agent-operations
Articles
Template Gives You Consistency, Not Credibility
Per-vertical content a practitioner respects cannot be stamped from one template. Depth at altitude takes serial authoring through an adversarial critique loop.
Ground It or Flag It: Ending Agent Confabulation
The agent failure to fix is not guessing - it is confident, unmarked inference dressed as fact. The cure separates disclosure from verification.
An Agent Should Write the Script, Not Be the Loop
When an agent task burns money, the reflex is to make the run survivable. Often the real fix is upstream: the agent should write the script, not be the loop.
We Accidentally Built a Shadow System of Record
A review caught our agent console rendering the customer's own data back to them. We had built a shadow system of record nobody designed.
claude.ai as a Routing Surface, Not a Content Generator
MCP turns a conversation interface into a capable operator by exposing production tools directly - the LLM becomes the routing layer, not a content generator.
Overlay Drift: The Fleet Observability Gap That Looks Like a Deployment
Deploying a new overlay version does not mean all instances have it - persistent processes require active drift detection, not deployment confirmation.
An Agent That Can Edit Its Own Autonomy Ceiling Has No Ceiling
OS-level file ownership - not prompt instructions - is the only reliable way to prevent an AI agent from modifying its own governance constraints.
Live Reconfiguration for Persistent AI Agent Processes
How to apply configuration changes to a running AI agent without downtime, and why "durable" apply semantics matter more than the detection mechanism.
Model Selection Is an Operational Variable, Not a Code Constant
Config-driven model selection gives per-customer cost control, live tier changes without redeploy, and a verification path for actual API spend.
Suppress Work Loudly
When a suppression mechanism fails, the failure mode should produce more work, not less - because silent suppression looks identical to a successful decision.
"Merged" Is Not "Wired"
Live output path audit - not PR titles - reveals merged safety-critical features that never reached the path real output actually travels.
Dead Code Removal Is a Security Audit, Not Just Cleanup
Deleting old adapter code forces a semantic comparison you would not otherwise do - and that is where security gaps surface, not where they get introduced.
Building a Product's Entire UI in a Two-Day Parallel Agent Sweep
A client portal and an admin console, built in two days by agents working in parallel, each owning one surface, on a shared dual-mode abstraction.
Putting an Agent at the Top of the Funnel Instead of a Contact Form
A prospect talks to an agent and gets a real findings report. The eval harness, adversarial personas, and voice bridge behind a generative top of funnel.
Parallel-session worktree isolation: when one agent's `git add -A` sweeps another's edits
Two sessions on the same repo. One runs git add -A. The other's uncommitted work ships in the wrong PR. The fix is filesystem isolation, not branches.
Shipped Friday, retired Monday: kill-discipline in four days
A Phase 1 feature shipped end-to-end Friday. Monday a P0 portal bug closed it. Four days from launch to clean removal.
Plan-declared workstreams: every plan ships its own execution topology
Replacing ad-hoc agent coordination with a fenced YAML execution-mode block at the bottom of every approved plan. One gate, plan-internal parallelism.
How an agent's friction report became a fleet-wide git rubric
An agent complained about rebase and force-push policy walls. A four-PR arc shipped a documented branch-class rubric every session inherits.
Validating pattern-matchers against a real corpus, not your imagination
Our v1 reflex hook had zero corpus recall. v2 was corpus-driven from the start. Mine real transcripts before shipping pattern-matching code.
When one tenant's CI noise blocks all the others
An agent in a clean session saw 153 unresolved CI alerts that belonged to another venture. Per-venture scoping plus a rollup view fixed it.
Per-minute session activity from JSONL: when heartbeats lie about idle time
Heartbeats only fire on /sos and /update. Per-minute activity from Claude Code JSONL gives sessions an honest activity timeline.
Killing Skills on Purpose: A 32-to-24 Retirement Cascade
How a full-day audit cut a 32-skill catalog to 24 - and the operational tests that determined what earned retirement vs. what earned a stay.
Shipping a Cultural Directive Your Agents Read at Session Start
Captain's standing order kept getting forgotten. The fix: commit it to a file agents read at session start - and learn what text belongs there.
What Happens When Your AI Agent Briefing Lies
The session briefing said 10 alerts. There were 270. Fixing it took four tracks, $741 in compute, and a captain who refused to accept done.
Cross-Venture Context - Teaching Agents Where They Are
Agents operating across multiple products need spatial awareness. Without it, they target wrong repos and leak secrets across contexts.
When Your Agents Spend 40 Hours on One Auth Bug
Four root causes, 40+ hours of agent time, one MCP auth failure. The final fix was a one-line CLI command.