Skip to main contentSkip to navigation
Veldarium
Work with us

Essay / AI-native shipbuilding

America should be able to build hard things

A founder’s thesis on rebuilding vessel-delivery capability, why VeldBuild exists, and what Veldarium has—and has not—earned so far.

1,186 words · ~5 minChristopher Azar-Brandes

What the public record says

The public record is serious enough without turning it into theater. In April 2026 testimony, the U.S. Government Accountability Office said Navy and Coast Guard shipbuilding programs had consistently fallen short of expectations over two decades and, taken together, were billions of dollars over cost and years behind schedule. GAO did not offer one magic fix. It pointed to a combination of design practice, program management, workforce, infrastructure, suppliers, and investment discipline.

GAO’s 2025 work on the private ship industrial base found that infrastructure and workforce constraints were making Navy goals difficult to achieve. It also found gaps in coordination, performance measures, and the strategy governing billions of dollars in industrial-base support. A separate 2024 review found that leading shipbuilders put more emphasis on mature designs, meaningful readiness measures, and shorter, more predictable design and construction cycles.

Those are government findings about government programs and the industrial base. They do not validate Veldarium, VeldBuild, or any particular technical architecture. They establish the scale and persistence of the problem we have chosen to work on.

What I believe it means

My reading is that American shipbuilding is not suffering from one missing dashboard, one weak model, or one group of people who simply need to work harder. Vessel delivery is a system outcome. Design maturity, material, supplier information, work packaging, production sequence, inspection, facilities, machines, and human decisions all meet on the same physical object. When those parts lose agreement, the cost appears later as waiting, disruption, rework, and schedule movement.

That is founder interpretation, not a finding GAO made for us. It is also why I do not think generic software positioning is useful here. Software matters only if it helps people build with a truer view of the vessel, preserves authority when records disagree, and makes the next piece of physical work easier to establish and execute.

Why VeldBuild exists

VeldBuild exists because Veldarium ultimately wants to become exceptionally capable at building vessels. The product is in development as the controlled build-state system we believe that journey requires: a way to connect requirements, design objects, material, work packages, production state, quality evidence, changes, and named human authority around the vessel being built.

Our company thesis is that better vessel delivery comes from making those relationships explicit and usable—not from placing a conversational layer over disconnected records. AI can help compare, explain, plan, and surface anomalies. It should not silently invent configuration, declare material available, waive an inspection hold point, or impersonate the person authorized to release work.

That thesis will be tested by whether it reduces uncertainty and friction in real build work. Until then, it remains a company thesis rather than a proven shipyard outcome.

People, machines, and responsibility

I do not think the answer is replacing shipbuilders with AI. Welders, pipefitters, electricians, machinists, shipfitters, riggers, inspectors, planners, supervisors, and engineers carry physical judgment that a model does not. A serious system should make their knowledge easier to use and their work easier to prepare, not treat them as friction to automate away.

I do want machines doing more where they have a real advantage: selected dangerous, repetitive, heavy, or precision tasks, with task-specific risk assessment, safeguards, validation, training, and maintenance. That is future ambition, not a claim that Veldarium operates robotics today. A machine acting on the wrong revision can automate the wrong work faster. Coherent build state has to come first.

AI belongs inside the same boundary. It may retrieve, compare, identify inconsistencies, trace dependencies, generate candidates, and help people understand the build. It does not turn generated output into approved engineering state. Missing evidence remains missing, and consequential release authority stays named.

Why begin with a vessel we can understand

A vessel forces abstractions to meet reality. A requirement has to become a design decision. A design decision changes material and work. Work has sequence, access, tooling, inspection, and acceptance constraints. A late change does not stay in a document; it moves through steel and through the people responsible for it.

We began by modeling one bounded vessel program because a concrete object exposes weak assumptions quickly. The point is not to pretend that a synthetic model carries the depth of naval architecture, production engineering, or yard experience. The point is to make VeldBuild answerable to a physical build structure from the start.

Why start smaller

I do not think a credible first physical step is to pretend a new organization should begin with a 400-foot combatant. The current VS-001 concept envelope is roughly 30–35 metres, about 100–115 feet. That is a synthetic starting envelope, not a completed design, cost estimate, regulatory category, or promise that the vessel will be built.

Smaller does not mean simple, cheap, safe by default, or lightly regulated. Mission, hull form, structure, stability, machinery, crew, route, flag, class, construction method, and operating rules would all have to be established through real engineering. The envelope only narrows the first learning problem while leaving enough physical complexity to make weak operating assumptions visible.

That is the company thesis, and it still has to survive reality: begin at a scale a new organization might eventually have a chance to understand, build, inspect, and learn from. Put what the steel and the people teach us back into the system. Then earn the right to attempt something harder.

What exists today

Today the public proof is deterministic and deliberately small. VS-001 is a synthetic conceptual vessel program, and WP-019 is a synthetic work package inside it. The records are hand-authored test data. This is not a customer vessel, not a live shipyard integration, not a class-reviewed design, and not evidence that Veldarium can deliver a vessel.

The proof answers one narrow operational question: can work readiness be established from configuration, material, predecessor, access, quality, and resource evidence, with named release authority and active holds preserved? Its result is three-valued—READY, NOT_READY, or CANNOT_ESTABLISH—so missing evidence cannot become an accidental yes or no. That demonstrates a control rule and an inspectable data model. It does not demonstrate production performance.

What we still have to earn

The future ambition is much larger: earn the technical, industrial, supplier, workforce, facility, safety, quality, and regulatory capability to help deliver vessels—and, over time, to become a shipbuilder. Veldarium does not operate a shipyard today, and this public proof does not represent a production deployment or vessel-delivery record.

Getting from a synthetic work package to physical capability will require partnership with the people who already know the work: naval architects, engineers, planners, buyers, welders, pipefitters, electricians, inspectors, suppliers, operators, regulators, and yards. It will also require learning where software should stop, where automation can safely help, and where judgment must remain unmistakably human.

The standard

America should be able to build hard things with competence, continuity, and respect for the people doing the work. I want Veldarium to contribute to that capability in a concrete way. The honest place to begin is not with a claim that we have arrived. It is with one vessel model, one bounded work package, one inspectable decision at a time—and a willingness to let reality correct us.

Sources