Engineering Intelligence Agent
A fully autonomous AI agent that assembles, visualizes, and delivers my engineering team's weekly status report — pulling live data from four internal systems every Monday morning, with zero manual effort required.
- 4
- Live data sources
- ~2hrs
- Saved per week
- 11
- Report sections
- 100%
- Automated
Status reports are a tax on engineering managers
Every week, producing a meaningful status report for a technical team means manually pulling data from multiple disconnected systems, formatting it consistently, and distributing it to stakeholders — a process that consumes hours of management time better spent on people and strategy.
Fragmented data collection
Support ticket metrics, project progress, task completions, and team narratives each live in separate tools — requiring manual lookup and copy-paste every single week to compile into a single view.
Inconsistent formatting
Manual assembly means quality varies week to week. Charts, DRI attribution, and drill-down details are time-consuming to produce and easy to deprioritize when time is tight.
Repetitive, low-leverage work
The same retrieval, calculation, and formatting steps repeat every week — exactly the kind of structured, predictable workflow that AI agents are designed to own, not senior managers.
How it works
The agent runs on a scheduled trigger every Monday morning. It orchestrates parallel queries across four live data sources, aggregates and normalizes the results, assembles a structured HTML report with interactive visualizations, and delivers it via Slack and cloud archive — all without human intervention.
Data sources
Support Ticketing System
New · Open · Resolved · DRI breakdown
Project Management Platform
Portfolio · Tasks · Risk status · Progress
Team Weekly Update Doc
Wins · Blockers · Looking ahead
Identity & Profile System
Team photos · Display names · Titles
Claude Sonnet 4.6
Hyperagent Platform
- Aggregate
- Normalize
- Summarize
Outputs
Interactive HTML Report
11 sections · Charts · Expandable DRI details
Slack Summary
TL;DR + report link → team channel
Cloud Archive
Versioned HTML copy → shared drive
Metrics History Doc
Week-over-week trend tracking
Two-phase Monday automation
9:00 AM PT — The agent checks the team's shared update document and posts a Slack reminder (with direct @-mentions) to anyone who hasn't filed their weekly entry yet.
10:00 AM PT — One hour later, the full data pull runs and the report is delivered. The gap gives stragglers a window to complete their updates before the data is locked.
What makes it work
The agent isn't just a data fetcher — it applies engineering judgment to structure, filter, and surface the right information for each section of the report, with production-level reliability every week.
Parallel data collection
Dispatches concurrent subagents to query each data source simultaneously, with pagination handling for large datasets (12,000+ task records). Each team member is queried individually to avoid record-count API limits.
Interactive visualizations
Renders stacked and grouped bar charts with labeled values on every bar, full names on Y-axes, and distinct chart types optimized for different data patterns (ticket queues vs. task activity).
Expandable drill-down
Every team member section uses native HTML <details> elements for progressive
disclosure — summary stats up top, full ticket and task lists one click away. No
JavaScript required.
Persistent knowledge base
Maintains structured memory of field schemas, filter rules, and formatting preferences — so it never re-discovers context it already knows, and rule changes persist automatically across every future run.
Risk surface & KTLO separation
Automatically separates keep-the-lights-on operational work from the active project portfolio. Surfaces a Risk Tracker distinguishing projects by outcome achieved — not just lifecycle stage — so nothing is hidden behind a "Closed" label.
Team identity integration
Pulls live profile photos at full resolution from the team's Slack profiles, renders a human team strip with the manager visually distinguished, and attributes every single report bullet to the responsible team member.
What it changed
Measurable outcomes after moving from manual reporting to fully automated weekly delivery.
- ~2hrs
- Manager time recovered per week
- 52×
- Reports delivered per year, on schedule
- 4
- Live systems queried in a single run
- 100%
- Format consistency
Under the hood
The engineering decisions that make the agent production-reliable — not just a one-off demo — and able to handle real-world data edge cases every week without supervision.
// Agentic architecture
- Named persistent agent with versioned identity — skills, memories, and formatting rules survive across weekly runs with no re-prompting
- Orchestrator + subagent pattern — the parent agent coordinates; specialist subagents handle parallel data retrieval independently
- Each subagent writes raw and aggregated results to JSON files on disk; the orchestrator reads these for accurate expandable-details population, preventing fidelity loss from prose-only summaries
- Skill-based modular architecture — reusable, versioned scripts for ticket summarization, project aggregation, and HTML assembly; independently updatable
// Data engineering
- Native API filter syntax (not formula strings) to avoid silent-failure patterns on 12,000+ record tables that return all records regardless of formula
- Cursor-based paginated fetching with maximum page sizes for reliable bulk retrieval
- Server-side field projection to prevent token-limit overruns on wide data tables with 50+ fields per record
- Client-side date and collaborator filtering for fields not supported by the native filter API
- Queue-level filtering enforced on all ticket queries — prevents roughly 2× overcounting from unfiltered results
// Report generation
- Full HTML/CSS written from scratch — complete control over layout, interaction, and visual hierarchy; no template engine or CMS dependency
- Chart.js v4 with datalabels plugin for annotated bar charts — all bars show numeric values, full names on all axes, zero label clipping
- Distinct chart types by data pattern: grouped bars for created-vs-completed task comparisons, stacked bars for queue and portfolio composition
- Native HTML details/summary elements for zero-JavaScript expandable drill-downs — works in every email client and shared link view
// Scheduling & delivery
- Two scheduled invocations per week using iCal RRULE format — the deliberate 1-hour gap between reminder and report is a product decision, not a limitation
- Quarterly report variant triggered 10 days after each fiscal quarter close — same agent, separate schedule and output template
- Three delivery channels per run: Slack channel post (short summary first, link second), cloud drive archive (versioned HTML), persistent metrics history document
- Week-over-week metrics tracking document updated automatically each run for trend analysis over time
Built with
A combination of AI infrastructure, integration APIs, and frontend libraries working together in production.
AI / agent layer
- Anthropic Claude Sonnet 4.6
- Hyperagent Platform
- Orchestrator + Subagent Pattern
- Persistent Skills & Memory
Data sources & integrations
- Airtable (Project Management)
- Airtable (Support Ticketing)
- Google Drive API
- Slack Web API
Report rendering
- HTML / CSS
- Chart.js v4
- chartjs-plugin-datalabels v2
- Google Fonts — Inter
Scheduling & delivery
- iCal RRULE Scheduling
- Slack Bot Delivery
- Google Drive Archival