A Practical Method: Build Verification Into the Loop

Agent output is a first draft, not a conclusion. In many cases, the fastest way to trust it is to attach clear acceptance criteria and have the agent produce evidence: automated checks, validation results, and targeted verification. This approach keeps iteration cheap, makes failures obvious, and preserves human accountability with artifacts you can review.

Habits:

• Add explicit acceptance criteria to agent prompts (inputs, expected behavior, and “done” conditions).
• Require the agent to run automated checks and report results (or explain constraints when it can’t).
• Make verification checks the default for changes; have the agent add or update checks alongside the work.
• For behavior that spans components, use integration or end-to-end checks where useful (smoke tests beat vibes).
• Ask for a short verification summary: what was checked, what passed/failed, and what remains unverified.
• Keep cost bounds: cap iterations/compute, and escalate to human review when the loop isn’t converging.

A Practical Method: Asynchronous Agent Runs With Review Gates

In many cases, agents are most useful when they can produce a first draft without a human in the chat loop the whole time. Treat these runs like any other automated system: define the inputs, set constraints, execute in a controlled environment, and require verification before changes land. The goal isn’t autonomy for its own sake, it’s predictable progress with clear accountability.

Continue reading...

A Practical Method: Build a Verified Template Library

When a task repeats, you can capture the pattern as a template that produces a solid first draft. Templates make expectations explicit: inputs, constraints, checks, and what “done” looks like. The goal is not to automate judgment, it’s to reduce rework and make outcomes easier to verify. Over time, the library becomes a shared baseline that teams can use and improve when appropriate.

Continue reading...

I wanted a simple way to write blog posts without opening VS Code. Obsidian turned out to be perfect for this.

Setup

1. Open your Hugo repo as an Obsidian vault - the whole repo, not just content/. This lets you work with images in static/ too.

2. Add .obsidian/ to your .gitignore - Obsidian creates a config folder in your vault. Ignore it so it doesn’t clutter your repo.

3. Install the Templater plugin - This lets you create templates with dynamic content like auto-filled dates and titles.

   Continue reading...

A Practical Method: Make Context a First-Class Input

Agents are most effective when they start with the same inputs a human would use: goals, constraints, relevant code and docs, and the right tools. The output is usually a first draft, so aim to make it easy to verify and hard to misunderstand. A good default is to ask: “With the context, constraints, and tools provided, can this task be completed without guessing?”

Continue reading...