Qontak | AI Agent | Autonomous AI Agent — ANCHOR

ANCHOR PRD — the initiative master index. It orchestrates all phases beneath it and carries no acceptance criteria of its own (ACs live in each phase PRD). Synced with the canonical Confluence Autonomous AI Agent — Phase 1: New Engine Migration (QON 51188335138) and reconciled against the actual codebase (chatbot, chatbot-fe) and Jira (Epic BOT-4191).

Scope: Autonomous AI Agent = the production autonomous-agent engine inside the Qontak chatbot stack — letting Chatbot Specialists and customer Bot Builders draft, configure (Behavior · Capabilities · Routing), and ship an autonomous agent, coexisting with the legacy session-based /ai-agent modal. Phase 1 (New Engine Migration) productionises the chatbot-ml-dev prototype; later phases add batch testing / automated judging.

HEADER BLOCK

Field	Value
PM	Dimas Fauzi Hidayat
PRD Version	1.0
Status	ACTIVE
PRD Type	ANCHOR
Epic	BOT-4191 — "26Q2 Autonomous Agent Migration" (In Progress)
Labels	epic:qontak-chatbot | module:ai-agent | feature:autonomous-ai-agent
Last Updated	2026-06-22

Phase Index

Phase	Goal	PRD Link	Epic	Status
Phase 1: New Engine Migration	Productionise the autonomous-agent engine + the new configuration model (Profile · Capabilities · Routing wizard) inside chatbot BE + chatbot-fe (List / Create / Config / Actions), coexisting with the legacy /ai-agent modal	prds/phase-1-new-engine-migration.md	BOT-4191	🔄 In Progress (shipped on /v2/ai_agents autonomous mode — BOT-4235, Done; Resources + preview pending)
Phase 2: AI-Assisted Refinement	Let tenants iterate a live autonomous agent's capability_pack through a conversational AI surface in the agent editor — paste an error or describe a misbehavior and get a conversational reply plus reviewable, apply-or-discard config changes — by adding a refine proxy on chatbot BE over the upstream refine-skill-pack, reusing the Phase-1 engine/config model and the drafter's validation pipeline	prds/phase-2-ai-assisted-refinement.md	BOT-4191	📝 Draft
Phase 3: Migrate Existing (Internal) Configurations	Define + execute the migration plan moving AI agents already configured internally on the legacy Tree-Diagram production config (and the intermediate Data-Team design) onto the new Autonomous engine + new config model — picking up Phase 1's deferred Non-Goals #1/#3 (legacy coexistence → migration), so internally-built agents move over without a full rebuild. Scope TBD — owned jointly by Engineering + Data/ML team; migration inventory + mapping data to be provided (separate session).	TBD	TBD	⏳ TBD — pending Eng + Data team input
Phase 4: New-Configuration Iteration	Iterate the new configuration from the Design Validation research (15 IDIs): a visual Logic Map for Milestone → Action → Routing, a full terminology overhaul (e.g. Milestone → Checkpoint, Performance Exit → Handover, model tiers → Engine Tier w/ credit cost), Milestone simplification, global-vs-capability routing-hierarchy clarity, inline validation, stepper navigation, and inline-snippet KB updates — lifting non-technical self-serve confidence from the research's ~50–60% to the independent-config threshold	TBD	TBD	⏳ Not started

Status options: 📝 Draft · 🔄 In Progress · ✅ Shipped · ⏸ Paused · ❌ Cancelled

Reconciliation (2026-06-22): Phase 1 landed on the existing FrontendService::V2 (/v2/ai_agents in "autonomous agent mode", BOT-4235) rather than the new FrontendService::V3 namespace the Confluence draft describes; FE routes are /bot-automation/*; rollout is gated by the org setting ai_agent_engine_version === 2 (not an autonomous_ai_agent system_preference). Shared Resources (KB) and the preview rail (AAA-S04 / S09) are not yet built. Full drift table lives in the Phase-1 PRD's "Implementation Reconciliation" section.

Phase 3/4 source of truth. Phase 4 (new-configuration iteration) is grounded in the Design Validation research — Qontak AI Agent Design Validation | AI Autonomous Agent Configuration Interface (Devina Amalia / Wulan Febyazzahra / Dimas; 15 IDIs; Google Doc). It frames the three-design lineage (existing Tree Diagram → Data-Team design → Designer-proposed Wizard = the new config) and a P1–P3 action plan (Logic Map, terminology overhaul, Milestone/Routing fixes, inline validation, snippets). Phase 3 (migration) moves agents built on the existing Tree-Diagram production config onto the new model.

Scope handed to sibling initiatives (not phases here): batch test-runner + /test/improvise + /test/judge (Phase 1 Non-Goals #4–#5) now belong to AI Agent Testing (Epic BOT-3351); the Shared Resources / KB surface overlaps AI Agent Knowledge. PM to confirm the hand-offs before each Phase-3/4 PRD is written.

