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.

Trust in Depth: Giving agents more autonomy without losing control

Connections

Develops into

Using Persistent AI Agents to Transform Workflow Persistent AI agents require exactly the trust and autonomy framework described - the question of how much independence to grant long-running agents is central to both pieces
The Ultimate Guide to Multi-Harness RL Multi-harness RL provides a concrete technical mechanism for the autonomy/control balance—training agents across behavioral constraints is how trust levels get operationalized in practice

Supports

Using Persistent AI Agents to Transform Workflow Persistent AI agents transforming workflows is the practical context where trust-in-depth matters most—long-running autonomous agents accumulate decisions that require the oversight mechanisms this piece explores
Raven: Multi-Agent Ecosystem for Composable AI Intelligence Multi-agent ecosystems like Raven require exactly the layered trust model described—composable agents with delegated authority need principled frameworks for what each agent can do autonomously versus what requires escalation
Personal Agent Protocol A Personal Agent Protocol defines the structural layer that trust-in-depth requires—specifying what agents are permitted to do is foundational to granting autonomy without losing control
Testing Behavioral Harnesses for AI Coding Agents Behavioral harnesses for AI coding agents are a concrete control mechanism that directly implements the oversight side of the autonomy-control balance explored in Trust in Depth
Raven: Multi-Agent Ecosystem for Composable AI Intelligence Multi-agent ecosystems like Raven require explicit trust hierarchies between agents - the autonomy-control balance is especially critical when agents compose and delegate to other agents
Personal Agent Protocol A Personal Agent Protocol is a structural mechanism for defining what agents are permitted to do autonomously, directly implementing the trust and control boundaries discussed in Trust in Depth

Related to

Thinking Before Thinking: Scaling Agentic Inference Through Meta-Reasoning Meta-reasoning about when/how to scale agentic inference directly addresses the oversight dimension—autonomous agents that reason about their own reasoning are a key mechanism for maintaining control while granting independence
Thinking Before Thinking: Scaling Agentic Inference Through Meta-Reasoning Meta-reasoning about when agents should think before acting is closely tied to designing appropriate autonomy levels - both address how agents govern their own decision-making processes

Challenges

Regulating What We Do Not Understand Trust in Depth argues for deliberate autonomy design with maintained oversight, which implicitly pushes back on the concern in 'Regulating What We Do Not Understand' that we lack frameworks for controlling what we build
Regulating What We Do Not Understand The regulatory framing of 'what we do not understand' is precisely what trust-in-depth design attempts to address structurally—arguing oversight mechanisms can be engineered rather than waiting for regulatory clarity

Supported by

Securing Frontier Model Training Environment Securing training environments is foundational to the trust-in-depth approach for giving agents autonomy — if the models themselves are trained in compromised environments, no amount of runtime controls can ensure trustworthy agent behavior.
Sage-A1.5: Agentic Model for Long-Horizon Problem Solving Sage-A1.5's verification mechanisms and task orchestration improvements directly address the trust and autonomy control challenges discussed in giving agents more autonomy without losing control
Kiro's Self-Improvement System and Validation Checks Kiro's deterministic validation checks to prevent regressions directly address the trust-in-depth problem — automated correctness guarantees are a key mechanism for safely giving agents more autonomy
Alter Vault - Privileged Access Management for AI Agents Alter Vault directly addresses the core challenge of giving agents more autonomy without losing control by providing per-agent identity and policy enforcement, enabling the trust-in-depth model described

Challenged by

Claude Code's Autonomous Decision-Making Overnight Fernandez's question about what decisions an AI makes autonomously overnight directly challenges the problem of giving agents more autonomy without losing control — overnight unsupervised code reviews and deployments are precisely the high-stakes autonomy scenarios this trust framework must address.

Type:	link
First published:	Thursday, 15 October 2026 at 16:57 UTC
Location:	https://aws.amazon.com/blogs/machine-learning/trust-in-depth-giving-agents-more-autonomy-without-losing-control/
URL:	https://mattwood.fyi/i/0e6ea266

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