{
  "id": 4220575,
  "title": "Why I separated live discovery from the AI chat box",
  "url": "https://urgent.news/2026/08/29/why-i-separated-live-discovery-from-the-ai-chat-box",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-29T15:32:58.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/zxhwolfe/why-i-separated-live-discovery-from-the-ai-chat-box-12c9"
  },
  "original_language": "en",
  "account": "Most AI workspaces begin with the familiar chat box interface. While utilizing AI Workstation for regular tasks, I encountered two inquiries that did not fit within a traditional chat flow. These queries required live data, and their failure modes differed from standard drafting. Models might generate plausible responses even with outdated information, mix project identities, overlook licensing restrictions, or assume popularity equates to quality. Consequently, I decided to divide AI Workstation into three distinct layers: a conventional workspace, public discovery Radars, and installable Agent Skills.\n\nLayer 1: The conventional workspace is where routine knowledge work occurs. It deals with questions, links, documents, images, drafting, proofreading, reusable templates, and exports. The objective isn't to conceal every operation behind a single, broad prompt. Instead, it aims to maintain accessibility for standard tasks while facilitating the integration of tasks that necessitate current data into a more explicit procedure.\n\nLayer 2: Public Radars for live discovery are designed to cater to creators and editors seeking up-to-date topic candidates and developers and researchers comparing active AI projects. The Global Topic Radar presents current topic options, their freshness, market context, evidential status, and original sources. The Open-Source AI Radar offers rankings, categories, collections, and project cards for AI projects, complete with direct links to their repositories. Crucially, the Radars do not claim to predict viral posts, guarantee project quality, or produce summaries that replace the original repository or license text. They operate without the need for a logged-in account, enabling visitors to examine real-time data before deciding if the workflow suits their needs.\n\nLayer 3: Agent Skills act as research contracts. While Radars identify potential leads, Agents require guidelines on subsequent actions. Topic Intelligence transforms a relevant Radar item into a structured content brief, encompassing research questions, verification requirements, claim avoidance, and visual specifications. These elements highlight the uncertainties that remain. AI Open Source Intelligence manages the open-source facet, resolving project identities, analyzing license evidence, creating comparison matrices, and outlining candidate stacks under explicit constraints. It also provides nine read-only MCP tools that allow an Agent to access public Radar data without executing external code. Both Skills are open-source and can be reviewed on GitHub, where users can inspect the Skill.md files, scripts, references, Agent metadata, and release assets. The aim is to enable users to review the operational contract before installation.\n\nThe reasoning behind this layered approach is to make potential errors more apparent. For instance, if the data is outdated, inspect the Radar data and source timestamps. If a project is mistakenly identified as another similarly named repository, review identity resolution. Should a claim not be ready for publication, place it in must_verify rather than attempting to make it appear confident in the final text. If an integration requires access to public data, keep it read-only instead of granting it execution privileges by default. This also explains why I do not characterize the system as a one-click content factory. Discovery and initial research occur within these layers. Subsequent stages, such as script review, asset creation, publishing, and performance optimization, are handled in later workflows, and the Radar cannot ensure performance.\n\nThe trade-off is a more methodical explanation of the product's functionality. Each component has a distinct responsibility, and there is no single button that claims to discover trends, verify every fact, produce production assets, publish results, and predict their reach. The advantage lies in the inspectability of the boundaries: the workspace manages general work; the Radars surface current leads and sources; the Skills structure repeatable Agent-side research; and the user retains final judgment. Feedback from those developing Agent workflows is particularly appreciated, especially regarding which aspects of live discovery should remain outside the model and which research checks should be enforced as part of the Skill contract.",
  "summary": "Most AI workspaces start with the same useful primitive: a chat box. I kept one in AI Workstation because it is still the fastest interface for many research and writing tasks. But while using the product for day-to-day work, I found two questions that did not belong in a general chat flow: What current topic is worth researching today? Which open-source AI project is worth evaluating now? Both…",
  "key_points": [
    "Public Radars provide live data for topic candidates, AI project rankings, direct repository links",
    "Agent Skills act as research contracts, transform Radar items into structured content briefs"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}