Today, the coder in me awoke like it had just been in hibernation. A strong urge to look at Claude's code, in IntelliJ IDEA. The body relaxing as I cmd+B through the code again.
Asking Claude to red team an idea with agy and codex is proving very valuable. I don't enjoy the friction as much as the dopamine hit of fast execution, but the solution is going to be so much better.
@stijnnoorman Then share your solve. That's been the missing piece for me. Have been building and problem-solving my whole life, hardly sharing the discoveries with anyone. Now - thinking that's a real shame.
Asking Claude to red team an idea with agy and codex is proving very valuable. I don't enjoy the friction as much as the dopamine hit of fast execution, but the solution is going to be so much better.
So how do you allocate time between playing the long game and shorter -erm games in your business?
Like writing, SEO, creation of ever-green offers vs getting on the phone and closing a deal today.
Let things happen more organically. Trying to control the message, the audience, the sale - being stuck in my ways is static, is hard. Letting go means also letting unexpected, pleasant side effects happen.
Realizing I've spent the last twenty years working on reducing friction in work, by doing the hard part up front, by abstracting away complexity. From designing APIs to building "lego blocks" for course creation and sales funnels.
Bad OODA loop: You set out to create something, see someone else doing something that seems smarter (greener grass syndrome), so you change what you're doing before you get results, in a loop.
Software evolves, every line can be rewritten, but when you write a book or create a course, they get old fairly quickly. How can we make it as easy to keep a book or a course evergreen, as it is to keep working on existing software, shipping new versions continuously?
Every time I have finished creating something, I have ideas on how to improve it.
So instead of shipping v0.1, I keep building. And building. And building.
I need to start shipping more often, get feedback from real users, then improve.
(Sure, I'm user #0, but still ...)
In my software factory, the bots have to *prove to me* they've done what they were supposed to.
Now I'm building it so a new, adversarial AI will check the proof - *not the code*, towards the spec.
A year ago, even with Claude, I had to do a lot of copy+paste to give Claude the info it needed. Now it can log in anywhere, fetch whatever's needed to debug, collate info from all the sources. It's easy to forget how incredible this tools has become so quickly.
I'm working on having my implementer run on multiple machines, waiting for fully-specced work to do, then one machine 1) claims the work, 2) does it, 3) reviews and re-works, and then 4) creates the PR, which simply becomes *proof that it works*