check out the new AI metrics
- AI Usage by Dev
- Lines of code generated vs written by humans
- Which devs are using which models
- How much those AI generated lines are costing in tokens
We're rolling out this dashboard to teams for Claude, Codex, Cursor, etc.
According to @WakaTime, I use AI to write 83% of my code these days, and recently used it to redesign the @WakaTime dashboards for devs and dev teams, with a new focus on AI metrics instead of time.
I used #Codex 60% of the time and #Claude 40% of the time, switching between the two when hitting their rate limits or when one model got stuck. Yes, they didn't always do things correctly and sometimes introduced security bugs, but thankfully I code reviewed everything, paying very close attention to the backend changes so I noticed when it left PII exposed in the API, made incorrect assumptions in business logic, or just introduced random bugs.
Having a different session code review the AI changes before committing helped, but not as much as I thought it would.
The new feature I'm most proud of: Cost by Agent
It shows your "real" cost by token usage for all AI agents you use so you can see if you're using the full value out of your Claude/Codex/Cursor subscription.
So @instagram decided it's not safe to let me login because I moved countries. How do I fix this? It even says my password is correct, it's just not "safe" to let me login to my own account! WTF @facebook
Most tools stop at tokens, WakaTime shows more:
→ AI Adoption per dev/team/agent
→ AI Spend per dev/team/agent
→ AI Effectiveness - when devs fix what AI generated
Only WakaTime can accurately attribute each line of code to AI or old-school Human.
wakatime.com/ai
What every engineering team wants to know about AI:
→ Is adoption going up?
→ Are we getting our money's worth?
→ Are we shipping faster?
WakaTime answers all three. Free for individuals.
wakatime.com/ai
Most AI dashboards show team-wide totals.
WakaTime breaks it down per dev:
→ Which agents each teammate uses
→ How much of their code is AI-generated
→ Their AI spend, by model and agent
Find your power users. Spot your cost outliers.
Most tools only work with one agent/harness, WakaTime works with:
→ Claude, Claude Code, Claude Desktop, etc.
→ Codex
→ Cursor
→ Copilot
→ Gemini
10 more
The draft of @WakaTime wrapped code stats is on GitHub at github.com/wakatime/wakatime… let me know if you find any improvements before it goes live tomorrow!