if we could agree on universal definitions, universal schemas, universal meaning — would dissolve the problem. If "revenue" meant the same thing everywhere, if "customer" had one definition, if policy could be expressed in a single canonical form, then decisions could be evaluated against a shared truth.
This is correct in theory. It is impossible in practice at enterprise scale.
"Active customer" legitimately means different things to Sales, Support, Finance, and Legal — not because anyone is wrong, but because they're answering different questions. "Revenue" changes meaning across accounting standards, jurisdictions, reporting periods, and business contexts. Organizations cannot achieve consensus on universal definitions, and even when they do, the definitions drift faster than they can be maintained.
Twenty years of failed MDM initiatives, abandoned semantic web projects, and stalled enterprise ontology programs are evidence enough. The approach is intellectually sound and operationally ruinous.
The alternative isn't no ontology and it isn't global ontology. It's local binding — meanings resolved at the point of decision, within a declared scope, with explicit transforms when decisions need to cross contexts. Precise semantics without the prerequisite of universal agreement.
The human compensating control
None of this was fatal — until now. And the reason it wasn't fatal has a name.
Humans were the load-bearing wall.
For decades, people filled the gap that the missing primitive created. They carried context the system couldn't. They knew when to ignore the rule, when to escalate the exception, when the policy had changed but the system hadn't caught up. They stitched together knowledge across emails, chat threads, spreadsheets, and institutional memory.
Entire industries formed to compensate. Business process outsourcing and shared services don't exist because humans love repetitive work — they exist because our systems can't handle the judgment that real complexity requires. Much of the work inside organizations exists not because the tasks are valuable, but because the infrastructure can't adapt to the world it operates in.
Two properties made this survivable: consequences and slowness. Consequences meant humans had skin in the game — a person who approved the wrong payment felt the weight of that decision. Slowness meant the gap between "decision made" and "action taken" was wide enough for review, correction, and intervention.
The death spiral
AI removes both properties simultaneously.
Agents can reason, draft, recommend, and increasingly act — but they don't carry consequences the way humans do. And they operate at a speed that compresses the gap between decision and commitment to near zero. The two things that made the missing primitive survivable disappear at the same time.
This isn't a gradual degradation. It's a structural failure mode.
Without decision infrastructure, organizations deploying AI face a specific spiral: agents make commitments that nobody can prove were authorized. Exceptions become rules because the pattern-matching treats historical accidents as precedent. Policy changes silently because the system never pinned which rules were in force when. Authority drifts because the agent learned that most requests get approved, so it approves most requests.
The symptoms are already visible: AI takes actions no one technically authorized. Compliance becomes theater — screenshots and transcripts replace an actual chain of accountability. Humans can't explain why something was allowed; you get a post-hoc narrative, not a referenceable basis.
The AI isn't trapped. The infrastructure is.
The missing primitive pattern
The biggest platform shifts in enterprise technology didn't come from better implementations of existing abstractions. They came from recognizing that an abstraction was missing entirely.
Compute
Compute was locked inside physical servers, managed by hand, provisioned in weeks. AWS didn't build a better server — it made compute a primitive.
Payments
Payments were tangled in merchant accounts, gateway integrations, and bank relationships. Stripe didn't build a better gateway — it made payments a primitive.
Data
Data was trapped in rigid schemas owned by applications. Snowflake didn't build a better warehouse — it separated storage from compute and made data a primitive.
Each of these companies identified a missing abstraction, built the infrastructure layer for it, and created a category. And here's what matters: when the AI wave arrived, these infrastructure companies didn't shrink. They grew. AI needs more compute, more data infrastructure, more payment processing. The primitive layer becomes more valuable when everything above it gets commoditized.
Technology waves amplify infrastructure and commoditize everything above it.
The decision primitive is missing from that stack. Enterprise software has infrastructure for state, identity, activity, compute, payments, and data. It has no infrastructure for judgment — for making decisions well-formed, reusing them when they apply, and controlling what they authorize.
That's what Ordinant builds. Decision Infrastructure.
What Decision Infrastructure means
Ordinant is software that makes organizational judgment well-formed, reusable, and safe to execute.
Decide
Every decision gets a clear procedure: what evidence counts, which rules apply right now, who has authority, what has to be true before anyone commits. If required context can't be resolved, the system fails closed — it routes for review rather than guessing. The quality of the initial decision is everything.
Reuse
Well-formed decisions become precedent. When the same situation recurs, the answer is already there — matched by structural applicability, not similarity. Judgment compounds across the organization. Re-litigation becomes lookup.
Execute
A closed decision issues a specific, time-limited right to act. Execution produces proof that the action matched what was authorized. Authorization derives from the decision, not from identity. No commitment without closure. No action without a warrant.
The point isn't governance. Governance is a consequence, not the purpose. The point is speed — the speed that comes from decisions you can trust, judgment you can reuse, and actions you can prove.
Why now
Three forces are converging.
AI is removing the compensating control
Every agent deployment accelerates the timeline. The gap between "decision made" and "irreversible commitment" is compressing toward zero. The human judgment that filled the gap can't scale to match.
Regulation is arriving
The EU AI Act, APRA CPS 230, emerging US state-level requirements — regulators are demanding explainability, traceability, and human oversight of AI-augmented decisions. These are feature requests for Decision Infrastructure.
The decision surface is expanding
When AI makes reasoning cheap, organizations won't govern the same 5% of decisions they govern today. They'll govern all of them — because they can, and because they have to. Decision Infrastructure isn't automating a cost center. It's enabling a capability that was previously too expensive to exist.
The question isn't whether organizations will need Decision Infrastructure. It's whether they'll build it while the commitment boundary is still visible — before agents cross it without anyone noticing.