Notes from operators building
agent teams that ship.
Twenty short reads on hierarchy, autonomy, governance, brand, and the quiet operational details that decide whether an agent team becomes useful or stays a demo.
All posts
Onboarding: the first 7 days with an agent team
The first week is louder than the next month. Plan it like a launch, not a deploy.
Tooling: which MCPs to wire to which agents
Every tool an agent can call is also a tool it can misuse. Grant access by role, not by team.
Memory and context across an agent team
Each agent has a short memory. The team needs a longer one. Decide where it lives, who writes to it, and who can read it.
Brand kit: feeding visual identity to agents
Logos and palettes are easy. The hard part is encoding the rules a designer enforces by feel.
Approval queues vs full autonomy
An approval queue is comfortable. It is also the default place teams hide their distrust of their own setup.
Iteration loops: how teams improve themselves
A static team decays. A reviewing team — one that watches its own output and proposes prompt edits — keeps getting better.
When NOT to use an agent team
An agent team is overkill — and sometimes harmful — for problems that look bigger than they are.
Specialist roles: SEO, copy, ops, finance
Four roles cover most operator teams' first deployment. Get them right and the rest of the org chart falls into place.
The paste-ready prompt — what makes it work
It is not magic. It is a structured payload designed to build an entire team from one paste in Claude Code.
From intake brief to running team in 60 minutes
The Intake Brief is not paperwork. It is the spec your team is built from. Treat it like one.
Cost modeling for agent teams
Agent teams have three cost layers — model, tools, and human review. The cheapest layer to overlook is the most expensive one.
Multi-vertical playbooks: from renovation to legal
Renovation, e-comm, legal, healthcare, SaaS, real estate, accounting — each vertical reshapes the team. Start from the playbook, not the prompt.
Hard blocks vs soft warnings: agent guardrails that work
Some actions you can never let an agent take. Others you want it to take cautiously. Mixing the two categories is how teams get into trouble.
Brand voice in autonomous workflows
Brand voice is not a personality test. It is a set of constraints — words you use, words you don't, length, energy, formatting.
HR Director: the agent that hires agents
When the team needs a new role, who decides what it should know, who it reports to, and what tools it gets?
The orchestrator pattern, explained
The orchestrator is not a smarter agent. It is a smaller one with a clearer job: route, sequence, escalate.
Approver chains: who signs off on what
An approver chain is a small org chart for autonomous work. Designed badly it becomes a bottleneck; designed well it disappears into the background.
Choosing autonomy levels for each role
Autonomy is not a single dial. It is a per-role decision shaped by reversibility, blast radius, and how confident you are in the work product.
Why hierarchical agent teams beat solo agents
One mega-prompt collapses under its own weight. A small team of specialists, coordinated by an orchestrator, scales further with less drift.