Nick Dobos

prompt engineer @browsercompany | I make apps. My code runs on 100mil+ phones | former iOSdev @Twitter | indie dev @mindGoblinStdio

[Nick Dobos @NickADobos]

Buckle up. The world is about to get very weird.

[Sawyer Merritt @SawyerMerritt]

Tesla says they have officially started Optimus production in Fremont on their high-volume lines.

"The bots coming off this line are the first production quality bots that we have manufactured. While we will continue to iterate on the hardware and manufacturing process, we have proven that the current manufacturing process for Optimus is suitable for a commercial product."

[Nick Dobos @NickADobos]

~ 70% internet runs php * 46% old PHP
= ~32%

[Nick Dobos @NickADobos]

1/3rd of the entire internet has a massive security vulnerability and is so old that it’s not getting updated LOL

[International Cyber Digest @IntCyberDigest]

The day the internet broke, and most of it is too old to fix.

[Nick Dobos @NickADobos]

Making cancer drugs in space because in microgravity the chemistry of crystals forming is more uniform which leads to better performing drugs.

That’s sick as hell. The future is awesome.

[Wajahat Mughal @Wajahat]

Let me explain why is @VardaSpace is such an important company

A patient who has cancer has cells that have proteins that are activated leading to growth, replication and then damage.

Targeted drugs work on these cells in 2 ways:
1. Inhibit the receptor (the protein), essentially jamming its function until the drug wears off. An example is Tamoxifen, used in breast cancer to block oestrogen receptor binding.
2. Degrading the receptor altogether so signalling doesn't work and the cell needs to spend time recreating the receptor before it can work again, an example being Fulvestrant which degrades the receptor itself.

The second option here is important to understand because the effect of it significantly out lasts how long the drug lasts in your system. One type of these drugs is called PROTACs - it's a newer type of cancer drug that destroys these proteins (proteolysis) removing functions often related to cancer signalling.

Now these drugs are complex, 15+ chemical steps and can be upto 4x more heavy than normal drugs. With each molecule taking many shapes, when packed together, they often don't form a uniform lattice shape. On Earth, gravity leads to convection currents that feeds these molecules unevenly into the forming crystal, too fast resulting in non-uniform crystal formation.

These drugs:
- Dissolve poorly when ingested
- Have difficulties crossing cell membranes
- Lead to unpredictable bioavailability
- Require high doses in order to get enough drug in the system

Varda here by @WillBruey and @zebulgar aims to solves this with undisturbed crystal formation growth in microgravity.

With a new purer drug you get:
- Predictable bioavailability
- A more stable drug
- A better outcome for patients

This is Varda

[Nick Dobos @NickADobos]

Game changer for chat tab hoarders.

I think this effectively 2-4x’es my usage???

Huge for scheduled tasks too, because now you can let big threads of context pile up and get retriggered after multi hour or day delays. So now your weekly planning thread with 700k preloaded context tokens is super cheap cached price.

[Tibo @thsottiaux]

Day 16/

You can now cache for 12h in the API and for logged in users across Codex & Work we keep cache forever and you don’t have any penalty for coming back to an old conversation or resuming after a few hours and are now guaranteed to hit the cache.

[Nick Dobos @NickADobos]

Oh fascinating. League of legends is being reverse engineer distilled Vibecode attacked to the point they have had to change how game replays work. Model labs aren’t the only ones who need anti distilled patterns now.

[Spideraxe @Spideraxe30]

Update to Replay Limits. Rather than mostly being restricted Replays will be capped at a download limit of 10 per hour.

[Nick Dobos @NickADobos]

If Claude was serious about not killing the world it would auto patch the entire internet

[Marc Köhlbrugge @marckohlbrugge]

if this PHP vulnerability allows remote code execution, doesn't that mean we can use the vulnerability to auto-patch all vulnerable servers?

[Nick Dobos @NickADobos]

@rileybrown Yeah it’s kinda wild the gap.

Not sure of this is simply the training data from customer feedback, or simply a compute and tech lead, but it’s very apparent.

[Nick Dobos @NickADobos]

Incredible public good. I’m so excited to see how health evolves for the world

[Ilya Abyzov @IlyaAbyzov]

ChatGPT is now the largest medical record sync from hospitals to consumers, passing Apple.

I'm glad we can all access our medical records through the apps we choose. Rightly so!

We're working with Epic and hospitals to reduce the API load on their servers. The current setup is archaic - the equivalent of refreshing your inbox instead of getting a push when email arrives.

