# The Radexus Journal

> Essays from inside deployments. What actually breaks, what the arithmetic says, and what we got wrong. Written by the people doing the work.

*Source: https://radexus.com/journal/*

---

Resources

# The Radexus Journal.

Essays from inside deployments. What actually breaks, what the arithmetic says, and what we got wrong. Written by the people doing the work, not by a content team.

[Institutional memory  /  28 August 2026

### Twenty people got faster. The company learned nothing.

Individual AI productivity does not compound into institutional capability, and the gap between those two things explains almost every disappointing deployment we are called in to fix.

9 min readRead the essay](https://radexus.com/journal/productivity-trap/)

[Measurement

#### We spend the first week of every engagement arguing about definitions.

It looks like the least productive week in the project and it is reliably the most valuable one.

21 August 2026  /  5 minRead](https://radexus.com/journal/definition-week/) [Institutional memory

#### A system that never forgets will eventually be confidently wrong.

Expiry is not a caveat on a memory system. It is a first-class feature, and leaving it out is why so many deployments quietly stop being used.

14 August 2026  /  6 minRead](https://radexus.com/journal/expiry/) [Plumbing

#### Four records, one company, and the two years of revenue in between.

Entity resolution is the least interesting work in the building and it is load-bearing for everything above it.

7 August 2026  /  7 minRead](https://radexus.com/journal/resolution/) [Governance

#### Insight that cannot become an action is entertainment.

The output landed in a spreadsheet nobody opened, because letting software write into the ERP was unthinkable. Here is how to make it thinkable.

31 July 2026  /  6 minRead](https://radexus.com/journal/write-back/) [How we work

#### We assert the problem instead of asking what keeps you up at night.

Discovery questions produce the answer the person has already rehearsed. Assertion produces a correction, and corrections are information.

24 July 2026  /  5 minRead](https://radexus.com/journal/first-meeting/) [How we work

#### The pilot missed its metric. We wrote it up anyway.

Both outcomes are in the contract, and the one nobody publishes is usually the more useful.

17 July 2026  /  4 minRead](https://radexus.com/journal/pilot-missed/) [Measurement

#### The four questions a CFO should ask any AI vendor.

None of them are technical. All four are difficult to answer if the answer is bad.

10 July 2026  /  5 minRead](https://radexus.com/journal/cfo-question/) [India

#### The Western cost case for AI does not transfer to India.

When labour is not the largest line on the sheet, displacing it is not the argument, and pretending otherwise is why so many Indian deployments stall after the pilot.

3 July 2026  /  6 minRead](https://radexus.com/journal/india-inversion/)