We're evolving how we build software — from writing every line by hand to directing AI agents with intent, judgment, and ownership. This is your map for that shift, and how we'll grow together from software engineers into AI engineers.
Start onboarding ↓ Jump to how KPIs workAI doesn't replace engineers — it raises the floor on what one engineer can ship and the bar on the judgment that matters. Our objective is to grow every software engineer here into an AI engineer. Accountability and measurement exist to support that growth, not to rank people toward the exit. A few principles are non-negotiable — read them first, because everything else follows.
What you ship — merged PRs, working software in GitHub — is the source of truth, not how busy you looked.
No metric or system ever auto-triggers an employment action. A human always reviews and decides.
Lines of code, commit counts, hours logged, and raw AI-usage volume are banned as targets — full stop.
Any telemetry is metadata, never screen content, keystrokes, or Claude's responses. You always know what's measured.
Observe → coach → formal notice → PIP with real support → human review. HR/legal gates are mandatory.
Weights and targets are tuned against a 4-week baseline plus a bias check before any score counts.
Same mission, new leverage. The mechanics of how you spend your hours change — but your ownership of correctness, architecture, and outcomes only goes up.
Two tools carry the AI-oriented SDLC: Claude Code for AI-assisted development, and GitHub as the single source of truth for what you ship.
Anthropic's agentic coding CLI — your pair for drafting, refactoring, debugging, and testing directly in your project.
The centralized record of output across all repos. Every KPI category reads from here, by author — so your work is captured fairly and consistently.
From the ground up: what you do each day building with Claude Code, and what happens every Friday. Accountability lives in GitHub; the weekly profile is a fast, honest self-check — coaching, never a verdict.
Repeat through the day. The work lands in GitHub; the KPI agent logs metadata passively in the background.
Two views come together: your GitHub output (technical) and your /weekly-kpi self-profile (human).
A fast 3-category pulse focused on coaching. Lightweight self-report plus the auto-pulled metrics — meant to spark a conversation, not to judge.
The full four-category composite, plus a trend adjustment (±5) and a small qualitative input (≤5). This is the considered view a manager reviews.
Performance rolls up from four categories into a single 0–100 score, capped at 100 and normalized so higher always means better. Here's the whole model at a glance.
Outcomes over activity — a score is an input, never a verdict.
A small, disclosed helper for company MacBooks. It records two metadata signals for the AI-adoption picture — time per app category, and your own Claude Code prompts (redacted on-device). One-time setup, then it runs quietly in the background. You stay in control, and it never touches screen content, keystrokes, or Claude's responses.
IT delivers the app; you open it once and acknowledge.
TechiesKPIAgent.app into /Applications (or you drag the signed, notarized build there yourself).After you acknowledge once, the agent launches on its own at every login and relaunches itself if it ever stops — you never have to open it again. Closing the menu-bar window does not stop collection; only Pause or Stop collecting & uninstall do.
From the menu bar: pause for 1 hour / 4 hours / until tomorrow with a short reason. Capture fully stops and auto-resumes. Pauses are logged as neutral context — never a score.
One click — “Stop collecting & uninstall” — removes the background helper and deletes the local queue. Fully reversible; just re-acknowledge to start again.
/weekly-kpi plugin in Claude CodeEach Friday you generate your own AI-Engineer Transformation Profile — an honest read on how your AI-engineering is growing. The plugin derives it from your own Claude Code history on your machine and submits only the metrics (six 0–5 stages + two short notes). Your prompt text never leaves your laptop.
📖 Full plugin guide — marketplace, install, updates, and the manager commands: KPI-Plugins.html.
/plugin marketplace add can't find it, you're not in the org yet —
see “If something doesn't work” below.)TECHIES_KPI_TOKEN you set in step 2.Two slash commands in Claude Code, then one environment variable.
~/.zshrc):
/weekly-kpi. It reads your week, scores six dimensions,
saves a local my-ai-profile-<date>.html you keep, and submits just the metrics./plugin update techies-kpi when P&C ships a new version.The plugin submits your email, the week, an overall stage, six integer stages
(adoption, intent, prompting_orchestration,
workflow_tooling, hygiene, leverage) and two one-line notes. It
never sends your prompt text, file contents, or the evidence quotes — those live only in the local HTML
you generate.
/plugin marketplace add says “not found” or permission denied → your GitHub account
isn't in the TechiesApp org yet. Ask P&C to add you, then retry./weekly-kpi fails with “TECHIES_KPI_TOKEN is not set” → you skipped step 2, or your
terminal was opened before you set it. Set the token and open a new terminal.Work through these in order. Your progress is saved in this browser, so you can come back to it anytime.