Publications and perspectives

How to Prevent Architecture Drift in Agent-Assisted Changes

Architecture drift, named in 1992, and a working answer: rules kept as files and checked on every change. A Kubernetes example and an eight-question checklist.

Kealu Vector Team, on November 21, 2026

Ariane 5 Flight 501 and the Assumption No Review Saw

The Ariane 501 Inquiry Board found traceable documents and three decisions whose reasoning stayed out of review. What it asked for, and what it means now.

Kealu Vector Team, on November 19, 2026

Change Failure Rate and Lead Time When Agents Write the Code

We recomputed DORA change failure rate and lead time on 4,545 pull requests in one public repository. Each figure moves with a choice the metric leaves open.

Kealu Vector Team, on November 10, 2026

Design Assurance Levels and AI-Generated Code Under DO-178C

DO-178C objectives stay the same when an AI coding agent writes the code. Which ones need new evidence, and where the agent sits in tool qualification.

Kealu Vector Team, on November 8, 2026

Automotive SPICE and ISO 26262 Software When Agents Write the Code

Automotive SPICE leaves tools to the team; ISO 26262 asks how an AI coding agent's errors get caught. The evidence an agent-made change carries.

Kealu Vector Team, on November 6, 2026

Independent Verification and Validation (IV&V) for Agentic Work

IV&V puts verification in the hands of an organization independent of the developer. Where each check on an agent-made change stands, and what it proves.

Kealu Vector Team, on November 4, 2026

Why Productivity Metrics Break When Agents Write the Code

In one GitHub dataset, lines and pull requests rise by multiples after developers adopt coding agents, releases by a fifth. End-of-chain measures still hold.

Kealu Vector Team, on October 25, 2026

Functional Safety and Safety Integrity Levels When Agents Write the Code

An AI coding agent leaves the IEC 61508 SIL unchanged. It cannot be qualified like a compiler, so independent means that rise with the level verify its output.

Kealu Vector Team, on October 23, 2026

IEC 62304 Software Safety Classification When Agents Write the Code

An AI coding agent does not change the IEC 62304 software safety class. What changes is the evidence, in five places, mapped to the FDA's documentation levels.

Kealu Vector Team, on October 21, 2026

How to Tell If Code Is AI-Generated, and Why Acceptance Is the Better Question

AI code detectors score near chance or fail beyond their training data. A provenance record made at the time answers who wrote it; acceptance is what matters.

Kealu Vector Team, on October 19, 2026

AI Coder: Four Categories, and Who Accepts Their Output

The four categories of AI coder, from completion assistants to prompt-to-app builders, on the V-model, with who accepts the output of each and on what evidence.

Kealu Vector Team, on October 9, 2026

AI Code Checkers: What They Catch, and What Acceptance Still Needs

What AI code checkers and AI code review catch, what acceptance needs on top, and a checklist for agent-written changes, applied to one public pull request.

Kealu Vector Team, on October 9, 2026

Agentic Engineering: What It Is, and the Acceptance Problem It Creates

Agentic engineering defined, and why production capacity and acceptance capacity run on different clocks in software whose failure has consequences.

Kealu Team, on October 1, 2026

A working terminology for agent systems

A consistent set of terms for talking about LLM-based agent systems, with explicit notes on alternatives in current use.

Fabián E. Bustamante & Stefan Birrer, on May 19, 2026