Future Infinitive ☸️

Idk I hope rocket money is the end of them

yeah we did actually have to reallocate the budget so my heart goes out to all the hardworking folks up at Netflix and Amazon

Got me right in the feels y’all plz donate #pbs #npr #publicbroadcasting #pbskids #cpb #publictv #kidsmedia

I think I will do some videos on it, people clearly need examples, but also probably running a class cuz seeing me do it in on a video vs a group setting is kinda key, I have tried to teach it with a video before and... idk

ah docs for what- the software development lifecycle (SDLC) or AI and spec based dev? this represents literally the absolute cutting edge of AI capabilities there are no docs and it took me 25 years of very weird and hard to replicate experiences to distill it

these are the docs

as usual this is chatgpt's zine version of a class I teach, I made it to give away as free learning materials and you may do with it what you wish, under the condition that you not complain about the use of ai to me (I will literally send my agents after you)

anyway this is some dos and donts

there are some security implications, use your best judgement when installing unreviewed code directly from the internet, and if your judgement isn't the best don't install stuff

also you can write your own

basically it allows you to let your ai use websites or whatever apps in a stable way

Claude, Cursor, lm-studio, every single agent framework, and a lot of AI assistants support this

ok lets do one more zine, on Model Context Protocol

The standard model? If you want it to explain *everything* you’re gonna need to upgrade to the premium plus model

Alright I’ll see you there maybe? #oakland #sf #bayarea #rave

Ok the smartest kid I know is obsessed with them, is this their generation’s animorphs?

What’s the hardest problem in computer science?

Incrementing a race conditions caching number by one

yo thank you for the kind words I love that this has helped you! Let’s learn clean open source AI together folks #llm #ai #environment #cleanai #professor #genai #chatbot #claude #chatgpt #opensource#greenscreen

yes can confirm this is happening, also it is def job security for #cybersecurity #infosec folks #jobsecurity #vibecode #teaapp #llm #ai #jobloss #layoffs #devs #databreach

All materials I make and release to the public with ChatGPT are free as in beer and free as in speech and free as in you're free to do whatever you want with it but by using it you agree that you won't bother me about it except to say thanks

Good agentic software development is, surprise, just good software development. And by that I mean rigorous and disciplined and well described with good process flows and a clear definition of what is done and secure. Code basically writes itself now- everything else was always the hard part.

The same things that matter for keeping a lot of software engineers working well together is the same stuff that makes a single agent *super* effective.

Oh and then after you lay out the plan, finally, sticking to it? That's the hardest thing. Just one more thing for this PR and *then* I'll stop. That's what you said last commit - make your changes tiny and trivial to revert.

yeah software is a lot of little tasks. So many that it will drive you mad. This is why we have so many lists. Even our lists have sub-specs about how to make the lists, and this is honestly to *save everyone cognitive effort* and it *matters that you do this*.

So you need rubrics and checklists or your own process will drift. Software engineering is less of a science and more of a discipline, which is why so many of us are literally ninjas (I am sorry to report I am actually not joking here despite what you might think about an engineer you didn't like)

The frustrating thing is that getting good results out of a language model is as hard as it is to get good results out of an engineer, and that is why engineering management is also a skill that basically only engineers are good at. Also your brain gets tired with too much context.

Everyone seems to think: My job is the only hard job- everyone else's job is stupid and easy and it is obvious to me that I would be immediately good at it without trying

All the things humans have to do (eg code review, peer review, checking again, re-checking a third time) are critical to getting robots to do the same job.

Surprising no one: if you don't know what you want you won't get it, engineering & management still a needed skill despite robots being able to do it

ok I made another zine about the agentic SDLC

Evergreen post idk we’ll see