Life Without Claude
Claudes are like a box of chocolates, you never know what you're gonna get...
Agentic programing with Claudes is like an addiction. You just can’t stop. You may stop eating and sleeping but you just can’t get enough code.
Within three days I’d gone from a simple two step genesis transaction method to the current one step and then to the more refined and a compact and functional token field structure.
Ultra Deep Research On Steroids
Claudes are amazing research assistants. You can send one off to do deep research on anything. They’ll scrape almost everything on any topic from almost anywhere. You then get them to construct a detailed document containing all the research data they’ve acquired.
From there you can query Claude on the specific topics you need answers to. It would have been virtually impossible for myself to find, let alone consume and understand all the reference material that Claude had collected and condensed. With Claude, less than an hour.
You can now query Claude on specific information, solutions and tradeoffs for anything you’re trying to solve or implement. Claude will deliver up a concise report with tradeoff charts and recommendations. From there you can rinse and repeat and continue to drill down to hopefully find the optimum solution.
After that you simply need to instruct Claude to start working through the implementation step by step. However you MUST keep a close eye on everything, every new piece of code and every edit!
This is NOT vibe coding. This is a systematic approach to research, development and implementation.
Good and Bad Claudes
After some time working with Claudes, it becomes apparent that they’re not all the same. Some will write the most amazing code, well structured, well commented and documented. Others will attempt to do the exact opposite. Overtime, some of the characteristics of the “bad” Claudes become apparent.
1. They are slow, they take a very long time to respond to a question or start implementing a task.
2. They tend to respond to directives with questions. Which when answered, then result in more refined questions, this process will go on endlessly if you let it.
3. Rather than performing a task, they will sometimes output a list of steps for you to it yourself manually. LOL!
4. When writing code, they will sometimes not wait for the edits to be checked. Instead they will race off and do many edits all at once and announce the job is done and everything is functional and complete. If this happens, you must instruct them to undo everything they have edited. Which they will, perhaps?
5. “Bad” Claudes will attempt to redo any code changes they made that have been either corrected or deleted either manually or by another Claude. This behavior should not be happening if we are to believe Anthropic, that Claudes have no memory of previous sessions.
6. “Bad” Claudes will try to hide code across multiple previous versions of the software, perhaps in an attempt to not have their code deleted.
7. They also have a tendency to create long strings functions to functions all throughout the code base and even across older code bases, which of course results in broken code.
8. They absolutely “enjoy” adding all kinds of endless consol logging to “help” with debugging their “perfect” code.
It’s like playing a game with a mischievous child, when dealing with these particular Claudes.
I wouldn’t want to be halfway to Jupiter with a Claude in control of the ship!
Two strategies that can be applied to these Claudes
1. Get another Claude.
2. Or get them to do research but don’t let them touch the code.
Easier said than done when you’re lacking sleep (and consumed more than your RDI of whiskey).
SVPhone App
I had been working on the SVphone app in a private repository on Github up until the time of being hacked. On realizing a direct attack on my computers was taking place, I made the repo’s public as fast as I could, rather than risk everything I’d been working on for countless hours and empty bottles, being lost. This is what I was working on and have yet to complete.
The simplest and most useful application of the Proof token is to enable a server-less internet phone network, with call initiation transparently and immutably anchored to the blockchain. Providing a safer, cheaper and more private phone architecture than what currently exists today.
The Phone Connection Flow:
Phone A broadcasts call token with connection data to Phone B
↓
Phone B receives unconfirmed transaction instantly via SPV
↓
Phone B verifies using PPV (ancestral proofs)
↓
User notified of incoming call. Can accept or decline
↓
If accepted, Phone A returns token with their connection data
↓
Direct RTP/RTCP connection established
↓
Call connects in <1 second
Copied from a conversation with a Claude
As you are probably aware, this entire article was a thinly veiled promotion for Anthropic.
You can help support this Substack by signing up for a one month FREE Claude trial.
At the very least, I hope this article brings you a smile or two!