On-screen text
Top 5 tips to build anything with Claude Code
Yesterday, Claude Code celebrated its first birthday.
After one whole year of building with Claude Code,
I'll share my top five tips for building like a 10x engineer.
I'm Pete. I own a software development agency in NYC,
we build products for startups and Fortune 500 companies.
So despite the event organisers not inviting the
Godfather of Claude Code,
yours truly, to its birthday party in SF,
I'm gonna celebrate my own way by sharing these 5 tips with you guys.
Let's start with number one,
a feature they actually dropped yesterday called remote control.
Remote control lets you connect your local Claude Code sessions to your phone
or any browser.
So you can walk away from your desk and keep working from anywhere.
Type /rc in your CLI.
It creates a secure connection.
It gives you a URL. Open it on your phone.
It launches right in the Claude Code app if you have it installed.
That's it. Now you can build on your local machine using your mobile device
while you're AFK. And by the way,
no need to take notes. I elaborated on all these things on my Substack.
Posted a free article today.
There's a link in my bio. Next tip agent teams.
So don't build alone. There are three things you need to be doing right now
to double or triple your horsepower.
First, custom sub agents.
So store specialised agents as markdown files in the .claude/agents folder.
A code reviewer, a security auditor,
or a test writer, each with its own system prompt, tool,
permissions, and model.
Claude auto invokes them when it thinks the agent would be a good fit for a task.
Second, cloud agents.
So prefix any prompt with an ampersand and it fires off a task to run
on Anthropics cloud. It runs async.
You can kick off multiple in parallel.
When they're done, run --teleport to pull the full session right back into your local CLI.
Third, agent teams.
So one session acts as team lead,
spins up teammates that each get their own context window
and can message each other directly.
Think frontend, backend,
security, all working in parallel on the same code base.
Next, #3: skills.
If you're doing the same thing twice,
you're wasting time. So skills let you teach Claude a repetitive workflow once.
It doesn't even have to be technical.
For instance,
we built a myriad of skills to collect SOC2 and ISO27001
audit evidence for a fintech client project.
It saved us months of manual work.
So make sure to skill-ify everything.
Number four app previews.
So testing is where most devs fall short.
Claude can now test its own work visually.
Five days ago, Claude Code desktop shipped app preview.
It starts your dev server,
opens an embedded browser,
and actually uses vision to look at what's rendered,
not just reading your logs.
It takes screenshots, clicks elements,
fills forms, catches errors,
and iterates on its own.
You can even click on visual elements and pass feedback directly to claude.
I mostly use a CLI
but for web work I'm seriously considering switching to desktop for this alone.
Last tip number five hooks.
I like hooks because CLAUDE.md
is nothing more than a suggestion.
Claude might ignore it, whereas hooks are deterministic.
They always run. Think pre task and post task automation.
I use them for notifications in Claude Code linting and code quality
and running tests after every code change.
The mentality you need to adopt for skills.
Do the same for hooks.
Okay, so that's it. If you're doing those five things,
you're well ahead of the pack. Again,
I elaborated on all these in my latest Substack article.
It's free, there's a link in my bio.
