AInsight

A privacy-first, vendor-neutral framework for classifying and aggregating AI-assisted engineering work by topic, depth, and progression.

Core principle

Leadership should see patterns, not conversations.

AInsight is a lightweight analytical layer between engineers, AI assistants such as Claude, ChatGPT, Copilot, and similar tools, and delivery systems such as GitHub. It does not measure AI adoption through prompt count, token usage, license activation, or raw AI spend.

Its purpose is different:

to understand what kinds of engineering work AI is supporting, at what depth, how this changes over time, and how those patterns can improve team learning, delivery, and technical capability.

What AInsight measures

• Topic distribution — what kinds of engineering problems AI is supporting
• Depth — boilerplate, applied implementation, debugging, design, foundational reasoning
• Progression vs. repetition — whether the team is moving toward deeper work or repeating the same basic questions

Why usage metrics are not enough

A team can use AI more often and still produce more rework, more review load, shallower code, hidden skill gaps, more technical debt, and more tool dependency without deeper engineering capability.

The important questions are not how many prompts were sent or how many tokens were consumed. They are: what kind of work did AI support? Was it boilerplate or deep technical reasoning? Did the team progress or repeat the same basic questions? Did AI usage expose learning gaps?

AInsight exists to make these questions observable.

Minimal architecture

AI-assisted work
-> local classification
-> structured weekly report
-> snapshot storage
-> aggregation
-> topic heatmap
-> trend analysis
-> retrospective / learning loop

Repository structure

README.md
METHODOLOGY.md
PRIVACY_MODEL.md
SNAPSHOT_MODEL.md
TAXONOMY.md
GITHUB_CORRELATION.md
TEAM_LEARNING_LOOP.md

ai-work-topic-quantifier/        reference implementation (local-first scripts)
  config/                        taxonomies and component map
  schemas/                       JSON schemas for daily and weekly reports
  prompts/                       prompt packs for daily, weekly, manager summary
  scripts/                       build_prompt, prepare_git_context, aggregate_reports, pseudonymous_id
  examples/                      sample activity note, git context, weekly report, team summary
  docs/                          architecture, collector contract, privacy model, rollout plan

Roadmap

1. Public concept — publish methodology, privacy model, snapshot model, taxonomy, sample reports.
2. Reference schema and taxonomy — weekly report schema, aggregate report schema, core taxonomy, example domain taxonomies, sample developer prompt pack.
3. Minimal aggregator — JSON/YAML report loader, team aggregation script, topic heatmap generator, basic trendline output.
4. Dashboard — individual and team dashboards, topic and depth trendlines, hierarchical drill-down, snapshot comparison.
5. GitHub correlation — PR data mapping, CI pass/fail correlation, component-level topic mapping, review/rework indicators.
6. Real team deployment — validate privacy model, taxonomy usefulness, reporting burden, retrospective value.

Example output (excerpt)

{
  "report_id": "2026-W17-dev-pseudo-01",
  "scope": "individual",
  "report_period": { "start": "2026-04-20", "end": "2026-04-26" },
  "taxonomy_version": "ainsight-core-0.1",
  "topics": [
    {
      "id": "operating_systems.kernel",
      "share_total": 0.14,
      "children": [
        { "id": "kernel.scheduling", "share_total": 0.06, "share_parent": 0.43 },
        { "id": "kernel.locking",    "share_total": 0.04, "share_parent": 0.29 }
      ]
    }
  ],
  "depth_distribution": {
    "boilerplate": 0.18,
    "implementation": 0.31,
    "debugging": 0.24,
    "design_tradeoffs": 0.17,
    "foundational_learning": 0.10
  }
}

A full example lives in SNAPSHOT_MODEL.md.

Slogans

• Patterns, not conversations.
• Insight before platform.
• Measure AI-assisted engineering work, not just AI usage.
• Understand what AI is supporting, not just how often it is used.

Documents

• METHODOLOGY.md — full conceptual model
• PRIVACY_MODEL.md — governance and trust boundaries
• SNAPSHOT_MODEL.md — time-series model and reporting logic
• TAXONOMY.md — core and domain-specific taxonomy design
• GITHUB_CORRELATION.md — delivery signal mapping
• TEAM_LEARNING_LOOP.md — retrospective and T-shaped skills use