Back to Activepieces

Chat Engineering Gaps — Priority Order

.agents/chat-engineering-gaps.md

0.85.02.6 KB
Original Source

Chat Engineering Gaps — Priority Order

Priority 1: Batch Execution for One-Time Tasks

Status: Done Impact: Critical — blocks 8 of 10 use cases (any multi-item task) Effort: Medium

Implemented: Extended ap_execute_action with items[] array (max 100) + description for progress label. Worker-side batch loop calls executeAdhocAction per item via RPC, pushes data-batch-progress stream events for a live-updating BatchProgressCard in the chat UI. Continue-on-error. Timeout scales with item count. Single-item path unchanged.


Priority 2: Value-Before-Auth Flow

Status: Not started Impact: High — users hit auth wall before getting any value, kills conversion Effort: Low (prompt-only change)

Problem: The agent calls ap_discover_action_auth → immediately requests connection. Users are asked to authenticate before understanding what the agent will do for them. Zero value delivered before the ask.

Fix: Add system prompt rule: "Before requesting connections, explain what you will do and show the execution plan. Only request auth after the user confirms the approach." This is a prompt-only change with outsized impact on first-time experience.


Priority 3: Connection-Based Suggestions

Status: Not started Impact: Medium — existing users get no personalized experience Effort: Medium

Problem: Chat does not suggest actions based on existing connections. A user with Slack, GitHub, and Google Sheets connected sees the same generic empty state as a brand new user.

Fix: On conversation start, query user's connections and show personalized suggestions via ap_show_quick_replies (e.g. "Send a Slack message", "Check GitHub PRs", "Add a row to your spreadsheet"). Inject connection context into the first-message prompt.


Priority 4: Analytics Events

Status: Not started Impact: Medium — flying blind on whether tasks succeed, cannot measure PMF Effort: Low-Medium

Problem: Current telemetry only syncs tool call counts to console. No outcome-level tracking. Cannot answer: "Do users complete tasks?", "Do one-time tasks convert to flows?", "Which task types succeed?"

Fix: Add 3 events:

  • CHAT_TASK_COMPLETED — fires when a one-time action or batch succeeds
  • CHAT_FLOW_CREATED — fires when ap_build_flow completes successfully from chat
  • CHAT_CONVERSION — fires when user converts a one-time task to a flow (Gap 2's "Automate This?" path)

Completed

Gap 2: "Automate This?" Suggestion

Status: Done System prompt suggests converting successful one-time tasks to flows via ap_show_quick_replies.