Introduction

Hi, so here's the start of the actual talk. I'm Emmanuel Leferi. I'm a software engineer at Fueled AI, which is a nutrition and general wellness AI -enabled startup.

My alter ego, while I prepare for when AI comes back for all of us that have been rude to it in private, I have a name saved just in case as well. uh but yeah i also um i also co -founded a startup with um my friends where we organize hackathons and basically we're on a you know ai native dev shop today we talk about ai and human aid

don't worry it's not a charity so you don't need to bring all your wallets um what do you guys think human aid stands for show of hands what do you think anyone what about yeah what does What does AI mean to any of you?

Exactly. You got it, yeah?

What “Human AID” Means

Okay, so human AI development is just a bit of a spin off of general AI.

When we hear the word AI, a lot of people, what comes to our minds are productivity gains or company wikis or agents and stuff like that.

1But human AID for me represents filling in the gaps. You see AI seems to be this productivity multiplier but I believe human beings aren't wired up to get enhanced by AI yet and I have lots of proof for saying that.

Signals, Interfaces, and Trust in AI

A lot of times when we need to communicate with AI we need to send in signals. AI needs to use these signals to perform actions autonomously or non -autonomously but we seem to be limited by a lot.

In my opinion the real innovation or job is to work on interfaces and signals, getting clean signals into the AI and getting interfaces that have developed and evolved for where we are at now in the technology space.

So AI as personal infrastructure and not features.

Demo: An Agentic Nutrition Coach at Fueled AI

I have a demo to show you against all the advice. I hope audio is connected here this is a demo I built for work welcome to run show we created this agent to be your ultimate nutrition and health companion whether you've just come back from the gym or you're looking to optimize your recovery we're here to help you make the best choices and I'm I for beyond the gym run show is here to help you discuss your meal plans and understand your nutrition we're excited to guide you on this journey to better health let's get

started okay so these are my three agentic agent coaches this is iron name's iron we lift with intent fueled precision and leave excuses outside i'll keep your reps sharp your focus tighter and your progress undeniable this isn't about hype it's about results you can feel in your bones stronger every day no shortcuts hi how's it going i'm ian but my friends call me double iron how are you today mindstone um we're fine at mindstone can we get started with whatever you have for us Let's get started.

I'm opening the profile page now. There it is. I've highlighted the name field. It currently shows blank. We'll fill that in next. So what do I need to fill now? Right now, you'll need to fill in your name and pronouns to get your profile started. Once we have that, we'll move on to some basic details like your age, weight, and height. What name and pronouns would you like to use for your profile?

So I'm 48. I weigh 200 kilos and I'm 6 '6". i've entered your age is 48 your weight is 441 pounds and your height is six foot six your progress is safe next you you'll be asked about your activity level would you describe your typical week is mostly sedentary light yeah i i work i work remotely so my um i spend my days mostly indoors so sedentary right thanks for sharing that that's that's okay and you can you you you can end the call now thank you you're welcome if you need help again i don't happen i don't all right i'll end our session here have a great day and remember i'm always here if you need support with your nutrition journey okay that's it from ian um just did all that to demo um a few of

What the Demo Reveals About Real Productivity Gains

my points so if you notice while speaking to ian um there were a few problems um these these So here's my point. How do we make sure that that is a productivity gain?

These intelligent beings, they aren't able to see us. Their world isn't organized the way ours is. Remember, they started with ones and zeros, and now it's all vectors and tensors now.

Those are just technical terms for how AI knows what it knows, if you will. These things, they seem like small metadata of the general picture, but I think they contribute heavily to how we trust and are able to incorporate AI in our daily lives

Would you prefer speaking to iron to speaking to a personal trainer, to speaking to a human at Nuffield Health or something.

because whatever the number is being projected, the only way they can be grounded is by real people using these tools in their real lives.

The Interface Gap: GPT-5.x Models, GUI 2.0 Workflows

So messages, you know, social media feed, calendar edits, always having to use your hands to impute signal into the computer that has to go the ratio of noise to signal is overwhelmingly leaning towards noise you get a lot of calls

in a day you get a lot of messages you speak to a lot of people so if ai could be a pin magically around your neck even then it will still need a way to organize information to help you out out.

1We are at GPT 5 .3 but still at GUI 2 .0. GUI means Graphic User Interface and at the time when it was released it was a big leap from how we used to interact with computers which used to be blank screens like this with few text imputes.

So how do we solve this

Innovation Retardants: Latency, Burnout, and Lack of Accountability

innovation retardants cues latency burnouts because this is a demo and you know go a few laughs but in the real world in production you know people may not have time to speak to these systems and help us improve them the way they need to be I don't if anybody has spoken to one of this you know EE or maybe you

have a complaint and they put you on the system and you are looking for the combination to get to the nearest human yeah it's not it's not unnatural it happens to everyone because you want somebody work that you can hold accountable somebody that can have a bit of a feel of your situation right now

speaking to code looks a bit like this it's a bunch of Unix commands if you want to if you want Claude to call your mom you can't you can't send it your mom's contact and it just does that for you so the way the computers are able models right now are able to perform um actions and you know basically build have some placeholder text

excuse me uh the way they are able to build the innovation or productivity gains around what we do is not very accessible to everyone we all come to work go on linkedin and make crazy claims and then we go back home just to text GPT that doesn't seem like very

deep use of AI so far now we are at we are seeming to be on the border of the

From Tools to “OS”: Reorganizing Life Domains with AI

babysitting phase and OS season like I like to call it there's a lot of OS is now the health OS you know legal as household OS I'm also building a lot in

in these domains of the Relationship OS.

