AI spend, attributed to work

See what your AI coding spend actually shipped.

Your AI tools can tell you who spent the tokens. None of them can tell you what the spend produced. RidgeLine meters every AI coding session to the code change it shipped, and puts a number on the spend that shipped nothing.

Read-only. Your code never leaves your estate. UK-hosted.

Finding · Stranded SpendIllustrative
≈ £4,210

of agent work last month never reached a pull request

  • payments-service412 sessions≈ £1,830
  • web-checkout198 sessions≈ £1,140
  • internal-tools164 sessions≈ £1,240
18% of monitored spendconcentrated in three repositories
01 · The situation

The bill went variable. The answers did not.

The bill

The AI invoice became usage-based.

Copilot, Cursor and Claude now bill the way clouds do. Last month's invoice is no longer a guide to next month's, and the line item is growing faster than any other tool you buy.

The blind spot

You can see who spent it. Not what it built.

Provider dashboards stop at people and days. When a board asks what the spend shipped, the honest answer today is that nobody can say.

The waste

Some agent work never ships. Nobody measures it.

Sessions that end without a pull request still cost real money. Across the industry, nearly half of agent-written fixes are rejected. That spend disappears into the total.

It is not just you.

Across the industry, adoption has outrun proof of return. The leaders running the largest AI budgets say openly that they cannot yet tie the spend to what shipped, and the numbers behind them are stark.

“It's very hard to draw a line between one of those stats and ‘okay, now we're actually producing 25% more useful consumer features’.”

Andrew Macdonald, COO, Uber · Fortune, May 2026
4 mo

for Uber to exhaust its entire annual AI coding budget, before imposing per-engineer caps

Fortune, 2026
15%

of enterprises forecast their AI costs to within ten percent of what they actually spend

Survey of 372 enterprises, 2026
46%

of agent-authored fixes are rejected before they ever merge

AIDev research corpus, 2026
02 · The answer

There is a ledger for this now.

RidgeLine is the attribution layer for agentic engineering. It joins the provider's meter to your development record, so every agent session lands on the commit and pull request it produced, with its cost attached.

Work that shipped carries a number. Work that shipped nothing becomes a finding with a name, Stranded Spend, and a value in pounds. Not an average across your team; each session metered to the change it became.

Read-only, connected in an afternoon. The how is below.

Spend Lineage · one release, tracedIllustrative
  • a3f9c12
  • 7c21d04
  • e9d40b7
#482 · Merged2.1M tokens · ≈ £38
  • b512f8a
  • 04c9e11
#489 · Merged1.7M tokens · ≈ £29
Releasev2.143.8M tokens
  1. 9 agent sessions3.8M tokens
  2. 5 commits3.8M tokens
  3. 2 pull requests3.8M tokens
  4. 1 release≈ £67
Five commits, two pull requests, one release. The meter rolls up with the work, and nothing is lost on the way.
03 · How it works

Connected in an afternoon. Answers the same week.

01

Instrument the project, once

A read-only connection your engineers review and approve. It takes about an afternoon, and nothing is copied out of your estate.

02

The meter runs on its own

From that point on, every AI coding session in the project is metered automatically, whoever ran it. No process change, nothing for your team to remember.

03

See what the spend became

Slice the ledger by feature, bug fix or release. Every commit, pull request and deploy carries its metered cost, so work that shipped and work that stranded sit side by side.

ridgeline.cloud/o/your-org/usageIllustrative
ChangeContributorTokens≈ CostStatus
checkout-api · PR #482D. Okafor2.1M≈ £38Shipped
auth-service · PR #217L. Fenwick840k≈ £15Shipped
billing-webhook · no pull requestR. Vance4.8M≈ £86Stranded
Metered per change, never averaged. Each entry also carries its estimated carbon.

Metered per change. Not averaged across your team.

04 · The difference

The number your provider dashboard cannot show you.

Provider consoles meter people per day. They cannot see which piece of work a session became, so they cannot tell you what the spend produced, or that it produced nothing. Put the two side by side and the question changes from who spent the most to what did we get.

A provider console

  • Spend per person, per day
  • No link to commits, pull requests or releases
  • An average across the team
  • One vendor's meter, reading its own bill

RidgeLine

  • Spend per commit, pull request and release
  • Work that shipped nothing surfaced as a finding
  • Metered per session, never averaged
  • One neutral ledger across every agent you run

Then open any pull request.

