Skip to content
GetShims

Coming soon · Private beta

Dashboards show you the gap.
GetShims closes it.

GetShims finds where your AI coding agents waste tokens, review time and budget. It ships the fix, then proves whether it worked.

No spam. We email you once, when your invite is ready.

Works with Claude Code, Cursor, Codex and GitHub Copilot

GetShims · InsightsSample data

This week's top action

Move small tasks to the small model tier

$31.8k

a month, range $24.6k–$38.2k

5,140 small tasks ran on the large model tier last month. Six teams that already route them to the small tier saw no rise in rework.

High confidenceLow effortReversible

Where the money quietly goes.

None of this shows up on an AI bill. All of it shows up in GetShims.

refund.test.tsmock not set up

  • Session 1failed$14.20
  • Session 2failed$15.80
  • Session 3failed$16.40
  • Session 4failed$15.60
$62 · same failure

The retry loop

An agent fails the same test four sessions in a row. The fix is usually one missing line in the repo instructions.

Session receipt

Task
Rename config keys
Changed
1 line
Model
Large tier
Time
9 min

Total$18.64

Big model, tiny job

The oversized model

Large models doing small jobs. A one-line rename doesn't need the most expensive tier.

payments-core#3041

Make refund writes idempotent

Waiting

14.2 h for a first review

R134%
R227%
+1539%
2 people · 61% of reviews

The review queue

Work piles up behind the same two reviewers, and the biggest PRs wait the longest.

Illustrations use sample data

Not another dashboard. A loop.

Here is that retry loop, found, fixed and proven.

    1. SessionImplement refund retry
    2. Commit+188 lines
    3. CIrefund.test.ts failed
    4. Retry ×4same failure · $62
    5. Mergedafter 2 days

    01

    Spot the leak

    GetShims links each agent session to the commits, CI runs and reviews it caused, and flags the step where money and time leaked.

  2. PR #3412 · AGENTS.md

    Add test setup note for ledger-sync

    @@ testing @@+ Mocks for tests: see tests/README+ Stop after 2 identical failures
    Approved

    02

    Ship one fix

    You get one change at a time, as a pull request or a setting. Nothing changes until the right person approves it.

  3. Repeat failures per 10 tasks

    −71%2.1 → 0.6

    no change−80% to −58%
    • First-try CI passunchanged
    • Reworkunchanged

    03

    Prove it worked

    Each fix is measured before and after, with an honest range and a check that quality held. Keep it, or undo it in one click.

Small fixes. Compounding gains.

Each shim is one small change, proven and kept. Together they add up.

Cost per shipped PR · 12 weeks

$66.90 → $47.50

−29% from three kept fixes
$50$60$70Week 1Week 6Week 12Small tasks → small model−13%Plan before large tasks−9%Minimal-code rules−7%
  1. Small tasks → small model−13%
  2. Plan before large tasks−9%
  3. Minimal-code rules−7%

Sample data. Each step was measured before and after, with rework and CI checked.

One loop. Every level.

Developers

A private coach. Only you see your own numbers.

Managers

See where work waits, without blaming anyone.

Leaders

Spend tied to shipped work, with savings you can prove.

  • Prompts and code never leave your machines
  • Nobody is ranked, ever
  • Results shown as ranges, not spin

Be first in line.

We are onboarding a small group of engineering teams before launch. Tell us about yours.

We only use this to plan the beta.