Insights
Short pieces, every number traceable to a named public study. We write when we have read something that changed how we work, which is less often than a content calendar would like. There is no newsletter and nothing gated.
What we have been reading, and what we changed because of it.
Why most enterprise AI pilots produce nothing
MIT researchers found 95% of enterprise GenAI pilots showed no measurable P&L impact, and Gartner expects over 40% of agentic AI projects to be cancelled by 2027. The blocker is not model quality.
Read it →Twenty percent faster, nineteen percent slower
A randomised trial found experienced developers were 19% slower with AI tools while estimating they were 20% faster. The 39-point gap is the finding that matters.
Read it →A year of model progress did nothing for code security
Veracode tested over 100 models in 2026. The security pass rate for AI-generated code moved from 55% to 56%. Syntax improved, security did not, and the gap is a review problem.
Read it →Four rules, mostly to keep ourselves honest.
- Every number has a named source with a date. If we cannot link it, it does not go in.
- We say when a source is weak. A working paper is not a peer-reviewed study and we will tell you which one we are quoting.
- No number is quoted without the thing it contradicts. A statistic that only supports our own sales case is a marketing asset, not an insight.
- Nothing is gated. There is no form between you and any of this.
Bring us the problem, not the spec.
Thirty minutes, no commitment, and a straight answer about whether one deployed engineer is the right shape for what you are trying to do.