Sunday, November 22, 2026Coding news · developer employment · the CodeCore archive · RSS
CodeCoreCoding news, employment and the Vancouver tech community

AI Decision-Making Framework: Setting Boundaries Before AI Acts

A seven-part decision contract is essential for product teams to define rules before AI systems classify, route, score, or pick actions in workflows.

· DEV Community

CANADA — AI is transitioning from generating text to making decisions within real workflows, prompting the need for a structured decision-making framework. This framework outlines the rules product teams must establish before an AI system classifies, routes, scores, or selects actions. The framework includes defining the outcome, the exact question, allowed answers, evidence the model can use, the threshold for acting alone, human handoff, and the record kept afterward.

Previously, AI in products primarily generated content like chat replies, summaries, or drafts. Now, AI increasingly makes decisions such as determining if a request is in scope, assigning it to a queue, assessing issue severity, or deciding whether an agent should continue or seek help. OpenAI's Decisions API, released in beta on October 6, 2026, and generally available on October 22, 2026, exemplifies this shift. Developers provide evidence and questions, and the API returns typed answers.

The Decisions API accepts text, images, and JSON, handling up to 400 questions per request (200 during the beta). It is described as approximately 10 times faster than OpenAI's Responses API, with uncached input costing $0.10 per million tokens and output tokens not charged. This makes AI decisions a composable primitive, similar to search or notifications. However, while calling the model has become easier, defining the policy around it remains challenging.

The core challenge lies in determining what the decision entails, what evidence can influence it, and the actions to take when the system is uncertain. For instance, in a support workflow with three possible outputs, the policy behind these options remains undefined until the team answers follow-up questions. The model may return a clean choice, but the product team owns the consequences of that choice. OpenAI's documentation emphasizes that the API returns typed answers, and the application runs any actions those answers select.

This framework is crucial in product development, where decisions like 'build the feature,' 'run the experiment,' or 'ignore the request' have significant implications for users, resources, opportunity cost, and future commitments. Product teams should use this seven-part template before connecting an AI decision to any side effect, ensuring clarity and accountability in AI-driven workflows. Reported with files from DEV Community.

News

Critical Check Point Vulnerability CVE-2026-123039 Exploited in PHP2Shell Campaign

· DEV Community

Security researchers detail post-exploitation evidence and recovery steps for the exploited Check Point CVE-2026-123039 vulnerability with CVSS 7.8 score.

News

Developer Reveals Method for Scraping App Store Reviews Without Proxies

· DEV Community

A Node.js developer shares a technique for extracting international App Store review data in JSON format, enabling sentiment analysis without complex scraping infrastructure.

News

Databricks SQL AI Forecast V2 Adds Holiday Support and Covariates

· DEV Community

Databricks SQL's ai_forecast function now supports holiday effects and covariates, streamlining time-series forecasting directly within SQL queries.

News

How a WordPress LMS Front End Was Rebuilt with React and Cloudflare Pages

· DEV Community

A team modernized a WordPress LMS front end using React and Cloudflare Pages while maintaining core functionality like logins and payments.