Sprawl
Code no one owns
Features a model wrote that nobody on the team fully understands, can safely change, or wants to be on call for.
New offering
AI Spring Clean
for the mess the gold rush left.
Your team shipped fast with AI. Now the codebase sprawls, the token bill creeps, and nobody quite trusts the parts a model wrote. I find what's actually costing you and clean it up.
The problem
AI got you moving fast. Now it's the tax.
Cost
A token bill that creeps
Prompts and calls bolted on without a budget. Inference spend that grows quietly and outpaces what the feature is worth.
Trust
Features that don't hold up
LLM features that demo well and break in production: no evals, no guardrails, flaky output, and no one sure how to fix it.
How it works
Audit first, then clean.
Step one
The Audit
A read of where AI is helping you and where it is costing you: the generated sprawl, the token and inference spend, the features that will not survive scale. You leave with a prioritised cleanup plan, useful whether or not I do the work.
Step two
The Cleanup Sprint
The work, on one fixed scope: cut the token bill, refactor the sprawl into code the team can own, and put evals and guardrails around the features that matter. Two to three weeks, one deliverable.
If it makes sense to keep it clean, I can stay on as a retainer.
Why me
Same problem, new cause.
Two decades of making systems that have to hold up in the real world: firmware on vehicles, a regulated medical platform, a commerce platform rebuilt to hold its load. AI sprawl is that same problem with a newer cause, code that grew faster than anyone could reason about it. Cleaning that up is the job I have always done. More on keeping systems healthy →
What it costs
Two steps, scoped to the work.
Audit
The read
from €6,000
One to two weeks. Findings and a prioritised cleanup plan.
Cleanup Sprint
The work
€18,000-28,000
Two to three weeks, one fixed deliverable. Net of VAT. Full engagement model →
Find out what AI is really costing you.
A 30-minute call to see whether a Spring Clean is worth it. If it is not, I'll say so.