AI & Agents
What coding agents do while they run, and what that costs in tokens and wall-clock time. Written from session data, for people using Claude Code, Codex and Cursor as part of normal work.
How to Track AI Token Spend by Project (2026)
To track AI token spend by project you need per-session records carrying the model, the token counts split by pricing tier, and the repo the session ran in. Here are five ways to get that, where each one breaks, and real numbers from a 13,674 dollar month.
What Is Agentic Engineering Intelligence? (2026 Definition)
Agentic engineering intelligence is the measurement of software work done by humans and AI agents together: session time across IDEs, terminals, and coding agents, tied to commits, PRs, and token cost. Definition, the five dimensions it measures, and how it differs from developer analytics and editor time trackers.
How to Track Cursor Usage (Time, AI Diffs, and Shipped Work) in 2026
Track Cursor usage three ways: editor time, AI-generated diff share, and shipped output. Because Cursor is a VS Code fork, editor plugins can see it, but they cannot tell your work from the agent's. Here is how to get a number you can defend.
How to Track Claude Code Usage (Time, Tokens, and Shipped Work) in 2026
Track Claude Code usage three ways: token and cost logs, session time, and shipped output. Here is what each method sees, what it misses, and how to get a number you can actually defend.
How to Measure the ROI of AI Coding Assistants (2026)
Measure the ROI of AI coding assistants by comparing what you spend (subscriptions, tokens, review time) against what actually shipped, not lines generated or suggestions accepted. Here is the framework, the numbers that mislead, and how to prove the payback.
How to Track the Percentage of AI-Generated Code in Your Project (2026)
Track AI-generated code percentage by measuring diffs at the commit level, not by eyeballing a codebase. Here is what actually counts as AI-generated, the methods that work, and why the number alone still cannot prove authorship.
How to Measure AI Coding Productivity in 2026 (a Framework That Holds Up)
Measure AI coding productivity by tracking leverage and shipped output, not lines or tokens. Here is the framework: what to measure, what to ignore, and why commit volume and token counts mislead in the agent era.
Changelog
All updatesAugust 26, 2026
The DevClocked Agent
Ask questions about your own data in plain language at /agent. It reads through 15 tools covering summaries, projects, time breakdowns, sessions, commits, clients, invoices, unbilled time, timesheet items, invoicing coverage, and workspace rollups. Eight more tools can propose actions, but nothing runs until you accept its confirm card. Ask it to close out the month and it walks the invoicing recipe one step per turn, from coverage check to invoice draft. Available on Pro.
August 23, 2026
Gemini CLI Agent Tracking
The Mac app now reads Gemini CLI sessions, so those agent turns land beside your Claude Code and Codex work instead of disappearing into plain terminal time. The daemon parses your local session files for the model, input, output and cached tokens, tool calls, and subagent nesting, then attributes each session to the right project. Message content never leaves your machine; only counts and metadata are sent. Requires Mac app v1.4.49 or later. Gemini models are not priced yet, so you will see token counts rather than cost.
August 13, 2026
Export Reports as PDF
Every report — Overview, Commits, and Issues — now has an Export PDF button that produces a clean, print-optimized document honouring your active filters and date range, ready to attach to an invoice or share with a client. Exports are generated locally in your browser; no data leaves your machine.
August 13, 2026
Figma Design Tracking
The Mac app now tracks time in the Figma desktop app as real design hours — automatically, with nothing to configure. Turn on optional file-name tracking to attribute design time to individual Figma files and auto-map it to projects by naming files "Project — File". Off by default, reading window titles of design apps only, behind an explicit macOS Accessibility grant you approve yourself.