
Dev Researcher | C++ Dev | ex-MSFT .NET PM Frmr Game Dev. I alt+tab to @ProdToProd Screaming into the X pillow & listening to the noise https://xitter.twitterapi.xyz/t.co/91ThNnhPeA
Brisbane, Queensland
Joined March 2007
- Tweets56.7K
- Following1K
- Followers5.3K
- Likes6.3K
We've given all Pro, Max, Ultra, and Team users one free full reset, across all models and limits. Use it anytime before December 10 at 2:00PM PT.
We recently fixed a few small issues that used more limits than intended, and the Fable resets didn't do much for those of you who'd already hit your weekly limits for all models. While these issues didn't affect everyone, we want to make up for the trouble.
You'll find it in Settings > Usage, or at clau.de/reset.
oh shit... now you picked a fight with @thsottiaux ...show em Tibo... make em pay, hit the reset button full week plus give 3 full resets.. show em your pain threshold is much higher... gawn.... do it..
35
For some reason... I remembered I did this podcast in 2015 with @michael_wise ..
We painted this vision of how software would change and .....re-listening again... omfg... we called Ai 10 years ago .... damn.... feels more like an accident though π
spacewelders.com/episodes/epβ¦
69
Hear me out:
We hated "waterfall" because after Big Upfront Planning, implementation took FOREVER, and by the time we were done, the plan was out of date
But with AI, implementation is rapid (100x faster, or more, than before)
So Big Upfront Planning can now make sense?
I mean.. can we all call it out for what it is atm
There is a striking similarity to the ye olde Colgate Toothpaste trope and AI Usage today.
Story is simple, people believed that Colgate had increased their diameter of the Toothpaste by a few mm so users without realising consumed more toothpaste.
It never happened but....
Fast Forward to today, Frontier Models are adopting the exact same fking strategy imho.
Here's the new model, omg look at how shiny it is.... whats that... you're no 20% usage in the first day... oh well.. what could of caused that I wonder.....
Somehow the new models "use more compute?" right???
RIGGHHHT?
1
217
Notes from the Jobmarket
------
Most jobs I've gotten calls for are:
1) Help make make my team scale with AI via can you either invent an AI-DLC or use one they've started or find us one
2) Build me a harness/software factor that makes us scale
3) Help me keep my budget under control. I am spending $$ on per token usage and can't us Subscriptions because privacy concerns
My Raw Notes
------
- If you're a solo dev with AI subs its easy to make end to end things. If you have to do this with a team or play with your ties in a bigger sanbox... there be fights and dragons in first 1-3 months
- Harness/Software Factory is hit and miss. It works great on greenfield but brownfield there is a lot of setup foundation work to bootstrap one in place. I think people just assume they download an AI-DLC harness/factory point at brownfield and say "make new" and it just works. It will do a lot quickly and give you a sense of "yeah its working... see.." but again, its a lie. As many say "junk in, junk out" still applies
- Product Management ... Years ago I went from a Developer Evangelist over night to Product Manager at Microsoft and I didn't suddenly "know" how to do it. I had to literally shadow brilliant people to figure it out and find a path here. I didn't realise it but now that skillset is coming back today, as that prompt in CLI terminal demands you give it something to chew on thats well thought out. No amount of skills or bullshit markdown files are going to solve this problem, you either know or you don't. Immature product management is what I think kills AI projects fast...
- YouTube rots your brain. Stop watching youtubers talk about AI vibe coding, harnesses, factories etc. 90% of them are just echo chambers and full of bullshit techniques that don't last beyond a demo SaaS solution. Good demo, great demo, production, not really. I think most already long into adoption have figured this out but the new comers.. just stawp..
- SaaS UI is done. No its not. You can spot a AI saas solution fast. For example AI doesn't know how to do gaps/spacing/padding correctly. It assumes everything generally is even and lives in odd pixel structures. A Good SaaS designer can also manage your behaviours, grid, spacing, tones and so on. Its night and day when you hand an Ai-gen SaaS to a designer. They come back with way noticeable differences. However, AI can throw out a good enough via shadcn anyway so its down to "it works kinda"
- Vibe Code works only when you invest in the foundation. I have it now setup where it produces code that fits my architectural needs but doesn't drift and it took a lot of schema and guardrails to make that happen. It still churns out terrible code if you let it, so thats why I prefer managed languages like C# for the backend. .NET is pretty much guardrail abstraction anyway, so it doesn't feel then need to become too creative with the language. You're mostly nudging it back into shape when it drifts not pounding it back in.
- AI Bot network. It can work but you have to have your foundation and build trust over time. Grok Bot was great at the orchestration / marshalling but now its usage is weak sauce. Sneeze it in one day you can get around 30% usage and thats for the same work that I make Sonnet do and it taps at around 4-8%.. so i'd prefer using OpenBot and just use my existing subs. Tbh i'm cancelling Grok, its a waste of time and money atm.
- Markdown files are foundation built on sand. I keep seeing people push their markdown ecosystem every inch of the way as guardrails. It can work, even better tell AI "Use compressed notation" and it kind of builds a rules language to keep that in check. Honestly the best unlock for me is to live and die by declared schemas in json. Nobody gets to write a markdown without an attached schema that defines the shape.
Wild how most of these are "help my team actually use AI" and not "build the AI". Are they paying more for that than regular dev work?
1
1
10
random thought.. am re-reading The Magician again (Raymond E Feist) ... at various points in the book I stop and think "I wonder what that would have looked like" so I use voice to narrate to Codex to "render me this scene"
Here's pug leaving the portal... its not really him though ... but ....if you build out a character sheet and read a book.. i mean... something there...
1
142
Replying to @Dorizzdt @PirateSoftware
Wtf bro, how are you eating them to have that even happen in the first place
1
1
67
Followup if I may
1) I accept this
2) I need more info on how to improve. If this means eating boxes to train. I am willing to donate myself for science
3) I am eating them now and the bbq particles are in my keyboard. Is it rock bottom if i suck them out between the keys like a drink through a straw?
1
16
ππ
Microsoft filed a patent to let PC gamers earn money by sharing their unused GPU power with AI companies.
Users could earn credits for contributing their GPU power, potentially exchangeable for money, software or access to AI models.
The system would monitor GPU temperature, power consumption and memory usage, pausing workloads whenever the PC is needed for gaming.
1
7
1,113
If you've ever done a contract or what i call a "tour of duty" in a govt department federal and state.. you will soon realise putting these folks anywhere near AI controls is like feeding Colin Robinson from What We Do in the Shadows
youtube.com/watch?v=6r80I1YDβ¦
Australia's Prime Minister Anthony Albanese just announced new AI controls.
They want to review models.. there's no way model providers allow that.
The 6 items are:
1. Maintain security control
2. Manage risk
3. Manadatory disclosure
4. Review of Models
5. Risk management and transparency
6. Updated rules as technology changes.
2
197
Code review is now the bottleneck. I'm experimenting with diff visualization tools to help me judge PRs faster and ship quality software.























