Tom Killi

The thread

In 1996 I founded x.hlp Technologies: enterprise software that captured how people used applications, paired those interactions with outcomes, and improved training from the results. We served 30+ Fortune 500 companies including JP Morgan and Shell, plus McKinsey and NATO, before the 2004 acquisition. Two acquisitions later the technology was inside SAP.

My patent from that work, published as US20010039552A1, has since been cited by 50+ patents, from Microsoft, Apple, IBM, Samsung, Nokia and SAP itself among others. The number is there so you can check it.

In 2014 I founded Wyetree, building energy infrastructure and trading software. That architecture, real-time coordination under uncertainty, became the foundation for CleanAim®, founded 2020.

The book

Silent Wiring

The Hidden Failure Mode in AI-Generated Code

Silent failure is a known family: code that produces wrong behaviour with no crash, no error, no page. Silent wiring is the member you get when an AI writes the integration code, the seams between parts of a system it cannot see end to end.

I found four of them in our own 1.1 million line codebase while 18,000 tests ran green.

A doubt engine registered, subscribed, never firing, and writing to a table no consumer read.
A calibration engine whose computation had stopped months earlier while a fallback wrote plausible, freshly timestamped, counterfeit numbers.
A fitness engine whose scores had collapsed into a three-cent band: real, computed, useless.
An evolution engine running on a population too uniform to evolve.

Nobody got an error message.

So I wrote the book. It documents 21 independent incidents at Google, Uber, Cloudflare, Knight Capital and others, plus the research on why AI-generated code is more exposed. Published 9 September 2026: 293 pages, pay-what-you-want, $0 minimum.

Get the book on Leanpub · More about it at silentwiring.com

What I do

CleanAim®

The book names the problem. CleanAim® is what I built to fix it.

Your monitoring tells you messages are moving. Your tests tell you a function returns the right value for inputs you invented. Neither answers the one that decides whether it works: is the data reaching the right handlers, and is the output real?

Verification splits four ways: structural or behavioural, static or runtime. Three quadrants are well served, some by billion-dollar companies. Behavioural verification at runtime is the one I have found no general-purpose tool for.

That is the gap CleanAim® closes, for teams shipping AI-generated code who need more than a green build. Two ways in: Silent Failure Audits and Codebase Due Diligence.

cleanaim.com

Get in touch

Write to me

Thank you — your message reached me. I will reply to the address you gave.

It reaches me directly. I read everything, and I answer most of it.