Alex Smagin (@asmagin) header
Alex Smagin (@asmagin) profile picture

Playing with technology for the better half of my life.

NJ, USA
Joined May 2009
Photo by Alex Smagin (@asmagin)Photo by Alex Smagin (@asmagin)Photo by Alex Smagin (@asmagin)Photo by Alex Smagin (@asmagin)Photo by Alex Smagin (@asmagin)Photo by Alex Smagin (@asmagin)Photo by Alex Smagin (@asmagin)Photo by Alex Smagin (@asmagin)Photo by Alex Smagin (@asmagin)Photo by Alex Smagin (@asmagin)
Alex Smagin retweeted
If you are using Grok Bot, don't have it do coding work directly. That is going to burn your tokens, and leave you upset about usage limits. Instead, pair your Grok Bot Ultra sub with an ultra level coding sub from one of the frontier labs. Both Codex and Claude Code have CLIs available. Grok Bot can install these CLIs on its virtual computer, and can orchestrate their use to do your actual coding work. Have them execute that coding work in their cloud environments, and not on the virtual machine itself (it can overload the VM ...). This allows you to spread the token burn across two subscriptions, and allows for more specialization at the agent level. Grok Bot is your orchestration layer, Claude / Codex are your implementation workers. The results are killer!
135
27
12
784
65,626
Finally jumped into @grok @bot over the weekend. Wired up a @telegram connector with media, buttons, and voice. Woke up Monday to find I’d already burned >50% of my weekly limits gone. Haven’t done anything useful yet 🙈. github.com/asmagin/grokbot-t…
Replying to @DogecoinNorway
The cost of serving Grok Bot was unsustainably high relative to revenue
1
54
20 years ago I heard the phrase: “You can’t automate chaos.” It still holds up perfectly against today’s AI hype. Organizations need a solid digital and data foundation first. AI doesn’t fix the mess — it just amplifies whatever is already there. More: asmagin.com/posts/2026/08/th…
1
46
Alex Smagin retweeted
183
820
62
9,270
337,965
Alex Smagin retweeted
There's a fruit fly walking around right now that was never born. @eonsys just released a video where they took a real fly's connectome — the wiring diagram of its brain — and simulated it. Dropped it into a virtual body. It started walking. Grooming. Feeding. Doing what flies do. Nobody taught it to walk. No training data, no gradient descent toward fly-like behavior. This is the opposite of how AI works. They rebuilt the mind from the inside, neuron by neuron, and behavior just... emerged. It's the first time a biological organism has been recreated not by modeling what it does, but by modeling what it is. A human brain is 6 OOM more neurons. That's a scaling problem, something we've gotten very good at solving. So what happens when we have a working copy of the human mind?
691
2,291
1,149
25,036
9,602,260
"program.md" sounds very similar to "program.cs" from my dotnet background. The more we trying to get better results from AI the closer we get to development best practices and project management approaches. We just writing now at the higher level of abstraction.
Karpathy just open-sourced autoresearch. One GPU. 100 ML experiments. Overnight. You never touch the code — just write a Markdown file. The bottleneck isn't compute. It's your program.md. gli.st/z3iakp3f
1
73
Alex Smagin retweeted
someone connected LIVING BRAIN CELLS to an LLM Cortical Labs grew 200,000 human neurons in a lab and kept them alive on a silicon chip, they taught the neurons to play Pong, then DOOM now someone wired them into a LLM... real brain cells firing electrical impulses to choose every token the AI generates you can see which channels were stimulated, the feedback from the neurons in choosing that letter or word
729
1,496
949
11,401
2,970,988
Alex Smagin retweeted
17,000 tokens per second!! Read that again! LLM is hard-wired directly into silicon. no HBM, no liquid cooling, just raw specialized hardware. 10x faster and 20x cheaper than a B200. the "waiting for the LLM to think" era is dead. Code generates at the speed of human thought. Transition from brute-force GPU clusters to actual AI appliances. taalas.com/the-path-to-ubiqu…
376
912
460
7,258
1,817,317
🚨 Our #1 most requested feature is here: You can now export designs from any Stitch agent directly to Figma as editable layers. Vibe Design is perfect for exploring many ideas in minutes. But sometimes, you need that final layer of polish. Now you can move seamlessly between vibe design and polish.
106
273
79
2,979
261,977
Because you wouldn’t let it slide… these are rolling out today for our most requested feature: Prompt-Based Revisions: Tweak, tailor, and tune your slides just by prompting the revisions you want PPTX Support: You can now export your Slide Decks (Google Slides coming next!)
437
1,555
1,046
11,729
3,679,827
Alex Smagin retweeted
welp, it was fun while it lasted folks
176
1,007
118
19,307
825,544
Alex Smagin retweeted
Google just killed the document extraction industry. LangExtract: Open-source. Free. Better than $50K enterprise tools. What it does: → Extracts structured data from unstructured text → Maps EVERY entity to its exact source location → Handles 100+ page documents with high recall → Generates interactive HTML for verification → Works with Gemini, Ollama, local models What it replaces: → Regex pattern matching → Custom NER pipelines → Expensive extraction APIs → Manual data entry Define your task with a few examples. Point it at any document. Get structured, verifiable results. No fine-tuning. No complex setup. Clinical notes, legal docs, financial reports, same library. This is what open-source from Google looks like.
