Matching agentic AI architecture to the right problem

Q: How do you decide whether a production AI use case needs an agentic architecture or a classical model?

The decision follows the shape of the problem, not the novelty of the technology. Classical, function-based approaches work when components change on different schedules and inputs are well-defined, as with a statistical pricing model paired with separate business rules. Multi-agent designs earn their cost when the task is genuinely interpretive, such as synthesizing hundreds of data points into a coherent narrative where ambiguity and multi-perspective review add real value.

Q: What causes AI systems to fail in production if the models themselves are accurate?

Most production failures are operational rather than statistical. Write locks from logging directly to a relational database, authentication tokens exposed in cleartext logs, and memory accumulating under load until a model crashes are software engineering failures, not AI failures. A Grafana dashboard providing AI observability into memory usage under load turned repeated crashes into a diagnosis instead of a guessing game; the same visibility later exposed idle infrastructure worth resizing.

Q: What is intent engineering and how is it different from vibe coding?

Vibe coding is fast, prompt-driven development that produces a working demo but skips authentication, validation, and observability unless explicitly requested, since it generates only what was asked for. Intent engineering adds a middle phase: a human-readable specification defining inputs, outputs, and failure modes before any code is written. Domino's own agentic engineering playbook documents a comparable phased discipline for production-ready AI coding agents.

The right AI agent architecture depends on whether a problem's ambiguity sits in the input data, where agents add real value, or in the system design, where it creates risk.

Thomas Reichert

Data Scientist

Helvetia Baloise

What you'll take away from this session

Separate prediction logic from policy logic

Statistical models and business rules change on different schedules and should be built as separate components. Combining them means retraining the whole system every time a policy changes.

Validate every agentic input and output

Typed data models replace free text at agent boundaries, and specialized validators review generated content before a human sees it. Quality problems drop sharply once ambiguity at interfaces is removed.

Observability turns crashes into diagnoses

A dashboard showing memory accumulation under load is what separates a diagnosed failure from an unexplained restart. The same visibility later caught idle resource waste worth fixing through hardware tier changes.

Freeze every dependency before production

Every dependency, including transitive ones, needs explicit versioning before production. An unpinned update once pulled in a dependency whose download was blocked by a corporate proxy, breaking a job that had run for months.

Keep AI out of deterministic execution

AI should accelerate the writing of deterministic code rather than run it, especially where output must be exact, auditable, and governed by explicit rules. This keeps AI-assisted speed without probabilistic risk in production.

Thomas Reichert, data scientist and AI solutions engineer at Helvetia Baloise, argues that AI amplifies whatever discipline or chaos already exists in a system rather than fixing what is broken. Drawing on two production systems built at the insurer, a classical dynamic pricing engine for motor insurance and a multi-agent financial planning tool for life insurance, Reichert shows that the choice of agentic AI architecture should follow the shape of the problem rather than the novelty of the technology.

The pricing engine uses no agents because its statistical and business-rule components change on different schedules and gain nothing from agentic reasoning. The financial planning tool uses multiple agents because synthesizing hundreds of data points into a coherent narrative is genuinely interpretive work.

Both systems failed in production for reasons that had nothing to do with model accuracy: database write locks, cleartext credentials, memory leaks, and unpinned dependencies. As Reichert frames it, the failure mode shifts once AI moves from notebook to production, and the fixes that matter most are ordinary software engineering discipline, not better models.

FAQ

How do you decide whether a production AI use case needs an agentic architecture or a classical model?

What causes AI systems to fail in production if the models themselves are accurate?

What is intent engineering and how is it different from vibe coding?

Transform the work that matters most

See how Domino helps the world’s most regulated enterprises build, scale, and govern AI-powered applications.

Watch demo Talk to an expert