Used to write code, now I dictate it. Sharing what I learn about aligning my ideas & code with agents.
undefined
Joined February 2009
- Tweets6.7K
- Following702
- Followers727
- Likes994
GPT always had autism because of their reasoning caveman speak.
Claude nerfed their legibility when they added the fingerprint.
Muse is trained on facebook posts and whatsapp messages.
I am not surprised it can write.
TIL in @t3dotcodes you can cmd+s to temporarily stash some text.
It's still the little things that differentiate vibe coding from agentic engineering.
OpenAI DevDay summary:
- 6.1 Still has autism, doesn't understand intent, but is smart, and cheap.
- Added partners to get more credibility
- Copied Jev
- Copied Grok Bot
- Rug pull because dev market is saturated, reallocate to provide tmp free dots usage
Am I reading this right?
"""Hence, the 500$ plan, for you to ve able to use the new AI model we announce tomorrow.
Nothing changes for you when you use previous gen models, but this new one, well, its expensive and you'll pay for it."""
Hi,
Tomorrow we are re-opening the Pro $200 subscriptions to new subscribers, but together with it we are also changing how we calculate the usage for it. In effect, if you do the math, it will net out at half the dollar in API spend compared to the old Pro $200 plan.
Now that it's said, let me explain why this is happening and why you will still get more work done than if you were on the Pro $200 subscription one month ago.
(a) We didn't want to compromise in other ways and are committing to not reintroducing the 5h limit, so that you can fully use the weekly usage when you want.
(b) On the subscription, we guarantee that over time you always get more work done and with an increasing level of quality. This means that you will continue to get more value per dollar spent as a result of models getting more efficient and us passing down the improvements in the form of API price reductions.
(c) We don't want to put an incentive on ourselves to artificially inflate the API list prices to make it look like you are getting a lot (and workaround it through discounts, etc). Instead we want to continue to both rapidly reduce prices and increase capabilities of models on the API. This week we introduced GPT-6 Sol and GPT-6 Luna at 50% of their previous price. Over time, we see prices go low enough that it makes sense for most to buy usage as needed without there being a significant gap between what you get in a subscription and what you get in the API for a dollar spent.
(d) Tomorrow, we are adding more things to the subscription that won't draw on the usage, I won't reveal what that is yet.
I wanted to be transparent before all the big announcements tomorrow. Lots of new exciting things are coming to the subscriptions that will make it super compelling, but I wanted to make sure to share this change ahead of time so you can all understand it before we shower you with good news.
Codexingly,
Tibo
I just reviewed this PR from Opus 5.5 (some of us still read code)
This is a Svelte.js codebase where I typed 80%+ code just a little over a year ago.
Opus cleaned up pain points that would have taken me DAYS.
Total review comments: 3, all easy minor nitpicks...
I'm floored.
Case in point.
Asked GPT 6 Astra if I can disable the f2 warning, along with a screenshot.
And I get an immediate "I misidentified" response.
This isn't a fluke.
This isn't a one-off.
GPT 6 is simply not ready.
I don't make decisions like this lightly.
I ran the same single task with a carefully constructed long form prompt that described a single feature
Every GPT model produced:
- Overengineered slop
- Not what I asked
Opus High got it right first shot with readable code 4x faster
🔥 nor retweeted
> if you're starting a business, code is the least important part.
Well yeah. Try to finish the product then.
Code quality is an artifact in the context for an agent. If you give it a shitty context it'll do shitty things. techdebt piles up fast and the agent is always happy to work around existing code, not modify it.
"Use a linter, bro" - not working. It'll rip out your cyclomatic complexity into small functions and pile them up in a single file. It's painful to watch.
You're coders, man. Get your shit together and steer the agent based on code changes you approve.
1. you'll understand your software better and know what change you need next (no weird spec bingo)
2. don't nit on every little code detail, roughly go through the code, let AI explain it to you for troubling parts
3. it's fast development - avoiding tech debt is smooth sailing with only one condition: read what the agent produces
Don't sell me this "today you don't need to look at your code". If you want to rely on frontier models and want to be lazy then be my guest.
Mediocre-complex software is not vibecode'able.
Easy slop is.
i'm appalled by how solo builders are approaching agentic engineering today. i've spoken to several people over the last couple months 1:1 to get an idea for how they're building with AI.
every single one of them is vibecoding like it's january 2026. they're writing huge specs. they have a kanban board neatly organized with hundreds of todos. they're having their agents take detailed notes after every change. i'm like, what are you doing??
why are you giving it specs? why do you have a todo list??
if you're starting a business, code is the least important part. half of these people are worried about AI code quality. i'm like, what are you talking about?? why the fuck do you care about code quality??
I think I am going to get a divorce from codex.
We had a good run, but - Astra is just a slopmachine, and GPT-6 Sol max is lower than 5.6 Sol high...
Is the slowdown here?