156
764
51
8,029
734,654
Alex Smagin retweeted
Disaster is coming. Thousands of ClawdBots are live right now on VPSs… with open ports to the internet… and zero authentication. This is going to get ugly. If your agent can: - browse the web - call tools - access files/secrets - hit internal endpoints …then an unauthenticated public endpoint is basically “please take over my bot”. This isn’t theoretical. The internet is a nonstop scanner. Fix it today: 1) close the port / firewall to VPN or IP allowlist 2) add auth (JWT/OAuth, at least a strong secret) + TLS 3) rotate keys (assume compromise) 4) rate limit + logs + alerts Agents are powerful. Demo-grade deployments on the open internet are not.
Readers added context they thought people might want to know
The image presents global data from the Terrorist Incidents Map of the World (1970–2015). It is not related to any specific events in 2026. factcheck.afp.com/no-approximate… commons.wikimedia.org/wiki/File:Terr…
317
524
218
4,554
809,326
Alex Smagin retweeted
DeepMind just did the unthinkable. They built an AI that doesn't need RAG and it has perfect memory of everything it's ever read. It's called Recursive Language Models, and it might mark the death of traditional context windows forever. Here's how it works (and why it matters way more than it sounds) ↓
294
1,074
164
7,690
957,104
Men, This is why understanding Debt is so Important ‼️‼️
35
1,090
56
4,364
342,474
Interesting turn of events, they use Google to kill Google
Google just made a subtle but massive change Last month, Google quietly removed the num=100 search parameter. This means you can no longer view 100 results at once. The default max is now 10. Why does this matter? - Most LLMs (OpenAI, Perplexity, etc.) rely (directly or indirectly) on Google’s indexed results, alongside their own crawlers. - Overnight, their access to the “long tail” of the internet was cut by 90%. The fallout: - According to Search Engine Land, 88% of sites saw a drop in impressions. - Reddit, which often ranks in positions 11–100, saw its LLM citations plummet. Its stock dropped 15%. For startups, this is brutal. Visibility just got harder. Reddit as part of AEO just changed entirely. It’s no longer enough to build a great product you need to crack distribution first. Because if people can’t discover you, they’ll never get to evaluate you. Most engineers seem to always neglect this reality, but a mediocre product with great distribution will always beat a great product with mediocre distribution. As Peter Thiel says: “Most businesses get zero distribution channels to work: poor sales rather than bad product is the most common cause of failure. If you can get just one distribution channel to work, you have a great business. If you try for several but don’t nail one, you’re finished. Superior sales and distribution by itself can create a monopoly, even with no product differentiation. The converse is not true. No matter how strong your product — even if it easily fits into already established habits and anybody who tries it likes it immediately — you must still support it with a strong distribution plan." Distribution > Product (h/t Adarsh Appaiah on LinkedIn)
1
191
Compiled prompt 👇🏼 Generate a professional, high-resolution profile photo maintaining the exact facial structure, identity, and key features of the person in the input image. Frame the subject from the chest up with ample headroom and negative space above their head, ensuring the top of their head is not cropped. Have the person look directly at the camera with their body also facing the camera, conveying confidence and approachability. Dress the subject in a contemporary smart casual blazer with crisp focus on fabric texture. Use a solid '#141414' neutral studio backdrop to avoid distractions. Apply bright and airy soft, diffused studio lighting, gently illuminating the face and creating a subtle catchlight in the eyes. Capture the image with an 85mm f/1.8 lens, featuring a shallow depth of field, exquisite focus on the eyes, and beautiful, soft bokeh. Ensure crisp detail on individual strands of hair and natural, realistic skin texture. Use clean and bright cinematic color grading with subtle warmth and balanced tones for a polished, contemporary feel.
Your next professional headshot is a prompt away. Generate polished, business-ready headshots and creative portraits with Nano Banana—no expensive equipment or complex editing software needed. Here are a few prompting tips to consider 🧵

2
222
None of it worked, but boy was it beautiful. 🤌🏼🤣
Claude 4.5 Sonnet just refactored my entire codebase in one call. 25 tool invocations. 3,000+ new lines. 12 brand new files. It modularized everything. Broke up monoliths. Cleaned up spaghetti. None of it worked. But boy was it beautiful.
65
It's super fast and cheap, but for some reason it gives so many errors in @cline, retrying the same thing for many times.
Grok Code leads with 57.6% of coding traffic on OpenRouter, more than every other AI code generator combined Grok Code Fast 1 ranks #1 Grok 4 Fast is at #4
1
74
Alex Smagin retweeted
I am once again begging you to put your database servers and application servers in the same region.
185
557
110
8,146
1,422,834