See how your team builds with AI

Agent sessions charted on per-user lanes, then compressed into metrics a lead can act on: output per prompt, leverage, and tool health.

Agent session segments flow onto per-user lanes, then compress into a team compare panel AGENT LANES · LIVE COMPARE · TEAM JT LEAD MS ENG KL NEW COMPARE · THIS WEEK PER PROMPT LEVERAGE ERR JT 4.2m 2.6× 3% MS 2.9m 1.7× 6% KL 1.1m 1.0× 11% team · 41 sessions · 9.4h agent time KL run:wait 52:48 · babysits one lane agent running wait ≤ 10 min subagents break · not counted
How the metrics work
Why this exists

Every engineer now commands multiple agents. Hours and review counts measure none of it.

agent running
wait ≤ 10 min · attributed fairly
subagents in parallel
break over 10 min · not counted
PER PROMPT 4.2 min LEVERAGE 2.6× ERR 3%

The work moved. An engineer's day is now prompts, parallel agent runs, and review. Commit counts and hours online describe the old job, not this one.

One board shows it. Every agent session lands on that engineer's lane: solid while the agent runs, dotted while they wait, striped while subagents work in parallel.

Rest is not penalized. Idle over 10 minutes counts as a break, not as wait. The metrics measure effectiveness in the zone, not time at the desk.

Compare, then coach. Select any set of engineers and see the same metrics side by side. The gap between a 4-minute prompt and a 1-minute prompt is a coaching conversation, not a mystery.

The three numbers that matter

Metrics built for agent work, not office work.

PER PROMPT
4.2 min
How much work one instruction buys: minutes of agent runtime and output tokens per prompt. Engineers who write prompts that carry show up immediately.
How it's computed →
LEVERAGE
2.6×
Average concurrent agents while at least one is running. A lead driving four lanes at once and a babysat single lane both show up here, honestly.
How it's computed →
TOOL HEALTH
3% err
Failed tool calls, redundant file reads, and rework episodes. Catches sessions that spin in circles instead of shipping, before the sprint review does.
How it's computed →