Skip to main content

Autonomous AI Agent

The production autonomous AI Agent engine inside the Qontak chatbot stack. Today the autonomous engine only runs in chatbot-ml-dev (a prototype with no FE/BE separation, rollout control, or plan gating); this initiative productionises it so Chatbot Specialists and customer Bot Builders can draft an agent from a prompt, configure its Behavior · Capabilities · Routing, attach knowledge + actions, preview, and ship — coexisting with the legacy session-based /ai-agent modal. The 26Q2 driver is 15 autonomous agents across 15 customer IDs inside the already-entitled Plus / Ultimate / 360 tiers.

Master index (ANCHOR)​

Phases​

PhaseScopeStatusEpic
Phase 1 — New Engine MigrationProduction autonomous-agent List / Create / Config (Profile · Capabilities · Routing) + Shared Actions — the new configuration model, coexisting with the legacy modalPRD draft; partial — shipped on /v2/ai_agents autonomous mode (BOT-4235, Done); deployed to production 2026-06-23 but gated to two internal sandbox accounts only (no client exposure, not GA) — a hard gate (unresolved autonomous/legacy staging runtime errors since 2026-06-03) blocks any client/CID enablement or GA claim; Shared Resources (KB) + preview rail pendingBOT-4191
Phase 2 — AI-Assisted Refinement (Confluence)Conversational "Refine" tab in the agent editor — describe a misbehavior or paste an error, get AI-proposed, reviewable apply-or-discard changes to a live agent's capability_pack (new POST /v2/ai_agents/:id/refine BE proxy → upstream refine-skill-pack); independently shippable on Phase 1PRD draft (v1.3); blocked on the upstream refine-skill-pack endpoint (Data/ML)BOT-4602
Phase 3 — Migrate Existing (Internal) ConfigurationsMigration plan moving agents already configured internally on the legacy Tree-Diagram config onto the new engine + new config model (picks up Phase 1 Non-Goals #1/#3)TBD — Engineering + Data/ML team; data to be providedTBD
Phase 4 — New-Configuration IterationIterate the new config from the Design Validation research (15 IDIs): visual Logic Map, terminology overhaul, Milestone/Routing fixes, inline validation, snippetsPlannedTBD

Reconciliation note (2026-06-22): the Confluence draft describes a new FrontendService::V3 namespace, /ai-agent/autonomous routes, and an autonomous_ai_agent rollout flag. The code shipped differently — FrontendService::V2 (/v2/ai_agents autonomous mode), /bot-automation/* FE routes, and an ai_agent_engine_version === 2 org-setting gate. The full drift table is in the Phase-1 PRD's "Implementation Reconciliation" section.

Sibling initiatives (not phases of this one): AI Agent Testing (BOT-3351) now owns the batch test-runner + judging scope this PRD deferred; AI Agent Knowledge overlaps the Shared Resources surface. Both are separate programs with their own anchors — confirm the hand-offs before the Phase-3/4 PRDs are written.

Jira Epics (multi-Epic initiative)​

This initiative spans more than one Jira Epic — the Phase 1 migration and the Phase 2 refinement surface are tracked separately. Jira's BOT project is flat (no tier above Epic), so the initiative is anchored by an umbrella Epic — BOT-4601 — AI Agent — Autonomous Agent (umbrella) (the jira_epic above) — and each phase hangs off its own child Epic below. The umbrella is a naming anchor only: the flat project cannot re-parent the child Epics under it. Because the initiative genuinely spans several Epics, the README frontmatter declares jira_epic_model: umbrella, which switches the delivery drift linter (lint-delivery.mjs CHECK GROUP H) into umbrella mode: it validates each phase PRD's header Epic against the child-Epic allow-set below (this table's Child Epic column) instead of requiring one Epic per initiative. A declared child Epic passes; an undeclared / typo'd one fires PRD_EPIC_NOT_IN_FAMILY, and a PRD that collapses onto the umbrella anchor fires PRD_EPIC_IS_UMBRELLA. Keep every phase PRD's child Epic listed here.

Phase / groupChild EpicNotes
Phase 1 — New Engine MigrationBOT-4191"26Q2 Autonomous Agent Migration" (In Progress). Shipped on /v2/ai_agents autonomous mode (BOT-4235, Done); 80/80 child issues Done per the 2026-06-25 reconciliation.
Phase 2 — AI-Assisted Refinement (Refiner)BOT-4602"26Q3 AI Agent Refinement (Refiner)". New Epic (2026-07-13) — Phase 2 no longer rolls up under the Phase-1 Migration Epic. Blocked on upstream refine-skill-pack (Data/ML).
Adjustment — Simplified Configuration (Steps/SOP)BOT-4733"26Q3 AI Agent Simplified Configuration (Steps/SOP)" (To Do, created 2026-07-22). PRD READY: prds/simplified-configuration.md.
Phase 3 — Migrate Existing ConfigurationsTBDEpic to be created when the Phase 3 PRD leaves TBD.
Phase 4 — New-Configuration IterationTBDEpic to be created when the Phase 4 PRD is authored.

Scope Changes​

  • Backend — FrontendService::V2 autonomous-agent endpoints, skill_pack adapter, capability validators, OSS-first KB ingestion.
  • Frontend — /bot-automation/* List / Create / Config (AiAgentEditor) + Actions; engine-version routing vs the legacy modal.
  • Data — upstream noncore-mrag / chatbot-ai / mekari-agent contracts; KB vectorisation.

QA Lane​

Lane B — keeps a human QA gate. Triggers: real-time chat flow (autonomous agent handles live customer conversations) + multi-org data isolation (agent knowledge and actions span customer workspaces). Classified by DRI on 2026-06-26. See .claude/reference/qa-lanes.md for the decision rules.

Contents​

  • prds/ — phase PRDs (each with its own ACs → Jira Epic) land here.
  • rfcs/ — Request for Comments (technical design proposals). The Confluence RFC (QON 51184074800) can be mirrored here.
  • tests/ — E2E / acceptance test specs.
  • delivery/ — TPM timeline, status reports, decision records.