This is insane...
Someone built a physical switcher for all their Claude Code agents.
A 4" touchscreen with a Raspberry Pi glued to the back. Every running session shows up as a little pixel critter in a 4x4 grid.
Hit “focus terminal” and the monitor jumps straight to that session. Hit “interrupt” and it stops.
16 agents running and you manage them like apps on a phone.
Check out the article below for more tips on Claude:
Adolfo Bolivar retweeted
Sometime in late summer 2026, I believe AI crossed a critical threshold: AI escape velocity towards superintelligence (SI), which is now within our reach.
The pace of progress especially at the three leading AI labs, namely OpenAI, Anthropic, and Google began accelerating at a rate that feels almost double exponential. Each new generation of models is arriving faster, with capabilities that would have seemed extraordinary just months earlier.
But I believe something much more profound is happening beneath the surface. We are approaching the point where intelligence itself becomes the engine of its own acceleration. AI is increasingly capable of writing sophisticated software, solving difficult scientific problems, designing experiments, discovering new algorithms, and assisting in the development of more powerful AI systems. As these capabilities improve, the feedback loop between intelligence and technological progress will become extraordinarily powerful beyond our wildest imagination.
And now, looking toward 2027, I believe we are entering perhaps the most consequential year in human history.
Imagine AI systems capable of conducting months of scientific research in days, teams of autonomous AI scientists exploring thousands of hypotheses simultaneously, and virtual laboratories where millions of experiments can be simulated before the most promising are tested in the physical world. Imagine the rapid emergence of increasingly capable robotic systems that connect machine intelligence directly to physical discovery and production.
In biology and medicine, fields especially close to my heart, the implications are staggering. We could begin moving from observing biological systems to predicting and eventually engineering them with unprecedented precision. Virtual cells, digital twins, AI designed therapeutics, and autonomous laboratories could transform how we understand disease, develop treatments, and ultimately confront aging itself.
The greatest transformation, however, may be even more fundamental. For all of human history, scientific and technological progress has been constrained by the limited number of human minds capable of pushing the boundaries of knowledge, and by the time those minds require to learn, reason, and discover. For the first time, we are developing intelligence that can potentially be scaled far beyond those biological limits.
Consider what this means. Scientific progress has traditionally produced better tools. We are now creating tools that can contribute directly to scientific progress and perhaps eventually improve the very process by which new knowledge and new intelligence are created. If that feedback loop becomes sufficiently powerful, the acceleration could surpass anything humanity has previously experienced.
My prediction is that 2027 will be remembered as a turning point, when AI begins evolving from an extraordinarily powerful assistant into an increasingly autonomous engine of scientific discovery, engineering, and innovation.
We are approaching a future in which breakthroughs that once required generations become achievable within years, and eventually perhaps months or weeks. Problems that appeared permanently beyond our reach may begin yielding to forms of intelligence capable of exploring possibilities no human mind could ever investigate alone.
I have never been more excited about the future. I believe we are witnessing the opening chapters of the greatest transformation in the history of our civilization.
2027 may be the year we begin to understand that the limits we have always accepted were never the limits of what is possible. They were simply the limits of human intelligence.
And those limits are about to change forever!
Adolfo Bolivar retweeted
Claude writes better code when it can verify its work. Checks that used to be expensive, like tests, benchmarks, and end-to-end runs, are now cheap.
Here’s how the Claude Code team thinks about verification:
claude.dev/blog/how-the-clau…
This is f*cking insane. Matt Palmer and Lauren Tan, the team behind Grok Bot, sat down for a full Q&A:
"Grok Bot is not a chatbot
It's an AI teammate that joins your meetings and works with your apps"
In 1 hour, they show how many bots Lauren actually runs, how to build a Chief of Staff agent and the biggest mistake people make with Grok Bot
Meetings → Your Apps → Chief of Staff → Cloud Coding Agents → A Team of Bots
Most people still use it like a chatbot. They use it like a company
Worth more than any $2,000 agent course
Bookmark and watch it today
Then read how one person runs 6 Grok bots and 200+ agents ↓
Replying to @mattpocockuk
A little bit extreme. Comments do have a purpose, and I think they can be there to mitigate intent debt, both for agents and future human devs.
arxiv.org/abs/2603.22106
Software is both art and utility. AI is both an Industrial Revolution and a Renaissance.
Factories are for utilities, at AIE Code I talked about how we can make better art.
Vibe coding a game is fun until you don't know what your own characters do.
@trq212 on keeping creative control as models get better: readable specs, custom languages, and a software renaissance.
The Sweet Side of the Bitter Lesson.
Watch: youtu.be/xRJMiaUoD_8
Adolfo Bolivar retweeted
Anthropic just opened its Claude Forward Deployed Engineer learning path!
I've already completed most of it, and it gives you a clear picture of how to deploy AI in enterprise environments, from initial pilots to production, with the necessary security and compliance safeguards in place
Here's the link 👇
Adolfo Bolivar retweeted
I am sharing a new paper today with Dario Caldara: “Can Peer Review Survive AI-Generated Text?”
sas.upenn.edu/~jesusfv/Peer_…
Dario and I, as editors, are increasingly getting papers submitted to our journals that are 99 percent AI-generated. Even worse, we are also receiving many AI-generated referee reports, both as editors and, more frustratingly, as authors.
