Skip to one-page overview
Iria MonitorAI CODE OBSERVABILITY

The entire ecosystem needs to measure. Now more than ever.

Microsoft is ending Copilot cost subsidies. Enterprises will see real AI costs for the first time. Iria Monitor measures what every euro invested actually produces.

The headlines your clients read

Every major consultancy is publishing the same message. Boards read it and ask for proof.

McKinsey, 2025
“AI will automate 25% of software engineering tasks by 2027.”
Goldman Sachs
“3–4x developer productivity gains with AI coding tools.”
Gartner, 2025
“By 2028, 75% of enterprise software engineers will use AI assistants.”
Accenture
“+84% more builds per week in AI-assisted teams.”

The board translates this into: how much will we cut development costs? And asks for proof.

What most enterprises measure today

Microsoft Copilot + Power BI. The default dashboard for 90% of large organizations.

ID
Who has a license
%
Acceptance rate
ORG
Team assignment

Activity metrics. Not business outcomes.

Two questions. No shared answer.

THE BOARD ASKS
How much will we cut development costs?
vs
THE CTO ANSWERS
How many developers use Copilot?

Measuring adoption is not measuring impact. The gap between these two questions is where decisions stall and budgets get frozen.

June 1, 2026: the “all-inclusive” model ends

Until today, the Copilot license ($19/user/month) included all usage. Microsoft absorbed the real inference cost. Starting June 1, costs are billed by actual token consumption. Multipliers for the most powerful models skyrocket:

ModelUntil todayFrom June 1Change
Claude Sonnet 4.51x6x+500%
Claude Sonnet 4.61x9x+800%
Claude Opus 4.53x15x+400%
Claude Opus 4.6 / 4.73x27x+800%

Clients like Caixabank, who are deploying Copilot, don’t know how to implement their plans with this change. The flat-rate cost model no longer works.

“GitHub has absorbed much of the escalating inference cost behind that usage, but the current premium request model is no longer sustainable.” — GitHub Blog, April 2026

Proving AI productivity in large orgs is nearly impossible

No baseline
You don’t know how long things took before AI.
Metrics that don’t correlate
Prompts and acceptance rates don’t predict real output.
Bureaucracy masks gains
Procurement, security reviews, and change boards absorb any improvement.

DORA State of DevOps 2025: org-level delivery metrics stayed flat despite widespread AI adoption.

Iria Monitor measures what AI actually writes. Line by line.

Not prompts. Not acceptance rates. The actual code that lands in production.

What code
Which lines the AI wrote, in which files, in which repos.
How much
% of AI vs human code, by team and project.
At what cost
Real token cost of each AI edit, annotated in the commit.
What survives
Code rejected, modified by a human, or still in production after 30 days.

Iria Monitor records real costs from day 1 — even when subsidized. When the June bill arrives, you’ll know exactly what each euro produced.

How it works

  1. 1
    Agent hook
    Captures what AI edits. Claude, Cursor, Codex, Windsurf.
  2. 2
    Git annotation
    Data in refs/notes/ai. Code never leaves the machine.
  3. 3
    Dashboard
    Authorship, cost, and durability per repo, team, and developer.

Your cloud or ours

Private cloud
  • Client’s own infrastructure
  • Data never leaves their environment
  • Compliance-ready from day one
SaaS
  • Onboarding in minutes
  • Ideal for pilots
  • SOC 2 / ISO 27001 on the roadmap