The AIWatcher Blog

Notes on running AI in production.

How teams keep AI cost and behavior under control across every coding tool they use and every AI feature in their product.

11 posts

July 17, 2026·4 min read

Why the developer-facing half of AIWatcher is open source, forever

AIWatcher Local reads your coding sessions. Here's the honest answer to why it's open source, what that commitment actually covers, and how to verify it yourself.

Read post
July 17, 2026·5 min read

The control loop for AI work: what it is, and why observability isn't it

A control loop for AI work supervises AI before, during, and after it runs: on developer laptops and inside your product. Here's how the five steps work.

Read post
June 25, 2026·9 min read

Four Failure Modes in Production AI Agents (and the Telemetry That Catches Them)

AI agents fail by running, not crashing. Four named failure modes (Retry Storms, Context Starvation, Prompt Bloat, Cost-per-Success Drift) and the telemetry signals that catch them.

Read post
June 15, 2026·9 min read

How to monitor AI agents in production

AI agents fail differently than traditional services — they don't crash, they keep working expensively or in loops. Here's how to monitor what actually matters: cost, behavior, patterns, and outcomes.

Read post
June 15, 2026·10 min read

Why AI is now COGS, not R&D: a CFO's guide to AI cost accounting

AI spend has crossed from R&D to cost of goods sold. The accounting choice changes gross margin, valuation multiples, and the unit economics conversation with your board. Here's the framework.

Read post
June 5, 2026·10 min read

How to attribute AI cost per customer in your SaaS product

A technical guide to building per-customer cost attribution for AI features — and why this is suddenly a board-level conversation in 2026.

Read post
June 5, 2026·8 min read

Why alerting on AI cost is harder than it sounds

A practical guide to catching runaway spend in real time across Claude Code, Cursor, Codex CLI, and direct API usage — before the bill arrives.

Read post
June 5, 2026·9 min read

How to track Claude Code costs across a team in 2026

Five real options for engineering leaders managing AI tool spend across multiple developers — what each one solves, where each one breaks, and what to use when.

Read post
June 5, 2026·7 min read

From token counts to business outcomes: the next era of AI observability

The first era of AI observability watched the model call. The next connects AI activity to cost, context, risk, and outcomes — from token counts to cost per useful outcome.

Read post
June 5, 2026·8 min read

AI agent monitoring is not enough. Teams need AI work attribution.

LLM observability describes model calls. Agents do work, locally and in production. The next layer ties cost to outcomes and evidence, and closes the loop by acting on what it finds.

Read post
June 5, 2026·7 min read

When 5 agents and a 30-second loop cost us $400 before we noticed

How a meeting notes pipeline ran unattended for 10 days, generated 31,762 session files, and what it taught us about the visibility gap in agentic AI.

Read post