NO WAY THIS COST JUST $1.52
SOMEONE JUST GAVE DEEPSEEK ONE PROMPT AND IT BUILT A FULL 3D SATELLITE TRACKING SYSTEM.
Not a basic demo. An actual browser-based satellite tracker called SATELI.
→ Tracks 16,846 active satellites in real time
→ 11,157 Starlink satellites included
→ Live 3D Earth with city lights and day/night simulation
→ Predicts when the ISS and other satellites will be visible from your location
→ Includes sky trajectory charts and night vision mode
BUT THE CRAZIEST PART IS THE ENGINEERING.
DeepSeek implemented SGP4 from scratch and passed 666 official verification tests across 33 satellites.
The result matched reference calculations to within 8 micrometers.
And the entire session cost just $1.52.
ONE PROMPT. ONE TRY. $1.52.
Built using DeepSeek V4.1 Flash with AgenC Core.
AI coding is getting out of hand.
ARE WE ENTERING AN ERA WHERE YOU JUST DESCRIBE WHAT YOU WANT AND AI BUILDS IT? 👀
SOMEONE JUST BUILT THIS ENTIRE GAME WITH AI. 🤯
Someone just spent 13 weeks building an ocean exploration and fishing game using AI on a $200/month plan.
29 ships in bottles.
120 fish planned.
5 zones.
Fishing, exploration, boat upgrades, pirates, storms and fog.
And the crazy part? Almost the entire game was built with Claude Opus.
The total cost so far is around $900–950.
AI game development is getting INSANE.
Are we entering an era where one person can build an entire game studio alone? 👀
I'm nitpicking but it seems every AI company has forgotten what "MAX" actually means.
If I'm paying for MAX, I expect MAX. Not "MAX until you actually need it."
We’re renaming our Max plans:
Max 5x → Max
Max 20x → Ultra
Max is still $100/month, and Ultra is still $200. If you’re a subscriber, there’s nothing you need to do.
We’re also removing the 5-hour limit for verified Ultra accounts, starting today with most existing subscribers. Newer accounts keep a temporary limit until they’re verified, which helps us prevent fraud and abuse. You'll get an email when it's removed.
Weekly limits remain unchanged. View our plans here: claude.com/pricing
THIS IS F**KING INSANE.
Someone's Android app had been broken and abandoned for 4 YEARS.
They asked GrokBot for help.
10 minutes later, they had a working APK.
12 HOURS later, they'd rebuilt the app and added around 50 new features.
No spending hours in Android Studio. No hiring an expensive freelancer.
Just describing what they wanted and letting AI do the heavy lifting.
People are still sleeping on how much AI coding is changing software development.
We're entering a whole new era.
DeepSeek V4 Flash just built a WALKABLE A320 cabin in 51 minutes for $0.80.
Not a static 3D model. An actual airplane cabin you can explore.
You can walk from the cockpit through business class, economy, the galley, and even the lavatory.
WASD movement.
Collision detection. Mouse look. Free-fly mode. Day/night lighting. Openable overhead bins and window shades.
But the crazy part isn't just what it built.
I let the agent run without approvals.
51 minutes.
166 model calls.
98% of input came from prompt cache.
$0.80 total.
And it tested its own build in a real browser before calling it done.
No console errors. ~132 FPS at 1440x900.
Built with AgenC Core, an open-source agent harness using DeepSeek's own API.
GitHub: github.com/tetsuo-ai/agenc-c…
I'm sharing more AI experiments, coding agent builds, and real-world results like this.
Follow me for more AI builds, experiments, and projects worth paying attention to. 👀
We cant even get 3 prompts with argon
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model.
Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys Intelligence Index at 59. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
We’re seeing significant improvements in long-horizon tasks even compared to Argon, and we’ve strengthened frontier safeguards.
As we did with Argon, we’re going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
Where is all the money for AI actually coming from?
The scale of AI spending right now is enormous.
Data centers, chips, talent, power. Companies and investors are pouring in capital at a rate that feels almost disconnected from traditional business cycles.
But where is the money actually coming from?
Is it mostly Big Tech recycling profits? Sovereign wealth funds? Private equity chasing growth? Or cheap capital looking for the next big story?
What makes me pause is how little we talk about the actual sources of funding.
If a large chunk of this spending depends on a handful of hyperscalers keeping the money flowing, how sustainable is the whole AI boom?
Is anyone actually keeping their $200/month ChatGPT Pro subscription after the Nov 1 price hike and usage cuts?
