AI Backlogs and Tool Ownership – A-CX Weekly
This week at A-CX, Ilpo has been writing about a new rule for AI-filed issues, and Ville has been thinking about who owns an AI tool after the pilot.
AI-Generated Backlog
Ilpo has been writing about the issues our AI agents filed on their own. He explains how we moved the filing rule to the point where issues are created, so each one must cite a real past consequence. Check it out here.
Test Isolation
Our development team has been focused on keeping AI generated tests from touching production data. A test that can reach live systems works fine until the day it cleans up or overwrites something that was never its own. It is worth looking at what your own test and staging runs can reach, and getting in touch if you would like help drawing that line.
Thoughts and Technology
Ville shared some thoughts about the question he keeps hearing at events: who maintains an AI tool once the models change, and who gets the call when it breaks. His answer is a clear owner, one view of every tool, and a plan for when a model changes. Read Ville’s post.
A-CX has been considering why meeting in person matters more as AI makes content cheap: trust still comes from people, and the best feedback comes from across the table. See what we shared.
What We’ve Been Reading
AI coding tools have made writing code fast, leaving planning, review, and deployment as the slow parts, as laid out in this playbook by Louis Claxton, with six stages that each end by committing an artifact, such as a spec or plan, for the next to read.
Bigger context windows do not guarantee better answers, as shown in this blog post by Drew Breunig, which names four ways long context breaks agents: poisoning, distraction, confusion, and clash.
AI is the answer that dominates a survey of maritime technology leaders, but this article from Splash247 argues the payoff depends on reliable data, shared context, and integrated workflows first.
