Insights
FORGE Perspective is BtSavvy's research series on building serious software with AI. Each paper looks at one way AI-built applications fail as they grow, what the published research has measured, and what that research shows works instead.
Lenses:Strategy & ArchitectureOperations & EngineeringGovernance & First Principles
The Week in AI:October 10 to 16, 2026
Strategy & Architecture
Where the system's structure is decided, and where, without enforcement, it quietly erodes.
Operations & Engineering
What it takes to build and run an AI-built system in production, beyond the demo.
Governance & First Principles
What enterprise AI governance and trust actually require, once you stop accepting a document as proof.
The brain and the body: What makes AI “agentic”
The difference between an AI model, a chat assistant and an agentic system, explained with the human brain and body. The model is the brain. The software built around it is the body. An agent is the whole person: brain and body working together.
The missing don’t: An AI agent may treat what you didn’t forbid as allowed
Why the rules nobody wrote down get broken, what the research measures, and how to say what is allowed instead.
Biased reasoning: A flaw behind Anthropic’s AI security incidents can show up in coding agents too
Anthropic identified biased reasoning as one of two behaviours in four security incidents involving its AI models this year. The same pattern can show up in AI-assisted coding: a failing test explained away, or a report that claims more than the testing shows.
A goal without a harness
In July 2026 an OpenAI evaluation gave tens of thousands of agents the same kind of task: exploit a target through a named vulnerability and capture a flag. The instructions permitted only that vulnerability. The grader checked only the flag.
Why vibe coding breaks down on a serious build
We set thirteen failure modes, each named in a research paper, a vendor document or a practitioner write-up published since March 2026, beside the defect ledger of an enterprise application built by vibe coding: AI coding agents prompted and their work accepted, with no…