As a longtime OpenAI user, this feels like the perfect way to push people toward Claude, Grok, and Muse.
I’ll use up my remaining credits, then cancel and switch.
Anyone else thinking the same?
I asked DeepSeek V4.1 Flash and OpenAI’s new GPT-6.1 Terra at Ultra effort to build a 3D car in Three.js. Here’s what each made.
Same prompt. Same task. Different coding setups.
I compared GPT-6.1 Terra using Codex CLI 0.164.0 in Ultra mode against DeepSeek V4.1 Flash at max effort using AgenC Core 0.19.19.
The challenge: build a realistic, interactive European family MPV/crossover in Three.js and WebGL, featuring:
- Detailed exterior and interior
- Opening doors and trunk
- Working steering
- Functional lights
- Interactive orbit camera
No prebuilt complete car models.
Terra used automatic task delegation and created three subagents for the body, wheels, and interior.
Both completed the task autonomously. Their generated applications are unmodified, and I checked the interactions in the browser.
The video shows both cars using their native automatic rotation controls. The interfaces were hidden for presentation, and the camera framing was adjusted. Geometry, materials, and lighting remain unchanged.
This compares the complete coding setups, including their harness behavior, rather than treating it as a controlled single-agent model comparison.
Which one looks better to you, and why?
DeepSeek V4.1 Flash was tested using AgenC Core.
github.com/tetsuo-ai/agenc-c…
Follow me for more AI model comparisons, real-world coding experiments, and hands-on tests.
Follow now so you don't miss the next experiment. More AI builds and comparisons coming.
Something I've noticed about AI coding: generating code is exciting, but coordinating everything around it can get exhausting.
That's why SuperPlane software factory approach caught my attention.
Pick issues. Run agents. Verify the work. Review the PR.
Fewer manual steps between an idea and shipped code.
🤝 Paid partnership
Last month, 52% of the pull requests our team merged were automated through SuperPlane.
Today, we’re launching @superplanehq 1.0: an open source software factory for engineering teams.
We built it because using a coding agent can feel like becoming its assistant.
Starting sessions. Checking its work. Chasing failing CI.
SuperPlane takes on that coordination:
- Scans your backlog and proactively picks up tickets suited to automation.
- Runs a coding agent and verifies the result.
Works through CI failures and review feedback.
- Your team reviews the evidence and decides what ships.
The setup, instructions, and standards are shared across the team. When someone improves the process, everyone benefits.
Choose your agents and models. Run it on your own infrastructure.
We spent more than a decade building CI/CD at Semaphore. We’ve seen how much teams can accomplish when routine work runs reliably in the background.
Now we’re bringing that approach to coding agents.
Here’s how it works:
Some tips to get the most out of Grok Bot 🧵
I burned through my $200/month plan in just 3 days because I had no idea what I was doing.
After some trial and error, here's what I've learned:
• Upgrade to the unified plan. I feel like I get much more usage now.
• Don't code with Cursor. I connect Grok to my ChatGPT and Claude plans via CLI and let them handle coding when needed. Better value for money, especially ChatGPT.
• Audit your routines. My meeting notes workflow was checking every 30 minutes without me realizing. Now it only runs after my actual meetings.
• Use fewer bots. Fewer handoffs and less communication mean less token usage.
• Reduce browser work. Browser automation is expensive, so use APIs whenever possible.
• Give feedback early. Ask for mockups before building the whole thing. This saves time and wasted tokens.
• Batch tasks. Explain the desired outcome upfront and let Grok ask questions before starting. Less back-and-forth, less wasted usage.
Still a work in progress, but going from burning through 100% usage in 3 days to something much more manageable feels great.
Shams retweeted
GRVT has made its first buyback official, and the next phase is already taking shape.
@grvt_io bought 1.1M $GRVT on the open market and permanently burned the tokens, nearly 1% of the circulating supply. The public dashboard lets anyone verify the buyback details and transaction hashes.
I am looking forward to seeing Grvt Earn, Stable Funding Perps, and yield-bearing margin come together in GRVT 2.0, giving users a way to earn and trade from one self-custodial balance.
Buybacks are intended monthly, but discretionary. So the next few rounds will say more than this first one 👀
Shams retweeted
Hi @Support @PremiumBusiness - we kindly ask the X team to review and restore our account @wallchain.
We’ve already submitted an appeal and are fully committed to complying with X’s policies.
p.s. any amplification for visibility would be appreciated