AIToday
AI Coding AssistantsAI Business & IndustryAmazon AI BlogPublished: Nov 18, 2026, 10:00 JST

AWS: ask if a step needs a model before picking one

Send on LINE
AWS: ask if a step needs a model before picking one

3 Key Points

  1. What happened

    AWS published guidance that routes workflow steps through three questions—can a rule check success, can a model write the rule once, is the answer from a fixed list—and sends only open-ended work to a right-sized Amazon Bedrock model.

  2. Why it matters

    Teams that default to model calls for every step pay per-token costs, add hundreds of milliseconds of latency, and get non-deterministic output that is harder to audit.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The guidance starts from a common pattern: a model call works for one step in a prototype, so it becomes the default for the next, and at production volume those small calls become one of the largest line items in the architecture. AWS frames the decision as a flowchart with three lanes—runs as code, questions, and runs on a model—and walks a customer support assistant through five steps to show how each maps to a specific AWS service. For steps that sit between a rule and open-ended generation, such as routing or classification, AWS describes a constrained decision call that picks from a fixed list and escalates when unsure, and it lists four implementation options including an experimental open source model, Strands Decider, whose README reports that answers at a confidence of 0.9 or more are right about 95 percent of the time on short classification tasks. AWS also recommends treating model choice as a tested configuration, with a golden evaluation set run in continuous integration and a configuration flag to reverse a swap.

FAQ
What are the three questions AWS recommends asking about each workflow step?
First, can a rule check success, such as a format, schema, keyword, or range? Second, can a model write the rule once, like a regular expression or JSON schema? Third, does the step pick one answer from a fixed list? If all three are no, the step runs on a right-sized model.
What is a constrained decision call?
It is a component that limits its output to a fixed list of options you define in advance and gives a signal for when to escalate uncertain answers to a right-sized model. AWS lists four ways to build one, including a small Amazon Bedrock model with structured outputs and embedding similarity.
How does AWS suggest teams start cutting model costs?
AWS recommends starting with one high-volume workflow, tagging every call as deterministic code, a constrained decision call, or a model call, and measuring how much of the workflow never needed a model.
Amazon 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 articleGE Aerospace's first AI chief builds single AI transparency platform