FHIR Subscriptions can solve this & reduce strain on the system.

[Nick Dobos @NickADobos]

@dennis_kortsch Better idea. Build a code maid to clean up automatically. That’s genuinely not something you should be wasting thought cycles on.

[Nick Dobos @NickADobos]

What if they train Claude 6.0 Fable in jail?

Probably haunted though

[Will Manidis @WillManidis]

strong america indicator that defaulting private prison operators are now marketing their sites as prospective data center locations

[Nick Dobos @NickADobos]

Oh that’s why Astra ultra fast drained it in 30 minutes lol.

[Nick Dobos @NickADobos]

Good point, but I’m not sure it holds for software and media in general.

Owning the customer relationship & how you make people feel and trust you is the one and true final most.

For example Zuck and Elon could make the best Ai but lots of people won’t use either of them simply becuase of their values & other companies.

Though I’m realizing as I’m typing this maybe that’s not the app layer lol.

[Nick Dobos @NickADobos]

App layer now wins.

[The Information @theinformation]

OpenAI, Anthropic and other leading AI labs face a growing challenge: Better models aren't necessarily attracting more usage. A 10-point intelligence advantage now earns just 20% more developer traffic, down from nearly 69% in late 2025.

Full story: https://t.co/V3pI2VYOLb

[Nick Dobos @NickADobos]

@gabrielchua @OpenAIDevs Say more?

[Nick Dobos @NickADobos]

@rileybrown How do they make money on this though?

Big models cost a lot

[Nick Dobos @NickADobos]

I’m just gonna point out that if you are stupid enough to think spam reply bots will work, you are therefore really bad at writing prompts

1. because of the simple fact that goal setting and aiming an AI agent at the most valuable task is probably the #1 skill

2. Becuase they wouldn’t have this problem of needing a bot to write for them if they actually knew how to present themselves and their ideas as interesting. If they are able to get attention in the first place, then they wouldn’t do this. So no surprise the Ai extensions of them act similar, and are uninteresting too.

[Nick Dobos @NickADobos]

Peculiar situation where every company wants to vertically integrate Ai software factory stacks, but also every dev tool and model lab is competing for it

[staysaasy @staysaasy]

Every single Eng leader out there is iterating on some kind of PR autoapprover, coding automations, AI cost optimization, and model switching/eval stuff.

Just a million people doing the same thing.

[Nick Dobos @NickADobos]

@nopranablem A light one usually takes me around 30-40

But it adds up as you go into long stretches.

[Nick Dobos @NickADobos]

People wanna hate on this but I do think 2-3 days, staying up late, of extended focused pour your soul and mind into the computer and do almost nothing else is the ideal coding flow now.

Unfortunately most jobs (and most people) are not setup to handle this or practiced with this intense on/off cycle, so it’s sounds and looks crazy. Not for everyone.

But like the standard tech co office hours of 9-5 or more, 5 days a week with crazy amounts of context switching and meetings is an insane working style too and produces many undesirable psychological effects. Neither of these approaches are without health concerns nor downsides.

Worst part of my job by far is the way hours are imposed as a regular low grade chronic stress grind and not these intense sessions.

I would prefer a more manic and obsessive work style personally.

It genuinely takes hours to get into a full proper vibecoding flow state. Similar to how meditating 20 minutes a day doesn’t really kick in until day 10 or so. Takes time for neurons to change and your body to get there. By hour 12 of vibecoding you reach areas of thought and creativity you can’t reach otherwise.

Alas the plight of the creative and the obsessed. People see out of distribution individuals and yet don’t seem to wonder if their quirks and flaws are what lead to their successes in other areas in the first place.

[Nick Dobos @NickADobos]

Is anyone else thinking we are going to get a bunch of smart system 1 models for slightly different use cases now?

[Nick Dobos @NickADobos]

New Jev like model from Cloudflare!

Decision models are out already lol.
We got system 1 models now

Cool thing about this, unlike Jev which can only return decision probabilities, this can also extract text, computed and estimated values. Because it's not generating new text tokens, it can't hallucinate, but it still lets you pull exact words and figures out into structured json.

Super handy for asking questions like how many, or how often. A new type of smart if statement similar to Jev but for different questions.

[michelle @michellechen]

a brand new model paradigm! introducing 𝚌𝚕𝚊𝚒𝚛𝚒𝚌: a system one model for extracting, computing, estimating, and predicting values from unstructured inputs.
https://t.co/xRE3gO6JeO