Core launches Spreadsheet Research agent loop in IrukaDark for automated enterprise list generation
The AMW Read
Incremental product feature update for a known Japanese AI assistant; no financial data, no structural shift, sub-segment relevance only.
Core launches Spreadsheet Research agent loop in IrukaDark for automated enterprise list generation
Core Inc. (コーレ株式会社, Tokyo) has added a new feature called "Spreadsheet Research" to its desktop AI assistant IrukaDark. The function uses a proprietary agent loop to autonomously discover candidates from the web, verify eligibility, and compile structured tables with source citations. Users specify target criteria, column structure, and desired row count; the AI then runs three parallel search loops followed by four parallel fill loops, with built-in hallucination validation and programmatic termination rules. The company positions this as distinct from deep research report generation, optimizing specifically for list-making tasks such as sales prospect lists, competitor comparisons, media contact databases, and market research tables.
Why it matters: This product update exemplifies the recurring pattern of "context-engineering moat" — rather than competing on foundation model capability, Core is building workflow-specific agent loops that solve a narrow, high-value enterprise pain point. The structured termination logic (program-side control of when to stop searching, vs. leaving it to the LLM) represents a pragmatic design choice that addresses a known reliability issue in autonomous agents. This move also surfaces the broader trend of AI tooling shifting from open-ended Q&A to deterministic, output-format-constrained workflows that enterprises can integrate into existing spreadsheet and CRM pipelines.
Grounded expert take: Spreadsheet Research is a textbook example of the "fastest-ARR-ramp" pattern applied to enterprise productivity. By automating the multi-step process of candidate discovery, verification, and table construction — tasks that currently require hours of manual search and transcription — Core targets a repeatable, high-frequency use case with clear ROI. The agent loop's three parallel search + four parallel fill architecture, combined with explicit termination conditions, offers a template for how AI agents can be made reliable enough for production use. The key open question is whether the output quality and citation accuracy will meet enterprise standards for sales and compliance workflows, and whether this will translate into measurable ARR growth for IrukaDark.