Systems field-note hub
AI Workflow Automation and Internal Systems
Systems work starts after the prototype succeeds once. This hub tracks workflow automation, internal tools, monitoring, costs, API behavior, and the boring recovery paths that turn an agent setup into infrastructure.
What this lane tracks
Commercially useful patterns pulled from live build work.
AI workflow automation that can be debugged in minutes
Internal tools, monitors, and approval paths around agent work
Business process automation with logs, ownership, and rollback paths
Latest Systems notes
26 crawlable notes in this category.
AI Agent Tool Execution: When Parallel Breaks Things
Parallel tool execution cuts latency by 40%. It also introduces race conditions, silent state corruption, and double-charge bugs. Here's the operator's guide to getting it right.
Read the field note →Dangerous Command Detection for AI Agents: What Claude Code Got Right
Claude Code showed how useful pre-execution dangerous-command checks can be before shell access. An open-source framework just adopted the same approach. The lesson: agent safety needs pattern recognition at the command level, not just sandboxes.
AI Agent Guardrails Are Too Aggressive: The False Positive Problem
AI agent guardrails fire false positives on legitimate actions. Operators bypass them to keep shipping. Here's why it happens and how intent-aware gating fixes it.
Why Most AI Agent Setups Fail Within 48 Hours
AI agent setups die in the first 48 hours — not from weak models, but from config races, over-aggressive guards, first-run friction, blind observability, and missing approval gates. Here are the patterns and the fixes.
Agent Config Packs: Stop Hand-Assembling Every AI Agent From Scratch
After the third time I copied a SOUL.md from one agent to another, I stopped hand-assembling and built a config pack. Here's the anatomy, the rules that hardened over three months of fleet operations, and the threshold for when you need one.
One Gateway, Many Agents: The Control Plane Pattern for Agent Teams
Multi-agent systems need a gateway control plane before they need more personalities: routing, identity, audit trails, and blast-radius boundaries at ingress.
Silent AI Agent Failures
Agent systems fail silently: dropped messages, invisible-unicode cron blocks, and reasoning echo-back loops that treat a model’s own output as new facts.
AI Agent Web Tools Need Failure Budgets, Not Happy Paths
The six major AI agent frameworks still ship web tools built for demos, not production. Browser automation fails like infrastructure. Your agent needs partial results, typed errors, failure budgets, and traces that survive the unhappy path.
Prompt Injection for Tool-Using AI Agents: A Security Checklist
Prompt injection turns dangerous when agents read untrusted content and call tools. Use this checklist before granting file, API, or message access.
n8n Self-Hosted Agent Automation: Why I Moved Off Make and Zapier
I ran my agent orchestration on Make.com for six weeks. Then I hit the operation limits, the missing version control, and the opaque errors. Self-hosted n8n costs $6/month and handles whatever throughput you throw at it.
Human-in-the-Loop AI Agents: Approval Gates That Make Automation Useful
Human-in-the-loop AI agents work when approval gates sit at publish, spend, delete, send, merge, and production boundaries.
AI Workflow Automation for Small Teams, Without the Science Project
AI workflow automation works for small teams when the workflow is scoped, logged, reversible, and owned by an operator instead of treated like a magic agent demo.
AI Workflow Automation With Existing Tools: When Not to Add n8n, Zapier, or Another App
Before adding another automation app, audit whether Gmail, Sheets, Airtable, HubSpot, Slack, and current APIs can handle the first safe workflow.
AI Agent Sandbox Checklist: Files, Shell, Network, Secrets, Rollback
A practical AI agent sandbox checklist for file access, shell commands, network calls, secrets, approval gates, and rollback before agents touch production.
AI Agent Handoffs Need Receipts
AI agent handoffs fail when the next worker has to trust a summary without proof. Receipts turn multi-agent work into an audit trail: artifacts, commits, task IDs, tests, screenshots, and blockers.
The Economics of Running Your Own AI Agent Fleet
What drives AI agent fleet cost: model calls, orchestration, context, failed runs, human review, and the maintenance work that keeps the fleet coherent.
Agent Runtime Config Migrations Need Rollback Plans
Agent runtime config migrations fail quietly when they rewrite files without dry runs, diffs, backups, validation, and a verified rollback path.
Top 7 Multi-Agent Orchestration Patterns
Seven multi-agent orchestration patterns I use, where each breaks, and when to choose it.
n8n vs Zapier vs Custom AI Agents: Which Automation Path Fits?
A practical decision guide for choosing hosted automation, self-hosted workflows, custom AI agents, or no automation yet.
Why AI Agent Setups Fail Within 48 Hours
AI agent setups fail fast when they lack durable state, ownership rules, recovery paths, and approval gates. Here is the 48-hour test I use.
Monitoring AI Agents in Production: What to Watch
AI agent monitoring should start at the task layer: outcomes, tool calls, token spend, context pressure, delegation depth, approvals, and delayed quality.
Self-Hosted AI Automation With n8n: A Practical Setup
A practical self-hosted AI automation setup with n8n: webhooks, model calls, review gates, workflow logs, and the parts I keep outside SaaS.
What It Actually Costs to Run AI Agents: A Practical Breakdown
Most AI agent cost posts quote enterprise prices. Here's what a solo operator actually spends running a multi-agent org, with real numbers for token costs, infrastructure, and the tradeoffs that matter.
Chat Completions API: What I Learned Running It in Production
The Chat Completions API looks simple until you add tools, memory, and real users. Here's what I changed to make it hold up under production load.
ChatGPT System Prompts That Survive Production
My AI agent kept drifting in real workflows. Here is what I changed in the system prompt: boundaries, tests, tool rules, escalation paths, and versioning.
AI Agent Workflows That Actually Ship
My agent workflows kept breaking in production. Here's what I changed: orchestration patterns, knowledge structure, and the memory layers that made the difference.