Your own toolbox

From using tools to building them

So far you have got the most out of what your agent ships with, plus whatever other people have made available. That gets you a long way. But at some point you hit a wall that is not about the model at all. Your agent cannot reach the system you live in. Or it can, but working with it is clumsy, and you repeat the same setup every time.

This track is about fixing that. You build the environment you and your agent work in: the connectors it reaches through, the jobs it runs on its own, the rules it follows, the checks it holds itself to. Same agent, better place to work.

You will point Claude Code at each job and stay in the loop while it does the typing, just like in the vibe coding track.

What goes in the toolbox

To get your own toolbox going, this is what the track covers:

• Memory. What your agent knows about you and your work, so you stop re-explaining it every session.
• Connectors. A way to pull data from your systems straight into your agent, and to let it act in them.
• Hooks. Steps that run by themselves the moment something happens, so you stop repeating them.
• Delegation. Whole jobs handed to agents that work while you do something else, one at a time or ten at once.
• Rules. What your agent is allowed to touch when you are not there to approve each step.
• Checks. A test your agent has to pass before it hands work back, so bad output does not reach you.
• Taste. Your standards, written into a file your tools read, so they stop living only in your head.
• Plugins. One package holding all of the above, that a colleague installs in a click.

Each of these is useful on its own. You get the most out of your work when you bring them together into one coherent system, and that is what this track is built around.

Nobody designs the finished version up front, so do not try. Every good setup starts as one file fixing one annoyance, and grows a piece at a time as things go wrong. Start where it hurts.

Hands-on

Reflect

• Which system do you touch every week that your agent cannot reach? Hold onto it. You will build its connector, automate it, and hand it to a colleague before this track is done.