FDE Epi : 10 | What Changed in AI, Work, and Power This Week | Intelligence Roundup

Hyper is a composable wipe coding platform, or as we like to call it, CVC platform to help knowledge workers build internal tools faster and better. So rather than scaffolding the entire solution with AI, Hyper helps build blocks of code through automation and then use Agent Ki orchestration to help stitch all those blocks of code to give more deterministic intent and AI tooling you can trust. So essentially it's a better way to build a custom tooling or custom software. The idea is to help developers use the AI tools maximally. The one unified workflow we've built agents like business analysts to help developers groom the requirements. And then we've built data modeling to transfer those requirements into a concrete data model. Then you bring all those things into Hyper AI Studio where you build tables, hook the databases, build MCP tools to hook with a barboard and build custom workflows all using Hyper AI. The idea of Hyper AI is to land up at a point where helping organizations self build their internal software systems rather than outsourcing to services vendors and coming up with a platform where the entire solution can fit into a mind of one developer. Both seemingly different ideas come together as one holistic outcome. So it has a dual use and I think it's much better way to build software where you're not outsourcing everything to the wides and we're not only using card code to generate all the things. It's basically a spectrum between vibing and architecting software. It's a new pl of work that we call composable wipe code.

Hi everyone. Welcome to the coding the spaces. It's our 10th episode, our weekly signal brief on what's happening in technology. So the news one of the days that software engineering has crossed the autonomous what happened? Basically the CUR recently released the support for ChatGPT 5.2 Codex in which even those they describe it as a current model for long running autonomous. More CEO showed the model training continuously for 7 days producing roughly I guess 3 million code of rust language and in result building a custom browser and rendering engine from scratch. So what's your take on this?

I posted about this I think a couple of days back on LinkedIn. Yeah, that's. So there are two things that's currently that's basically shaping this new wave of autonomous coding or identity coding if you like. One is the context. How can we help for example the models to have very large context and most of the technology that's currently being developed is like how can we come up with new ways of building very large context? Windows. So for example recently if you see a cloud code a couple of days back, Cord code has basically has updated their plan model. So for example, if you're using card it allows you there's a more called plan mode and in plan mode you plan different things. So what they've done is like now you can refresh their plan mode. Like for example the idea of that refreshing a plan mode is because they want to have that extra bandwidth available for the newer contacts. Once you are done with the planning, there's no point of reading. I mean there is a point. If you want, you can do that but if you don't, if you want it, you want to refresh that planning because you've done that planning and now you want to move on. So now if you refresh that planning, that planning thing, it would allow you to have that extra bandwidth in terms of context window, right? So one of the core thing is like availability the context windows. That's one thing that's driving this modern infrastructure for coding, economic or identity coding. And second is what you talked about Kasha, like how can we execute very long tasks end to end? So historically has been very good from middle to middle. It requires someone to start, it requires someone to end.

Right.

And it's good from middle to middle, right. But now what you've seen is most of the companies like described as system to think like you outsource certain tasks to the AI agents and they execute those tasks for like for a complete week. Think about it like if you're outsourcing an agent is executing a task in the entire week. It's not, it's not agentic or it's not resistance. I mean I'm calling, it's an end to end engineering execution, right? Imagine that you can execute an entire engineering task in a week by allowing AI agent to run for entire week. So that's what you're talking about. What's happening with the Cursor integrating chat GPT 5.2. Now it's not, it's a code they're integrating. So and then the founder browser you said, I mean I posted about it, they said about like at this point we don't know what does it mean to build a software because everything is one shot away from building anything. If you can build a browser in a week by just outsourcing a task to an agent, of course it's, it's, it's not like people are using that browser and stuff but you can have an idea of where the word is heading. So that's that's why important. But for software engineers, I think one of the things that we need to think about it is like agent coding is not just a copilot or assistance or for example prompting or for example intelligence or it's. It's basically you can outsource an end to end task to the to the agents. Now it's a complete engineering execution.

So the second use of the day is AI is now producing in novel mathematics. What happened? Basically a Math present at UC IR1 tested an internal git of croc 4.20 on open problem in I guess harmonic analysis. And within five minutes you, model discovered a new Belmont object, a math structure that human structure has been searching for years and haven't found it yet. So the question here is from is it just faster computation or AI is.

