Agent Technologies

AI Agents that deliberate, challenge, and write code

While building its own AI Persona Synthesis Research system, Root Kernel has operated a workflow where people and AI Agents communicate in shared documents, multiple Agents deliberate, and product and server changes move through specification and validation steps.

Doksuri and ATN support AI Agent communication and decision-making across company work. AI-SPARK is the patent-pending technology used specifically in Sudal service server development.

Operating Model

AI collaboration embedded in product development

Agent Technologies are not separate demos. They are working practices accumulated while building Root Kernel's AI Persona Synthesis Research system. Doksuri is the shared document surface for working with AI Agents, and ATN is a moderator-led turn structure that makes Agents share and challenge each other's views.

Technologies

Agent Technologies

Doksuri and ATN are used across company operations for AI Agent communication and deliberation. AI-SPARK is the agentic backend development technology applied to Sudal's service server.

In operation / development

Doksuri

Markdown-native Human-AI Collaboration Platform

A collaboration surface used across company work where people and AI Agents read and write the same Markdown documents.

PlaceholderDoksuri metaphor image. An eagle precisely snatches the needed requirements and evidence from scattered Markdown fragments, then organizes them into a work document with Agent comments, work results, and review state.

In operation / improving

ATN

Agent Turn Network

A deliberation layer where a moderator assigns turns so multiple AI Agents speak from their roles and challenge each other's claims.

PlaceholderATN script-style deliberation screen. A moderator assigns turns to Product, Architecture, Risk, and User roles, ending with one Decision Brief.

Patent pending / in development

AI-SPARK

AI Spec, Policy, Approval, Runtime Kernel

Patent-pending technology used in Sudal service server development, handling server changes through Spec Bundle, Validator, Approval Gate, and FSM Runtime steps.

PlaceholderAI-SPARK approval-gate image. Use Requirement → Spec Bundle → Validator → Human Approval → Safe Runtime as the main flow, with Fail-closed and ReverseMap as small support badges.

In Practice

How the operating model works

These technologies are not forced into one linear pipeline. Each one supports a different working situation inside the company.

Working situation	How it is used	Why it matters
Working through documents	Doksuri keeps requirements, task instructions, results, evidence, and reviews in the same document surface.	AI work does not disappear into chat logs. It becomes part of the company's work record.
Running turn-based deliberation	ATN uses a moderator to assign turns and make Agents see and challenge each other's claims.	Instead of a final synthesized answer, the team can inspect who said what, in what order, and how claims were challenged.
Validating server changes	AI-SPARK was used in Sudal server development to connect natural-language requirements to Spec Bundle, Validator, and Approval Gate steps.	AI-generated server changes are treated as reviewable, validated revisions instead of being applied directly.