For software agencies
You quote fixed price.
Your AI spend isn't.
Your developers' AI tooling is now a real cost of delivery, and it arrives as one org-wide invoice with no client attached to any line on it. Codelitics splits that bill across the repos you scope, so the engagement that burned three times its share stops hiding inside the average.
AI spend this month · $8,000 tracked
One fixed-price engagement burned 2.9x the AI budget it was quoted with. The pooled invoice shows none of this.
Engagements in scope
a client project is a repo- 2.9x its quoted shareravenna-replatformFixed price$4,310 AI spend · $61 / realized change
- In marginatlas-mobileTime & materials$2,150 AI spend · $24 / realized change
- In marginhelios-apiRetainer$1,540 AI spend · $19 / realized change
On the invoice, these three engagements are one number.
View engagementsThe problem
The bill arrives by the month. You get paid by the project.
That mismatch is the whole problem, and it only exists for people who sell delivery. Four ways it shows up:
Fixed-price work absorbs every overrun in silence.
Nobody escalates a token bill. It lands in a pooled invoice, gets averaged across the portfolio, and the engagement that caused it never carries the cost.
T&M can't bill what it can't evidence.
You could pass the tooling cost through, but only with a defensible per-project number. Without one it stays overhead, which means you eat it.
Your next quote is a guess.
Pricing AI-assisted delivery means knowing what it actually cost you last time. A seat count doesn't tell you that. Neither does a token total.
You can't tell leverage from spend.
Two teams on the same tools, same seats, same bill. One is shipping code that lasts and one is regenerating the same module all quarter. On the invoice they're identical.
One repo, 30 minutes
See the number for one of your own projects first.
We install on a single repo of your choosing and show you the delivery cost and what survived. No client work needs to be involved.
What you get
Delivery cost per repo, and how much of it lasted.
Cost per realized change
What a shipped, surviving change actually cost you in AI spend, per repository. The number a quote can be built on.
Which tools earn their bill
Which of the tools you pay for produce code that is still load-bearing weeks later, and which are billing you for churn.
Survival and Code Half-Life
How long AI-authored code stays in the codebase before it is rewritten or deleted. Regeneration is the cost nobody invoices for.
Exportable, traceable figures
Every figure traces back to how it was computed, so it holds up in a rate conversation with a client or a partner meeting internally.
Every figure is scored per repository and rolled up, so a client engagement reads as its own P&L line instead of a share of an average. The measurement behind it is public in the Return on Code standard.
How it works
A per-seat agent, the repos you choose, and nothing in your pipeline.
Install the per-seat agent
Each developer runs a lightweight agent on their own machine. It uses git hooks and plugins for the AI tools your team already runs, so the workflow stays exactly as it is. There is no CI component and nothing to add to your build.
Scope the repos you choose
The dashboard connects through a GitHub App or GitLab OAuth and reads only the repositories you put in scope. Most agencies start with a single internal repo to see the shape of the numbers before any client work is involved.
Read the numbers per engagement
Cost per realized change, tool yield, and survival for every repo in scope. A client project is a repo, so this is the per-client view: which engagements your AI spend sits on, and what it bought.
See it before the call
Two minutes of proof, then the dashboard itself.
The walkthrough shows how a month of AI spend gives up its savings. The live demo is the same workspace: open it and see what a per-engagement view looks like before a single client repo is involved.
The live demo
Behind the blur: a team's month of AI spend, with the leak marked.
Your work email unlocks it right here. We'll also send you one email with a link to book a walkthrough.
FAQ
The questions agencies ask first.
We're an agency, not a product company. Does this still apply?
It applies harder. A product team absorbs AI coding spend into a general engineering budget and asks whether it was worth it at renewal. An agency bills the work. Every hour your developers spend with an AI tool sits inside an engagement you quoted, so the tooling cost lands directly on the delivery margin of that specific project. You feel a bad month in the P&L, not in a dashboard.
Can we see cost per client or per project?
We scope by repository, and for most agencies a client project is a repository, so in practice that is the same view. You put the repos you care about in scope and get cost per realized change and tool yield for each one. If a single client spans several repos, they group. What we do not do is guess at a client boundary you have not drawn for us.
Do our developers have to change how they work?
No. Capture happens through a per-seat agent on each developer's machine using git hooks and plugins for the AI tools already in use. They keep working in Claude Code, Cursor, Codex, or whatever they run today. Nothing is injected into how code gets written or reviewed. The agent derives the metrics after the fact, from the repository and local AI activity.
Our client code is under NDA. What goes in scope?
Only the repositories you put there. You decide, and you can start with a single internal repo to see the shape of the numbers before any client work is involved. The dashboard connects through a GitHub App or GitLab OAuth and reads the repositories in scope to derive metrics.
Can we show these numbers to our clients?
That is the most interesting thing agencies do with them. Every figure on the dashboard is exportable and traceable to how it was computed, which turns an argument about whether AI-assisted delivery is real into a document. Some agencies use it to defend a rate, some to justify a faster quote than a competitor. Both work better with evidence than with a claim.