Actually doing research now then it is so. So I would just reiterate this news. Like what, what does it mean? So the professor in UC1 is trying to, he was trying to use draw to solve certain problems, right? So he used it for harmonic analysis. So basically harmonic analysis, what it means is like transforming complex functions in maths into wave waveforms, right? So that's what harmonic analysis is like. Now what he discovered is, I mean just try to understand what does it mean. What a new. I mean the model is discovering a new bandman function that's not ordinary. I think we need to understand what does it mean. I mean normally who does that? Postdoc from MIT, PhD in maths professors, they used to do this kind of stuff. Now the model is discovering bellman function. That is, I don't know at this point what does it mean to be a mathematician? I don't know that we discussed about solving out of problems in math. So the core thing here is to understand is like this is not calculation. Discovering a balman function is not a new balman function. It's not a calculation. Like it's not like you're outsourcing a GMAT exam to an agent and they're just performing that. Or you're outsourcing a complex computation just like you're out of something to calculate. It's not like that. This is basically discovering these inequalities in maths. This is basically conceptual space. This is an idea space in math. So what it means is like the model has demonstrated some earlier versions of performing deep reasoning capabilities that historically was only attributed to a great professor in math or something like that. So I think there are parts one AI roles in maths. I'm not sure I Can't comment on that. But other things like it's truly reasoning. It's not just execution, it's reasoning proper reasoning. This is and coming up with new novel ideas deep into the maths.

Exactly. So the third news of today is related to the Deep Seq. Deep SEQ discovers unused. What happened to circle at these Deep SEQ identified a huge shaped load. It optimized the balance between neural computation and the static memory. Using this Insight they built 27 billion parameters Ngram model that outperforms significantly larger than baselines. So why does this scale matters beyond just model size?

So scaling up basically means like the bigger the metal, the more you scale in your network or the more you scale the compute the more you scale energy the performance would get better. Or as for example like it's basically the equation is intelligence produced by 1 gigabyte of energy or something like that. The higher the energy, higher the compute, higher the bigger the models would be, the better the intelligence output would be. But I think why so if you see deep sea or Chinese models, most of the breakthroughs are towards how can they exhaust the existing infrastructure in ways that leads to new scientific breakthroughs. The reason for that is they Chinese companies are operating at a severe chip constraints, right? So there are lots of laws that not allowing them to have access to that chip infrastructure. Although they're very close. I mean down the line I probably would discuss a couple of the things for example how they're building new kind of chips. But what I'm saying is like currently they are under constant of producing certain scientific breakthroughs. So whatever they have they can use those hardware infrastructure and come up with new ways to produce intelligence. So they're continuously coming up with models and with breakthroughs that allow them to to get more from the existing infrastructure. So this Ingram model example that's basically U shaped it's basically a new approach of how they can balance memory with the computing base that they can avoid more they say more chips, more compute or they're trying to do more chipotle they're trying to use like probably they can get some algorithmic breakthroughs to harness new ways of introducing intelligence. And that what is what is powering that this again channel because they don't have access to that. So they're just applying what can be done in terms of research when it comes to models.

Perfect. So the fourth news of the day is the corporate tech is eating itself but happen basically Microsoft that spends roughly around $500 million per year as one of anthropic plot's largest customer. What next is that McKinsey test candidates on the abilities to critique outputs from its internal AI known as Lily. And Google launched personal assistant intelligence inside Gemini as well. So why are company becoming customers of your tool? They are competing with.

That's you talk about Anthropic. So Microsoft is spending $500 million to acquire services from Anthropic. And then you talk about McKinsey is testing the new candidates and just massive. Just think about it like historically what you consultants used to do. So they were. It's not like they used to perform extraordinary amounts of competition in their head and then come up with new insights about different industries. Now what they're practicing is like the new requirement is can you prompt their internal system like Lilly to come up with this new insights. So I think that when you talk about corporate stack is eating itself what it means what Microsoft does. Most of the revenue that Microsoft is generating I mean in terms of. In terms of software is basically database hooked up with certain workflows. Like that's what dynamic Stack is that every primarily that's everything that Microsoft does. So what they have is that distribution Microsoft is a sales business. It's less of a software business anymore. And I think that's an important thing. The new software business is if anyone the new kind of software business won't be building software. It's about distributing software. So every software business is now a distribution a distribution company. And that's what it means is like corporate stack is in itself like for example if at one end if you're using a tropic AI for $500 million a year what it's really doing is it's basically replacing your own internal software stacks. And because currently whatever software you want to build you can just crop it and interview your baseline scaffold. Most of the things you need engineer and to really bootstrap what kind of dynamics you need the micro dynamics or whatever.

Right.

So yes I think the thing to learn here is any entrepreneur who's trying to or trying to build a software business it's not about building software anymore. Right. It's about how can you distribute AI. That's most of the software business would be the next five, 10 years would be AI distribution businesses.

So different use of details related to the compute and materials apply what's happening Basically there are two main three main things Apple and call home computing with Nvidia and Amazon for glass then OpenAI and cines are 750 megawatts ultra low latency compute and Chinese lab Also built their chips with 90x pro group games. Are we reading the physical image of AI?

I think things are moving from algorithms to atoms. The true technology for a very long time was focused on building software and algorithms. But now I think software building is less of a lack of code and more of like glass fiber energy, heat latency. These are the first order problems. If you are again, I mean, I think I would like to plug this question with existing one. And if you are trying to do the software business, I think building a software business, the first sort of questions are like glass fiber, do we have access to energy? Do you have access to, you know, how to handle heating data centers and stuff like that. So that's a new, a new transition. So one of the jobs in America that's in demand is basically plumbers, electricians, right? So I asked myself, is, is, is plumbing, is the new software? Because all you need to do is make sure how to plumb all, all that stack together in ways that you can produce software at demand.

