INGOAMPT AI News · Original developer analysis published Saturday, 7 November 2026 · Vienna (CET/UTC+1). The source’s publication and event dates are distinguished below. Illustrations are original editorial artwork rather than documentary photographs.
In brief: New Chinese reporting tests Comma’s persistent coordinator, parallel workers and cross-app workflow, comparing the architecture with DeepSeek’s recently discussed Super-agent direction.
What’s new in today’s reporting
On Saturday 7 November, Chinese technology publication QbitAI published a hands-on examination of Comma, an agent workspace built by AFK AI, alongside DeepSeek’s Super-agent architectural proposal from earlier in the week. The product and original research proposal were not both launched today; the new contribution is the comparison and first-person test.
Router, workers and longer-lived context
The article describes a persistent ‘router’ conversation that delegates tasks to specialized workers with separate task histories. The reporter explored workflows involving writing a weekly summary, coding a game and tracking tasks through Linear. Cross-app connectors, ongoing events and worker status reporting may reduce the need for an operator to supervise multiple independent chats.
What is independently demonstrated
This is hands-on tech journalism, not a reproducible benchmark of coding success rates or a security audit. Statements about long-term memory learning and self-improving task models are presented as product and publisher claims. Before connecting a business account, verify the permissions model, whether local files leave your machine, retention, tool limits, licensing and the handling of failures.
The agent architecture lesson
A straightforward architecture for a solo founder is one dispatcher that classifies the user’s task, assigns a tool-limited worker, receives structured results, and asks for approval before irreversible changes. Separate long-term task state from model context, so the system can resume accurately without exposing a complete private conversation each time.
Mini exercise
Create two safe tasks, ‘summarize five public sources’ and ‘draft release notes from a sample changelog’. Let a coordinator pass narrowly scoped instructions to two workers. Log execution, reject actions with missing authorization, and compare time to completion, corrections and cost against two separate manual prompts.
Sources, limitations and further reading
INGOAMPT’s recommendations and test exercises are independent editorial analysis, not claims by the cited publishers. Product demonstrations, company statements and reported financing are attributed rather than treated as independently proven. This article is educational, not investment or legal advice.