In RidgeLine, each pull request carries its own cost record: which models worked on it, on which days, and what it cost to ship.

ridgeline.cloud · your-org / checkout-apiIllustrative
Rework payment retries #482Merged

D. Okafor merged feature/payment-retries into main · opened Monday, merged Friday

Cost to ship
≈ £38
Tokens
2.1M
Agent sessions
6

Tokens by model, per day

  • Opus
  • Sonnet
  • Haiku
A provider console cannot draw this page. It cannot see the pull request.
05 · The cause

The waste has a cause. The cause is fixable.

Stranded spend concentrates where the setup is weak: repositories without required pre-merge checks, stale steering files, agents working without a review gate. RidgeLine grades each repository's configuration and shows which fixes recover the most.

We rate the setup, never the person.

Harness Score · payments-serviceIllustrative
C
  • Read-only telemetryenabled
  • Pre-merge testsnot required
  • Steering filestale for 142 days
  • Review gaterequired
Two fixable causes account for most of this repository's stranded spend.
06 · Governance

Governs the harness and the work. Never the person.

Governance here means assurance: that AI-assisted software is built under known controls, by accountable people, without uncontrolled exposure, at accountable cost, and that this can be proven to a board or a regulator. Every session already carries a receipt, cost and provenance captured at the moment it ships, so the ledger that finds stranded spend is the same one an auditor can read.

Harness Assurance

Are guardrails actually being applied.

Exposure Control

Is code exposed to unapproved tools, accounts or jurisdictions.

Provenance & Accountability

Who did what, with what AI, traced to the commit.

Spend Accountability

AI cost, per pull request. The same receipt shows what shipped and what stranded.

Attestation & Reporting

Proof a board, an auditor or a regulator can read.

Every token, traced to the code it became. On the Enterprise tier, findings become gates: pull request checks, policy enforcement, and the vendor reconciliation that joins what you pay an AI vendor to what shipped. From token to commit to release, it's the only AI spend that's git-anchored.

07 · What it unlocks

One ledger. Answers for every room.

The same session-to-change record that finds stranded spend also answers the questions finance, delivery and procurement keep asking. And everything the ledger knows is available over the API and webhooks, so your own tooling can ask too.

DeliveryOn the roadmap

Put a cost on a feature, a sprint, a fix.

Every pull request and release already carries its metered cost. Linking your issue tracker rolls that up to epics and sprints, so a piece of work has a price rather than a guess.

Cost allocation

Attribute spend to clients, teams and cost centres.

A cost allocation code maps spend onto the labels your finance team already uses: client, product, lifecycle stage. Chargeback and showback without a spreadsheet.

Procurement

Walk into your renewal knowing the number.

Consumption by team, repository and model gives you a run-rate you can defend, so committed-use negotiations start from your data rather than the vendor's.

The board

Give finance the feedback loop it is missing.

Value returned on AI spend, as a report a board can read, with standard-format exports for the tooling your finance team already runs (FOCUS, the week the 1.5 standard ratifies).

08 · Trust

Designed so your engineers will say yes to it.

Your code stays yours, and stays put.

RidgeLine works on usage metadata: tokens, timestamps, repositories, sessions. Where it grades a repository's guardrails it reads in place; it never copies your code out, never stores it, never trains on it, and never passes it to another system. The change that enables telemetry is a pull request your engineers can read before they approve it.

Numbers you can put in front of a board.

Token counts are measured exactly. Money is always shown as what it is, an estimate marked ≈, never dressed up as an invoice. Your data stays in the UK, in London (eu-west-2).

Read exactly what RidgeLine sees
09 · Founding partners

Six places. A fair trade.

Why six. The programme exists because RidgeLine is early and is being shaped with its first customers. A fortnightly call with the founders does not scale past six teams, so six is the honest number, not a marketing device.

Who we are looking for. Engineering organisations in the UK or EU running AI coding agents across real repositories, typically between fifty and five hundred engineers, with a leader who wants the stranded-spend number and a team willing to say what is working and what is not. Smaller teams are welcome to start on the free Instrument tier; the waitlist is below.

The deal, stated plainly. Each partner starts with the two-week audit, free, then pays half the list price for the first twelve months, locked at signature, with a direct line to the founders and a real voice in the roadmap. In exchange we ask for a fortnightly feedback call and permission to publish anonymised numbers as a case study.

The audit is how both sides find out whether there is a partnership worth having. If the number is boring, you keep the memo and owe nothing.

Start with the audit

Start with your own number.

Two weeks, read-only, ending in one sentence: how much of your AI coding spend never reached a pull request. If the number is boring, you have lost nothing.

Request the audit

Two weeks. Read-only. One number.

Join the waitlist

No audit, no partner programme, just the product: leave a work email and we will write to you as RidgeLine opens to more teams.