AIToday
Large Language ModelsAI Regulation & PolicyDatabricks AI BlogPublished: Nov 20, 2026, 06:01 JST

Databricks' Nagamuthu maps four agent patterns for regulated firms

Send on LINE
Databricks' Nagamuthu maps four agent patterns for regulated firms

3 Key Points

  1. What happened

    In a Databricks blog series, Principal Forward Deployed Engineer Molly Nagamuthu described four patterns — deterministic control with agentic reasoning, agentic execution with a deterministic judge, agentic reasoning with human authority, and bounded autonomy — for placing agents in controlled systems.

  2. Why it matters

    The framework gives architects a way to decide where an agent may reason or act and where a reproducible control or accountable person must keep authority, rather than treating agent adoption as a model choice.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The series is aimed at the gap Nagamuthu describes between agents that can reason through complex problems and enterprise systems, especially in regulated industries, that must stay controlled and auditable. Rather than framing this as a choice between rules and agents, she presents it as a question of how the two work together, and she notes that customers keep asking how other organizations handle the same constraints.

The patterns are organized around two recurring questions: where determinism must remain, such as when actions are irreversible or errors carry asymmetric risk, and where an agent can be given authority, particularly when actions are reversible or cheap to remediate. Her examples run from financial reconciliations, where an agent may investigate a failure but the deterministic control still decides pass or fail, to legacy migration judged by automated validation, financial crime investigations where an accountable person keeps the final decision, and governed data access granted inside a defined operating envelope.

Nagamuthu also argues that existing deterministic systems need not be discarded. Policy engines can become control planes, validation frameworks can become deterministic judges, and operational limits can become autonomy boundaries. She adds that the boundaries can shift as systems mature, with judge rules refined from usage data and human authority narrowed as feedback builds confidence in agent reasoning.

FAQ
What are the four patterns?
Deterministic control with agentic reasoning, agentic execution with a deterministic judge, agentic reasoning with human authority, and agentic execution with bounded autonomy.
How does Databricks say its platform supports these patterns?
It cites Omnigent for composing agents, Unity Gateway for governing access to models and tools, Unity Catalog for data access and residency rules, and MLflow for preserving traces and evaluation evidence as audit records.
Who wrote the series?
Molly Nagamuthu, a Principal Forward Deployed Engineer at Databricks who works with global banks on data and AI transformation.
Databricks AI BlogRead Original Article
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleOsaka Prefecture to back municipal AI agent rollout