And we cannot emphasize this enough: the real problem is not that the papers and reports were AI-generated; it is that no one had done the thinking behind them, and you can so easily tell. The papers were eminently forgettable, and the referee reports were disappointingly uninsightful. We have seen too many reports of the form: “The paper is well-written and addresses an important problem. However, the authors should consider additional baselines.”
This deluge of AI-generated text puts peer review in jeopardy: journals will be flooded with low-quality submissions, economists will decline requests to referee papers or routinely use AI to referee them, and which papers get published will become less and less linked to their quality.
We do not want to defend the status quo entirely. Peer review is slow, many reports are poor, and editors (including us!) make bad calls all the time. Still, well-executed peer review remains the best system we have for assessing each other’s work and advancing the profession. AI could improve peer review if used deliberately for that purpose, for instance, to check proofs and code. That is not what we are seeing.
The difficulty is that readers, including editors and referees, cannot see how AI was used. As a result, in practice, the burden of verification shifts from the author to the reader.
Since we reject this shift, Dario and I did what economists always do when something bothers us: we wrote a paper.
So, what do we do in it? First, we check whether AI detectors such as Pangram get this right: is a referee report flagged as AI-generated indeed AI-generated? Or is Pangram just flagging proofreading and editing? No. We run an experiment with 25 referee reports from the pre-AI era. Proofreading and editing do not trigger AI-generated flags. Only when we ask the LLMs to substantially rewrite the referee report or to generate a new report do we get the AI-generated flags. Pretty compelling evidence, if I may say so.
Second, we build a theoretical model to explore who should bear the burden of verification: the author or the reader? (Yours truly always enjoyed the courses on the rules of evidence in law school; yes, as Billy Wilder would say, nobody is perfect.)
The answer is that it is much more efficient for the author to do it, because the author can verify at a much lower cost and only needs to do it once (rather than each reader repeatedly). Before AI, the authors more or less had to verify the results, even if only because they had to type them in. This was not perfect (and all papers ever published in economics probably have at least one mistake). But the problem is that, with AI, the author does not even need to read the text. Or can just skim it. Writing is thinking, and without writing there is usually not much thinking.
So, a paper flagged as AI-generated is prima facie evidence that the author has not verified it. Should this lead to an automatic rejection? No. But there should be a zero-tolerance policy for serious mistakes. The first serious mistake we find in the paper, and it is out! In comparison, for a non-flagged paper, a serious mistake is something that the editor might determine it can be fixed in a revision.
We are stricter with referee reports. There is absolutely no point in the referee generating a report using an LLM. As an editor, I can do that directly and save myself the trouble of asking you for it. Now, of course, you can use the LLM to help you with the report (if the journal allows it! Many journals do not allow you to upload the paper to an LLM, and you need to follow the rules). For example, you can use the LLM to search for existing results in the literature.
But peer review is not about finding one sentence that is not entirely precise or a slightly sloppy mathematical proof. Referees are asked: “Is this idea important despite its flaws?” And, relatedly: “Is the paper a good fit for the journal in terms of quality and focus?” That judgment is the task for which you are selected as a subject expert.
We also propose that journals require that, before a paper is submitted, the author submit it to an LLM using an approved prompt and include the LLM report and a discussion of what was and was not fixed. That way, the referee would not need to check (or at least not check so carefully) every technical detail, and they could focus on the big picture: Is this a good paper? Is this a good fit for this journal? What is fundamental for this paper to become publishable at this journal?
Anyway. The paper has many other arguments, and we try to be nuanced. So, before ferociously typing the first complaint that crosses your mind, would you consider reading it? Chances are that we have already discussed your objection. We have been thinking about this for a long time.
But by all means, leave good comments!
Adolfo Bolivar retweeted
Una sola persona acaba de jubilar a los bootcamps de programación.
Hizo una web que enseña con animaciones cómo funcionan por dentro las apps con millones de usuarios. Un ejemplo: una estrella del pop con 50 millones de seguidores publica algo. Su publicación tarda más de 8 minutos en llegar a todos y, mientras tanto, las publicaciones de los demás se quedan esperando. La lección te enseña cómo bajar esa espera a unos 2 segundos.
40 lecciones así, animadas. Gratis, sin cuenta y en español.
Lo que antes era un bootcamp de miles de dólares ahora parece una pestaña del navegador.
Adolfo Bolivar retweeted
MIT has an entire course on the stuff you should know at your first engineering job
shell, git, debugging, profiling, shipping code
the 2026 version even has agentic coding. all free.
link: missing.csail.mit.edu/
Adolfo Bolivar retweeted
Awesome Robotics! 💾
Building your own robot just got a lot easier.
@MathWorks have released an open-source GitHub repository packed with robotics resources for anyone interested in getting hands-on.
The repo includes examples for robot arms, ground vehicles, and drones, with projects that show how to connect with ROS and ROS2 or even deploy Simulink models directly as ROS nodes.
There are also more advanced demos, like modeling off-road environments and testing navigation algorithms in photorealistic simulations.