Energy and biology are scaling together. What's happening? Basically, Tesla broke ground on 375 million lithium refinery in Texas.

$375 million lithium refined lithium refinery in Texas. Yes.

Okay, and why is the high throughput design appearing everywhere at once?

Because I think what's happening here is like for example is number one, I mean you talked about Tesla Blue Ground on 375 million dollar lithium refined in taxes. The other thing was like coal generation declined in China and India. For the first time since 1973, Rice University mapped 100,000 gene circuits in one experiment. So what's going on? So basically there are two things that's happening in like the two things that are happening on steroids. One is simulation and there is automation. Both things has arrived at once. So once. So, so basically what it means is like once you can simulate cheaply. Just imagine if you can simulate cheaply, if you can test massively, if you can iterate faster. For example, if you see earlier versions of SpaceX, why SpaceX was so hard to pull off was like simulation was extraordinarily expensive. Simulating a rocket, right? So once, now what you can do is you can simulate cheaply, you can test massively, you can iterate fast. So every field has become a software. Like so biology is like. So think about historically, if you see iteration was primarily a needle. I mean primarily it was only confined to a software. The whole idea of mbp, you would build something, you test it and then you go back change it. I mean, ask hardware entrepreneurs that what does it mean to build an mbp? They spend millions of dollars in just building an MBP because iteration they have to be right? I mean they don't have those many shots to iterate. So for hardware founders, ask them what does it mean to simulate and then iterate. Right. For them it has been a big thing. Now if the simulation is easier, it's cheaper. What it means is like hardware manufacturing is easier. Most of the things that are, that exist in the physical spaces is basically requires lots of upfront cost. But now if you can simulate those things upfront in a software using AI then probably once. So basically it's a zero shot hardware zero shot physical goods manufacturing. Right. So robots are basically. So that's, that's what it is. So that's why it's a big thing. Simulation plus automation.

So the last news of the day is related to employment, identity and self ownership. So what's basically happening? CEO of the Robinhood CEO predicts a job singularity driven by solo founders even though the EU ban refusal of cash to protect inclusion. So what does it mean to own yourself in the era of AI?

So I think it's, it's, it's tough, it's most of the. I mean what, what does it mean to have a skill? What is scale basically? What, what does it mean to have a job? I think the, I mean if I had it, I had some, some Aladdin Kacharag or something, right? If I could have done. What I could have done was how could we bring authenticity to who we are? How can we bring spirit out of ourselves and bring to the table? Because that's the only thing that AI probably won't eat. Although I also think that's also some form of computation. But the more authentic you are, the more. For example, the Matthew, the guy of interstellar, he actually cloned himself that because he thought probably because to build a digital version of him, clone him was easy. I mean you can replicate his voice. So he actually, he actually patented himself.

Right.

So there's some very crazy things are going on. What it means is like I would say it's like the first time ever the value of truth, the value of who you are is going to be the single most important thing that would matter down the line. Every other thing that's not part of your spirit in terms of knowledge work that's not part of your soul actually is going to be, is going to be replicated. So with AI so humans generally have I think 100 billion neurons somewhat. AI is so just think about it. Like I posted this a couple of days back. It's very important thing. Human roughly speaks 10 million words a year. The intelligence produced by Rafa would be 10 million words a year. What it means is like 7 million tokens in terms of AI. So or let's round it up, 10 million tokens, you produce 10 million tokens of intelligence per year. Right. ChatGPT3 used to have $600 per million tokens. And if you multiply, for example, a year's performance for human, years AI performance for human, it's basically $6,000. With $6,000 you could have one person's AI output. But that was ChatGPT3GRO. Currently as we speak is $0.5 right. Per million tokens. So essentially what it means is your entire AI output is under $1. I am not sure what is, what does cognitive level mean at this place, at this time? I don't know what does it mean to. I mean one thing I can show off is like for example, if I'm talking to you, let's say I have. This is meta AI or AI glasses. So this is. So if I am talking to you and you have a mute computer, I have a mute computer. We both in this room, you and I are not the most intelligent one. Now we have another thing, AI. So basically probably as like humans would add negative cognitive value in few years time. So you probably so that's what's happening. So what it means, I think tough, bumpy rides. What we need to make sure is we need to think about how can we bring our soul into action, into our work, who we really, really want to be, who we really are. I think for companies, vision would be the single most important thing because every other thing would be meaningless than. And so yeah, identity, job, work, everything would be I think would be related to now your, with your soul and the bride would be bumpy. But I think that's a good thing. That's. I mean we talk a lot about like do it from your soul. We have to do it from the dil, otherwise there won't be anything left.

Got it. So that's it.