---

2. One-liner + Problem

One-liner: A production autonomous AI Agent flow inside the Qontak chatbot stack — draft from a prompt, configure Behavior / Capabilities / Routing, preview, and ship — replacing the prototype-grade chatbot-ml-dev app and coexisting with the legacy /ai-agent modal.

Problem: The autonomous-agent engine only runs in chatbot-ml-dev, a prototype Vue/Vite app calling the ML services directly with no FE/BE separation, no rollout controls, no audit trail, and no plan/tier enforcement. Chatbot Specialists tasked with delivering 15 autonomous agents across 15 customer IDs in 26Q2 cannot run the prototype safely against customer workspaces, and the legacy modal is a different product (session-based, Redis, V1 endpoints) that cannot serve the autonomous use case. This is the second iteration of the AI Agent line; competitor pressure makes shipping the production autonomous flow inside the already-entitled Plus / Ultimate / 360 tiers a 26Q2 commitment.

---

3. Target Users + Persona Context

Persona	Role	Goal	Pain	Workaround
Primary — Chatbot Specialist	Internal Qontak specialist (technical) configuring autonomous agents for/with customer accounts; owns the 26Q2 target of 15 agents across 15 cid	Stand up an autonomous agent end-to-end inside the production stack, with audit trail + rollout control, and hand it to the customer	Only flow today is the chatbot-ml-dev prototype — no plan gating, no auth surface, no rollout flag, not exposable to customers	Builds in the prototype outside the customer workspace, then describes config manually; or degrades into the legacy modal
Secondary — Dedicated Bot Builder (customer-side)	Technical or non-technical Plus/Ultimate/360 admin maintaining their company's AI agents	Self-serve an autonomous agent in their own admin: draft, add knowledge, preview, ship — without a Mekari specialist for every change	No production path; legacy modal lacks the autonomous engine; prototype is internal-only	Hands the request to a Mekari specialist and waits, or accepts a degraded legacy-modal build

---

4. Success Metrics (initiative-level)

⭐ Primary KPI: Number of Autonomous AI Agents successfully built and live

• Definition: distinct ai_agent_uuid with at least one successful save and one successful full preview (configured + validated end-to-end)
• Baseline: 0 (no production autonomous agents today)
• Target: ≥ 15 agents across 15 distinct cid by end of 26Q2

Activation: % of rolled-out Plus/Ultimate/360 workspaces that save + validate ≥ 1 agent within 30 days — target ≥ 60%.

Quality: Draft success rate ≥ 90%; KB upload + vectorisation success ≥ 95%; preview success ≥ 95% (all within 30–60 days of GA).

Efficiency: Median time from Create-entry to first successful save ≤ 30 min for specialist-led builds (informs the later self-serve phases — Phase 3 migration / Phase 4 iteration — readiness).

---

5. Key Decisions + Alternatives Rejected

5a — Decisions Made

Date	Decision	Rationale
2026-05-22	Ship the autonomous agent as a new surface coexisting with the legacy /ai-agent modal (untouched)	The legacy modal is a different product (session/Redis/V1); bolting the engine on would conflate two data models and add regression risk
2026-05-22	Phase 1 is proxy/BFF-only (no new Rails DDL); BE fronts noncore-mrag / chatbot-ai / mekari-agent	Reduces blast radius and accelerates the 26Q2 15-cid target; audit/analytics deferred upstream
2026-05-22	Public vocabulary uses capability_pack / capability_id; upstream skill_pack / skill_id translation stays in the adapter	Decouples customer-facing language from ML vocabulary
2026-06-22	Implemented on /v2/ai_agents ("autonomous agent mode") rather than a new FrontendService::V3 namespace	Team decision under the same Epic BOT-4191 (BOT-4235, Done) — extend V2 with engine-version awareness instead of standing up a parallel V3. Supersedes the Confluence draft's V3 framing
2026-06-22	Structure Autonomous AI Agent as an ANCHOR; Phase 1 = New Engine Migration, Phase 2 = migrate existing (internal) configurations, Phase 3 = new-configuration iteration	The autonomous line is multi-phase: ship the new engine first, then migrate agents built on the old config, then iterate the new config from research. Batch test-runner / judging moved to the AI Agent Testing initiative (BOT-3351)
2026-06-22	Phase 3 scope anchored to the Design Validation research (15 IDIs, Devina/Wulan/Dimas)	The research is the canonical input on what the new configuration needs — Logic Map, terminology overhaul, Milestone/Routing fixes — so iteration runs against evidence rather than opinion (this phase is now Phase 4 — see the 2026-06-29 reorder below)
2026-06-29	Reordered the Phase Index: AI-Assisted Refinement promoted to Phase 2; Migrate Existing Configurations → Phase 3; New-Configuration Iteration → Phase 4	Refinement builds only on Phase 1 (drafter + engine/config) and is independently shippable, so it is the next concrete deliverable; the Migrate/Iteration phases remain TBD and were renumbered to follow it. Supersedes the 2026-06-22 ordering (Phase 2 = migrate, Phase 3 = iterate)

