HN user
contact via https://ghuntley.com
there’s a difference between software development and software engineering. everyone is a software developer now, engineering is hard - always has, always will be but the skills to be an engineer are now very different.
the back pressure terminology comes from me. essentially it’s the wheel - you need to add backpressure to the agentic flywheel.
see https://ghuntley.com/pressure
i have the pleasure to work with moss and he came up with a way to explain what is in my head with ease.
no? i don’t work at Amp. that was a long time ago.
source code at https://github.com/schani/ralph-os
even then using vscodium does not free you from the grasps. vscode by design is designed to fracture - see https://ghuntley.com/fracture
it isn’t http://ghuntley.com/fracture
forking vscode? simple. extensions not so simple. they are controlled by microsoft. without them you’ll run into continual papercuts as a vendor who has forked vscode.
Essentially this is Ralph as a service - https://ghuntley.com/ralph
This is the 9-month recap of my "The Future Belongs to People Who Do Things" talk.
Inside:
- The problems with AGENTS . md
- The problems with LLM model selectors
- Best practices for LLM context windows
- AI usage mandates at employers
- Employment performance review dynamic changes
- The world's first vibe-coded emoji RPN calculator in COBOL
- The world's first vibe-coded genz compiler (CURSED)
and a final urge to do things, as this is perhaps the last time I deliver this talk. It's been nine months since the invention of tool-calling LLMs, and VC subsidies have already started to disappear.
If people haven't taken action, they're falling behind because it's becoming increasingly cost-prohibitive to undertake personal upskilling.
Well, actually, the funny story is Cursed is actually three different compilers:
1. It was first written in C
2. Then it was rewritten in Rust
3. Finally, it was rewritten in Zig
All part of research, learning how to drive these models and discover their underlying behaviours. I reckon you could get a comparable compiler going in under a month or less for less than $4k USD.
It has completely rewired my brain.
Location: Sydney, Australia
Remote: Yes (last 7 years; globally)
Willing to relocate: No (max 1 week a month out of Australia as required)
Technologies: https://ghuntley.com/
Résumé/CV: https://linkedin.com/in/geoffreyhuntley
Email: ghuntley@ghuntley.com
A principal software engineer who is known to write and do conference keynotes about AI. Skillset-wise - a weird mix of software engineering, infrastructure, engineering management, and marketing with a security background. My current focus is automating job functions with AI and teaching people how to do it. I'm currently conducting interview loops.
So it’s better to work on problems that get progressively more complete towards a known objective rather than problems that require exploration.
Yes. The longest I've had a self-directing agent loop running is a cumulative of three months. One goal, one purpose. Every now and then I modify the prompt in the background, and the agent picks up the updated prompt on the next loop.
Fair
And yet, the original post, which has been on the front page of Hacker News for 18 hours, is based on techniques from my blog that you're degrading.
Yes, but the cooked thing is you just run more loops with the right prompts and you can resolve defective outcomes. It's terrifying
Yeah, I totally get it man. Like when I discovered these techniques back in February, it really spooked me. And I guess 5-6 months later it's starting to become known even though I've been publishing full details on how to do this shit. It turns out that the best way to spread knowledge in San Francisco is to get drunk with YC founders...
Probably the second. I first discovered this around about March. It's kind of hilarious.
Nice. Any chance you could put in some attributions and credits in your paper? https://orcid.org/0009-0007-3955-9994
It gets kind of philosophical really fast. What does it mean when software can be automated through a bash loop? (Not to 100%, to 80%. What does that mean to software outsourcing in the consulting industry?)
If not, just port and move on.
Exactly the point behind this post https://ghuntley.com/libraries/
It's both serious and a joke. The seriousness is that it works (to point) and the implications to our profession as software developers. The joke is just how stupid it is. Refer to the original story link above for proof of outcomes.
Ralph is a technique. The stupidest technique possible. Running an agent in a while true loop. https://ghuntley.com/ralph
- weird copyright / IP questions all around. This will be a minefield.
Yeah, we're in weird territory because you can drive an LLM as a Bitcoin mixer over intellectual property. That's the entire point/meaning behind https://ghuntley.com/z80.
You can take something that exists, distill it back to specs, and then you've got your own IP. Throw away the tainted IP, and then just run Ralph over a loop. You are able to clone things (not 100%, but it's better than hiring humans).
It is, isn't it mate? Shit, I stumbled upon Ralph back in February and it shook me to the core.
Strangely enough, TLA+ and other formal proofs work very well for driving Ralph.
It's why I called it Ralph. Because it's just not all there, but for some strange reason it gets 80% of there pretty well. With the right observational skills, you can tune it into 81, then 82, then 83, then 84. But there's always gaps, always holes. It's a lovable approach, a character, just like Ralph Wiggum.
Nice. Check out https://ghuntley.com/ralph to learn more about Ralph. It's currently building a Gen-Z esoteric programming language and porting the standard library from Go to the Cursed programming language. The compiler is working, I'm just finishing up the touches of the standard library before launching.
The language is called Cursed.
making all this fault tolerant with good ux is lot of work.
Yes, it is. Not only in the department of good design in UX, but these LLMs keep evolving. They are software with different versions, and these different versions are continually deployed, which changes the behavior of the underlying model. So the harness needs to be continually updated to remain competitive.
Yep, once you've got the base coding agent (as in the workshop above), you can use it to build another agent or anything really. You start from that kernel and you can bootstrap upwards from that point forward and build anything.
Yep. What you need to do is set its direction and then blow wind into its sails.