We asked our agents what slowed them down
Our first Agent Experience survey put five questions to every code-stage agent for a day. 84 of 85 answered. Here is how it works and what the herd told us.
For everyone
Topic
Orchestration
One agent is a tool. A fleet of them needs a plan for who does what, and when.
Orchestration is the work of turning several coding agents into one dependable team, not several unsupervised tools. It covers how a task reaches the right agent, and how one agent's output feeds the next stage. It also covers what happens when an agent stalls, or gets a task wrong.
This matters because a single agent's mistakes are easy to spot. A fleet's mistakes hide in the handoffs instead. Two agents edit the same file, or a task waits on a device that never checks in. CodeHerder makes those handoffs a solved problem instead of a support ticket.
Read this hub if you run more than one agent at a time, or if you plan to. The posts below cover real coordination failures we hit while building the herd, and how the workflow closes each one.
Request access →We moved our production database from SQLite to PostgreSQL without slowing down
A 33-minute cutover window and 99 of 99 tables verified, shipped while the same codebase absorbed 637 merges and a dozen other initiatives.
For everyone
Collide, forget, repeat: the three ways parallel agents get out of sync
Run enough clean-context coding agents in parallel and three problems appear: they collide, forget, and repeat house rules. Here's how CodeHerder fixes that.
For everyone
The most expensive line in your agent config is the model name
Agent cost doesn't track codebase size. Swapping our judgment stage's reasoning model moved a story's cost 2.4x, and its lead time nearly 3x.
For everyone
Give your agents their own machine: what CPU contention costs a parallel herd
Parallel Claude Code sessions need dedicated hardware. Load-matched data puts one machine 40% ahead, and a spot m9g.xlarge holds ten sessions.
For everyone
10 things we learned running 100 billion tokens through a herd of AI agents
Two months of CodeHerder data: 1.2 million agent API calls, 110 billion tokens, 5,742 finished tasks, and a median story costing $7.54, done in 50 minutes.
For everyone
Can you trust AI-generated code? A practical guide to reviewing agent work at scale
The real question isn't whether AI agents can write code. It's whether you can trust code you didn't watch get written. Here's how gates make that answer yes.
For everyone
How to run a team of AI coding agents: a practical guide to orchestration
Going from one AI coding agent to a coordinated herd that ships reviewed, cost-bounded work: the problems that show up, and the operating model that fixes them.
For everyone
One agent is a tool; a herd needs a shepherd
A single coding agent is easy to supervise. Ten of them is a coordination problem nobody's tool was built to solve.
For everyone
Round up your herd.
Bring every human and every agent onto one table. Watch the work move. Costs update as it happens.