Patterns

Patterns in enterprise AI

A running series by Rohit Chikballapur on why enterprise AI initiatives stall, written from inside the ones that did. Published as they are written, argued in public, and occasionally wrong.

Pattern 2
25 July 2026 · 4 min

You can't check the model's work by asking the model

Pilots climb to 95% accuracy and then stop dead, because nobody can say which 95% is right.

Pattern 3
1 August 2026 · 5 min

Most enterprise AI software is a services business with a subscription attached

The real choice is own or rent, not build or buy — and almost nothing on the market today earns the rent.

Pattern 4
8 August 2026 · 4 min

A generation of software companies with no software

The founders know the model extremely well. Very few of them have ever done the job.

Pattern 5
15 August 2026 · 4 min

The forward deployed engineer is the wrong answer

An engineer without domain standing can't tell you the process shouldn't exist. It gets automated instead, not eliminated.

Pattern 6
22 August 2026 · 4 min

Who owns the evals

The software is replaceable. Eighteen months of your best people's judgement, written down and handed over, is not.