Everything is well-documented, with tutorials and links that make it easy to go from concept to prototype. 📑
Whether you’re a student, researcher, or just curious, there’s material here for every level. And since it’s an open community project, you can not only explore but also share your own contributions.
For anyone looking to learn robotics by doing, this is a solid place to start!
Here’s the link: github.com/mathworks-robotic…
~~
♻️ Join the weekly robotics newsletter, and never miss any news → ziegler.substack.com
Adolfo Bolivar retweeted
(Finally) wrote up "Understanding Positional Embeddings in Modern LLMs."
A RoPE from-scratch tutorial missing from my LLM building blocks collection + what recent architectures do (NoPE, partial RoPE, hybrids).
Prob not the flashiest topic, but essential for understanding LLMs
Adolfo Bolivar retweeted
Till date, the best hands-on tutorial for
"Attention Is All You Need" that ever existed, tbh. It's certainly tedious, I won't lie but it's certainly worth it.
Link : deep-ml.com/projects/attenti…
Adolfo Bolivar retweeted
"Ignore the Writing: A Proposal for Peer Review in the Age of AI."
arxiv.org/abs/2610.38062v1
my AIE talk is up! come hear how the grok @bot team ships so quickly without compromise on quality
youtu.be/O_zcc6Xl12g
Adolfo Bolivar retweeted
Han vuelto a subir nuevas charlas del @aiDotEngineer de NYC y del Code Summit a Youtube.
El nivel de las conferencias es SoTA ahora mismo y para mí, las charlas más relevantes de IA a nivel mundial.
He vuelto a seleccionar las 10 que me parecen más relevantes. Va hilo 🧵 con todas.
Like, RT y guárdatelo para poder verlas luego. 👇
i learnt this lesson this year. it's helped me follow through on my ideas and not get stuck overthinking it
don't ask if you should do the thing. do it, then come back with a demo or something tangible
you can just do things
Adolfo Bolivar retweeted
🚀 گوگل یک PDF نهصفحهای درباره Harness Engineering منتشر کرده؛ مفهومی که به نظرم باید هر کسی که با AI Agent کار میکند، جدی بگیرد.
فرمول اصلی ساده است:
Agent = Model + Harness
یعنی مدل هوش مصنوعی بهتنهایی ایجنت نیست. چیزی که آن را به یک نیروی اجرایی قابلاعتماد تبدیل میکند، زیرساختی است که دور آن میسازیم.
اما این زیرساخت از چه چیزهایی تشکیل میشود؟
۱. Guides: قوانین و دستورالعملهایی که تجربه شکستهای قبلی را به دانش دائمی تبدیل میکنند.
۲. Sensors: تستها، Linterها و ابزارهای اعتبارسنجی که اجازه نمیدهند هر خروجی اشتباهی به مرحله بعد برود.
۳. Agentic Loop: چرخه برنامهریزی، اجرا، بررسی و اصلاح؛ با محدودیت بودجه و امکان ارجاع به انسان.
۴. External Memory: حافظهای خارج از مدل برای حفظ تصمیمها، وضعیت پروژه و دستاوردهای جلسات قبلی.
۵. Permissions: کنترل دسترسی ایجنت به ابزارها و تعیین مرز استقلال آن.
۶. Observability: ثبت و پایش تمام فعالیتها، هزینهها، خطاها و تلاشهای مجدد.
اما داستان اینجا تمام نمیشود!
چند پروژه متنباز برای ورود عملی به این حوزه:
🔹 Google Harness²
github.com/google-research/h…
🔹 Learn Harness Engineering
github.com/walkinglabs/learn…
🔹 Awesome Agent Harness
github.com/Gloriaameng/Aweso…
🔹 Harness Engineering
github.com/TrishamBP/harness…
و حالا یک مفهوم جذابتر:
Graph Engineering
تصور کن چند ایجنت مستقل داشته باشی که هرکدام Harness مخصوص خودشان را دارند.
یکی تحقیق میکند، دیگری تحلیل میکند، سومی کدنویسی میکند و چهارمی خروجی را اعتبارسنجی میکند.
حالا اگر یکی شکست بخورد، شکستش میتواند به سیگنالی برای اصلاح عملکرد بقیه تبدیل شود.
اینجا دیگر با یک چتبات یا ایجنت ساده طرف نیستیم؛ با یک سیستم اجرایی چندایجنتی روبهرو هستیم.
یک نکته که نباید فراموش کنیم:
Hermes را میتوان بهعنوان محیط اجرای ایجنت در نظر گرفت؛ جایی که ابزارها، مهارتها، حافظه و گردشکارها کنار هم قرار میگیرند.
Harness Engineering کمک میکند این محیط را قابلکنترل، قابلاندازهگیری و قابلاعتماد کنیم.
به نظرم آینده متعلق به کسانی نیست که صرفاً مدل قویتری انتخاب میکنند؛ بلکه متعلق به کسانی است که سیستم بهتری دور مدل میسازند.
مدل، مغز است؛ Harness، سیستم اجرایی؛ و Graph، شبکه همکاری مغزهاست.


