5b — Alternatives Rejected

Alternative	Why Rejected	Date
Extend the legacy /ai-agent modal with an "Autonomous mode" toggle	Legacy modal is session/Redis/V1-only; conflating two data models risks regression on a live product	2026-05-22
Productise chatbot-ml-dev directly as the customer app	Prototype-grade — no auth surface, plan gating, rollout control, audit, or OSS-first compliance	2026-05-22
Mirror autonomous agents into local Rails tables in Phase 1	Adds DDL + migration + dual-source-of-truth risk during the 26Q2 window; revisit in a later phase under this anchor if admin tooling needs it	2026-05-22

---

6. Open Questions (initiative-level)

#	Type	Question	Owner	Deadline
1	Open Question	Phase 3 (migration) — TBD, owned by Engineering + Data/ML team. How many agents are already configured internally on the legacy Tree-Diagram config; is the migration automated (mapper old config → new capability_pack) or specialist-assisted; what is the cutover / coexistence plan? Migration inventory + mapping data to be provided by the user in a separate session before this phase is scoped.	Eng Lead (Eko) + Data/ML team	TBD (awaiting data)
2	Open Question	Phase 4 (iteration): which P1 research items ship first — visual Logic Map, terminology overhaul, or Milestone simplification — and what confidence-uplift target (from ~50–60%) gates "done"?	PM + Design (Wulan/Devina)	before Phase-4 PRD
3	Risk	Batch test-runner + judging (Phase 1 Non-Goals #4–#5) handed to the AI Agent Testing initiative (BOT-3351). Mitigation: confirm that initiative owns it so it isn't dropped between programs	Dimas (PM)	2026-07-15
4	Risk	Shared Resources / KB (AAA-S09) overlaps the AI Agent Knowledge initiative and is not yet built under /bot-automation. Mitigation: decide which initiative owns the surface	Dimas (PM)	2026-07-15
5	Open Question	Spec-vs-code drift (V3→V2, ai_agent_engine_version flag, shipped DELETE vs no-delete AC, unbuilt preview rail) — reconcile the Phase-1 PRD body to V2 reality	PM + Eng Lead (Eko)	2026-07-15

---

PRD CHANGELOG

Version	Date	By	Section	Type	Summary
1.0	2026-06-22	Claude	All	CREATED	ANCHOR created from the Confluence "Autonomous AI Agent — Phase 1: New Engine Migration" draft (QON 51188335138, Epic BOT-4191). New Engine Migration set as Phase 1 (shipped on /v2/ai_agents autonomous mode, BOT-4235); batch testing/judging seeded as Phase 2 from this PRD's deferred Non-Goals. Reconciled against chatbot / chatbot-fe + Jira; V3→V2 and other drift captured in §5a and the Phase-1 PRD's Implementation Reconciliation table. AI Agent Testing (BOT-3351) and AI Agent Knowledge flagged as separate sibling initiatives, not phases here.
1.1	2026-06-22	Claude	Phase Index, §5a, §6	MODIFIED	Phases 2 & 3 redefined per PM: Phase 2 = migrate existing internally-built configurations (legacy Tree-Diagram config → new model; picks up Phase 1 Non-Goals #1/#3); Phase 3 = new-configuration iteration grounded in the Design Validation research (15 IDIs — Logic Map, terminology overhaul, Milestone/Routing fixes). Batch test-runner + judging moved out to the AI Agent Testing initiative (BOT-3351). Open Questions reframed around the migration + iteration phases.
1.2	2026-06-29	Claude	Phase Index	MODIFIED	Added AI-Assisted Refinement ("Refine with AI" / refine-skill-pack, RFC §10.3b — QON 51153994292 / 51226214880) as a phase under this anchor. New chatbot BE refine proxy over the upstream refine-skill-pack + FE diff-preview chat; reuses Phase-1 engine/config + drafter validation; independently shippable (depends only on Phase 1). PRD: prds/phase-2-ai-assisted-refinement.md (numbered Phase 4 when added; renumbered to Phase 2 in v1.3).
1.3	2026-06-29	Claude	Phase Index, §5a, §6	MODIFIED	Reordered phases per PM: AI-Assisted Refinement promoted Phase 4 → Phase 2 (next concrete step after Phase 1, independently shippable); Migrate Existing Configurations → Phase 3; New-Configuration Iteration → Phase 4. PRD file renamed to phase-2-ai-assisted-refinement.md; source-of-truth note, success-metrics note, and Open Questions renumbered; reorder decision recorded in §5a.