Agents · token economics · measured July 2026
Where your agent's tokens actually go.
You pay for a coding agent by the token: a $200 Claude Max plan, an API bill, a seat. You assume you're buying reasoning. On a real session we measured on a 4.47-million-file machine, Claude Code spent about 58% of its tokens on the least glamorous job in software: finding files. This is that session's token budget, drawn, and why more than half of it never touches your actual problem.
The anatomy of a session
Four things spend your tokens. Only two do the work.
Inside one agent turn, tokens go to four places: the fixed tax of the system prompt and tool definitions; the reasoning you actually want; the output it writes; and the tool results: every directory listing, every grep dump, every file the agent pulls in while it hunts. Three of those are small and roughly fixed. The fourth is the one that balloons, and here is what one real session looked like, tokens to scale.
One agent session · 100% of the tokens your plan pays for
Measured: Claude Code on the benchmark machine, one real file-hunt task. 58% of the session's tokens went to search-and-locate. The reasoning-versus-output split shown inside the remaining 42% is illustrative; only the file-hunting share is measured.
Why it reaches 58%
Every search result is billed again, and again, and again.
An agent works in a loop, and each tool result is appended to the conversation and re-sent on the next turn, and the next. A find that returns four thousand candidate paths adds four thousand lines the model re-reads on every step that follows. So file-hunting tokens aren't paid once; they compound. By the last calls, the agent spends more of its budget re-reading its own search history than reading the file it's looking for.
The compounding is the whole story. Turn the same drawing into dollars per developer per day and the bill is ugly; watch the same content query run on the tool your agent actually shells out to and it takes 93.8 seconds on ripgrep.
The money argument, drawn
A $200 plan that spends more time searching than thinking.
Claude Max is $200 a month. Claude Code runs on it. If most of a session's tokens go to locating files, then the plan is doing more searching than reasoning: you're renting a frontier model to run ls and grep. The model isn't the bottleneck. The floor it stands on is.
Split by session tokens, not a line on your invoice: usage doesn't map to dollars that cleanly. But the proportion is real, and it was measured on the machine below.
THE FIX ISN'T A SMARTER MODEL. IT'S A FASTER FLOOR.
What changes with an index
One call. The other 58% goes back to work.
Interlinked keeps a live, always-fresh index of every file on the machine: names and contents, every repo, including the git-ignored files a project-scoped search never sees. The agent asks once and gets the answer, so the hunt never enters the context in the first place. Same session, same answer. The tokens that were searching are now thinking.
| Per lookup | Walking the disk | With the index |
|---|---|---|
| Time to the file | 6m 57s | 16 ms |
| Tool calls | 71 | 1 |
| Session tokens on hunting | ~58% | ~0% |
| Single-file lookup | walks the drive | 85 μs |
| Edit → searchable | re-index / re-walk | ~1 ms |
That is roughly a 7,200,000× speed-up on the measured task. A single-file lookup returns in 85 microseconds; 139 microseconds is the median across 20 queries on the 4.47-million-file drive, best case 2. A save is searchable in about a millisecond, under 30 milliseconds worst case, so the answer is never stale. The whole engine sits around 44 MB at rest (less than a browser tab) and its index takes under 1% of the drive.
Methodology: Ryzen 9 9950X3D · 64 GB DDR5 · NVMe · Windows 11 · 4,470,000 files. The agent run is the measured Claude Code session; lookup times are the indexed engine on the same corpus. The same picture, as a stack, is in the backbone write-up.
Do this today
Stop paying to look for files.
Install once. One signed installer auto-configures 19 AI clients: Claude Code, Cursor, Codex, Copilot, Windsurf, Zed, Cline and the rest. No JSON editing, no keys.
Let it index. Filename search works immediately; content search fills in behind it. At rest the whole engine sits around 44 MB and its index takes under 1% of the drive.
Scope when you can. Whole-machine when you don't know where something lives; scoped to a folder when you do. Either way the answer is one tool call, not seventy-one.
Watch the budget move. The tokens that were hunting are now reasoning. Same plan, same model, more of it spent on your actual problem.
Everything local is free forever, on all your devices, no card. Hosting starts at $5.99/mo. No AI, no keys, no metered lookups. The index never leaves the machine. Download for Windows.