The long answers.

Start with original research and reference material, or choose a practical buying or governance question. Sources and the limits of each result stay with the article.

Original research and reference implementations

Fig. 11

One spec in. Six systems out.

One frozen specification, six implementations, different architectures and security failures. Inspect Taiga’s July 2026 experiment and its subsequent security review.

Fig. 14

Six systems passed their tests. None was safe to deploy.

Part 2 of the six-systems experiment. All 71 tests passed, yet four of the six AI-built services let anyone become an approver and five let one account reach another's data. Where variance hides in the security boundary, and why scanners miss it.

Fig. 05

HÄLYTIN: the weekend build of an app the state scheduled for 2027

Finland's drone-alert app is scheduled for 2027. The factory built a reference implementation in a weekend. What compressed was upstream of code.

Buying and evaluation

Fig. 01

What is an AI software factory?

An AI software factory turns business intent into governed, production-grade software: code, infrastructure, tests, and an audit trail. What it is and how it works.

Fig. 02

AI coding assistants vs. an AI software factory

AI coding assistants make individual developers faster. An AI software factory delivers a governed production system. How they differ, and why they work together.

Fig. 13

Ten questions to ask an AI development vendor

Vendor security questionnaires audit the organization. These ten questions audit the delivery itself: provenance, variance, the bar, and the model chain. Built to be pasted into an RFP.

Governance and security

Fig. 04

Is AI-generated code secure? What the research says

No credible study finds AI-generated code secure by default. The vulnerability rates, the flaws that are getting deeper, and the policy layer no benchmark measures.

Fig. 07

Governance debt: what the EU AI Act delay actually bought you

The high-risk deadline moved to 2 December 2027. If a calendar date was the only reason you were building audit trails, the sixteen months are interest, not runway.

Fig. 08

The off switch is not governance

Microsoft ordered engineers off the tool they chose. Uber spent its 2026 AI budget in four months. Cost is the only ungoverned AI risk that reports itself.

Fig. 03

AI coding incidents: the file

Deleted production databases, a leaked source tree, 6.3 million lost orders. Eight incidents, dated and sourced, and the control that was missing in each.

Fig. 12

Better AI models haven't written safer code

Two years of flagship model releases barely moved the security pass rate for AI-generated code. Why waiting for a better model is not a security plan.

Ideas and company perspective

Fig. 06

The Henry Ford moment for software

Ford did not invent the car. He invented the assembly line. AI writes code at machine speed; the scarce thing now is the production system around it.

Fig. 09

The condor Steve Jobs remembered

The famous bicycle for the mind chart has no condor on it. What the 1973 data really shows, and why the rider is the point for AI.

Fig. 10

Why Taiga is called Taiga

The boreal forest that survives minus forty, the Kalevala's machine that produced on its own, and the word the Finns never needed. The name, explained.