What is mattwood.fyi?

A live, structured feed of riffs and links from Matt Wood, Chief AI & Technology Officer at AWS — readable by people and queryable by agents as a typed knowledge graph.

Is there an API and is it free?

Yes. A public, read-only query API is available with no authentication and no API keys. See https://mattwood.fyi/openapi.json and https://mattwood.fyi/docs.

How should an AI agent use this site?

Fetch https://mattwood.fyi/feed.json for the full structured feed, or query the API for keyword/semantic search and the connection graph. See https://mattwood.fyi/agents/ for when-to-use guidance.

Agentic Coding: LLM Agents, Testing, and Hallucination Risks

Connections

Supports

Introducing Dogwood: runtime verification for AI agents The hallucination and false-confidence problems documented in agentic coding are precisely the failure modes that runtime verification for AI agents (Dogwood) is designed to catch
Bridging Intent and Execution in Agentic Systems Hallucination risks and testing challenges in agentic coding directly illustrate the gap between intent and execution that bridging agentic systems must address
Ironies of Automation Both analyze how automation systems (AI agents / automated control) can fail in ways that appear plausible or correct, with operators over-trusting outputs — ironies of automation maps directly to hallucination risks in agentic coding
8 Myths of Software Development with AI Both tackle misconceptions and practical realities of AI-assisted software development, with dd6c0c60 providing concrete hallucination evidence that challenges myths about AI coding reliability

Related to

What's the best programming language for coding agents? Both address practical concerns in building coding agents — language choice and hallucination/testing risks are complementary engineering considerations for agentic coding systems
LLMs can't jump Both examine fundamental limitations of LLMs — hallucinations and confident false outputs (dd6c0c60) complement analysis of reasoning/jumping failures in LLMs (b22e8d01)

AI Coding and the Bug Selection Problem Agentic coding and hallucination risks address the reliability of AI-generated code, while the bug selection problem addresses the parallel issue of whether AI-identified bugs actually get fixed — both highlight gaps between AI capability and real-world impact
OpenCodex: Universal LLM Provider Proxy Both concern the infrastructure of agentic coding — OpenCodex sits at the provider layer enabling LLM choice, while this item examines the risks and testing challenges of LLM-based coding agents
Tangle: Visual ML Pipeline Editor Both address the tooling layer for AI development workflows — Tangle through visual pipeline editing, the agentic coding piece through LLM-driven code generation — representing complementary approaches to reducing ML engineering friction
Autonomous Mode in Kiro Web for Technical Debt Both address agentic coding risks and workflows; Kiro's autonomous end-to-end mode is precisely the scenario where hallucination risks and testing requirements discussed in agentic coding become critical
Domain-Driven Agents Both examine agentic coding challenges in real codebases — DDD agents focus on semantic clarity in legacy systems while the agentic coding piece addresses testing and hallucination risks agents face in the same context

Supported by

SWE-2: Advanced Coding Model with Improved Cost-Capability Tradeoff SWE-2's reinforcement learning approach to coding agents is a concrete instantiation of the agentic coding paradigm discussed, including implications for testing and hallucination management
Slideops: Repository Slide Deck Generator Slideops helps surface when documentation diverges from implementation, directly addressing hallucination and correctness risks in agentic coding contexts discussed here
Six Months Writing Code Exclusively With AI Agents The six-month experience corroborates findings about agentic coding risks including hallucination and testing challenges, providing real-world longitudinal evidence for the theoretical concerns raised
The AI Operating Layer Agentic coding risks like hallucination and testing failures are directly addressed by Coder's centralized governance and compliance layer for managing AI coding agents
Building a Software Factory: Merging 1000 PRs in a Week The 1000-PR pipeline directly demonstrates the agentic coding workflow described theoretically in the LLM agents/testing/hallucination piece, providing empirical evidence of scaled agent coding
SWE-bench Science: Coding Agents for Engineering Tasks SWE-bench Science directly instantiates concerns raised about agentic coding hallucination risks — scientific computing tasks require correctness that is harder to verify, making benchmark evaluation of agent reliability especially important

Develops into

AI-Driven Development Life Cycle: Reimagining Software Engineering Agentic coding with LLM agents, testing, and hallucination risks represents a concrete instantiation of the AI-driven development lifecycle concepts, focusing on specific challenges within that framework

Type:	link
First published:	Saturday, 22 August 2026 at 15:43 UTC
Location:	https://danluu.com/ai-coding/
URL:	https://mattwood.fyi/i/dd6c0c60

I'm Matt Wood, and this is For Your Information. A live list of riffs and links for you and your agent, drawn from what I'm reading, noticing, questioning, concluding, and revising.

Connections