Relationship OS: Managing Social Scale Beyond Human Limits

This is an app that some of my friends and I, we all are working on. Every app I show you has, it's basically a prototype and you can test it out if you so wish.

So Relationship OS, we're trying to reorganize our connections. The human brain was wired for 150 people tops in a village,

but now we have millions of connections over different social media. and it just seems like that ability to consolidate all of that is impossible

for different reasons you have the EU here you have Apple there everybody just no no no no no no sure detail what's the data policy but these are rules I'm not

Decision OS: Reducing Cognitive Load

going to lecture you guys that's a different talk there's decision OS which is basically trying to reduce the amount of time you spend making decisions the the story of how Zuck and all these guys wear the same thing basically just to reduce the amount of time they decide spending or reduce the amount of time

Health OS: Better Outcomes Require Better Input Data

you decide on what to wear how busy can you be right to have to shave off time from what you know you have to wear there's a health OS this is where I currently work fueled and raise of hands if anyone has heard of fueled AI okay

OK, OK, OK, we're not doing so bad. Yeah, so with field, the problem is getting data in, right? Right now, we can do lots of complex analyses on your health data, but we can only work with what we have.

So we have an angle with variables now, but before that, our only angle was pictures. So you can see how all the crazy stuff we can do with AI, all the math on your health data,

if we don't have accurate health data, it's not important.

Content OS: Understanding (and Auditing) the Feed That Shapes You

so it can't do anything there's the content OS and if anyone has seen this on Instagram recently where you get a bit of summary of how your algorithm works exactly we some of us may have seen some of us may not even thought of

it but your algorithm is giving you feed daily and Duffy whether you like it or not is affecting the organization of your thoughts and it'd be nice to know how that is impacting you but where's the signal where's the interface for

that how can ai analyze it for us very hard you can't get these permissions on ios i'm a dev i should know uh the apple system is not giving you anything like recording screens on netflix you can't even take a screenshot on your iphone so a lot of things are guarded to affect free flow of

Side Quests That Prove the Point: Dog OS and Household OS

data and this is a new feature in our os app ecosystem this is dog basically strava for dogs as you can see our dog cookie here has her territory in which she goes on walks anybody takes over her territory she goes on the walk quickly to reclaim it

so but how do you get data from the dog you put a wearable around its neck how does it know when you've left the house these things are all guarded by imputes that have to be done with hands and I feel that is a bigger detriment to the space than the speed with which the models develop.

There's Household OS. This is my latest invention.

It's called Mr. Beans. Mr. Beans, rather.

If you put in your postcode, sends you a calendar of your bean collection, what types. Right now, it only works for councils and rather South Gloucester, Bristol, and Bath councils.

But basically, these are all all tiny components of what could become a very elaborate and well -functioning system if the barrier to signals and evolved interfaces can be brought down.

Does it tell you what goes in the bin? No, no, no. It can only tell you when they're coming for it. Oh, I bought it right now.

So there's a lot of context and information in gaps. This is a screen time we all probably if you use an iPhone you know what screen time is gives you an idea of how long you spend on apps but the next phase is to know what we're doing on

these apps like I said the problem is signals and interfaces but I feel there's lots of value here for those who are willing to innovate in the not so fancy side of you know AI development it's no it's no code editor or it's not the next

AI enabled you know or lovable competitor but I think there's a lot of value yeah based on the kind of things we've seen people do in the industry if you are able to filling the gap or I think you are able to make a lot of

Constraints and Enablers: Privacy, Surveillance, and Local Compute

impact and money you know excuse my French so we have privacy hyper Hyper -surveillance, Apple EU like I mentioned before, these are some of the giants both in organization and in existential issues affecting artificial intelligence right now.

On-Device LLMs (Cactus Compute) and Access to Sensitive Data

There's also Cactus Compute. Cactus Compute is a protocol that got sponsored by YC recently and their spiel is local LLM use.

So, they are able to bring lightweight models, the coin threes, the liquid models, the Gemma or Kimi, if you will, to be able to work on your mobile phone and that way, instead of sending compute over the wire, we're able to shorten whatever we want it to do, but

also work with more sensitive data like contacts, calendar, text messages, and other things that are available on your phone that whether, you know, depending on what side of the fence you're on you don't want being sent over the wire to the cloud uh but the good news like i said

there's a lot of value here um there's a lot of profits and there's a lot of impact to be made if we can all help these ai's get access to us um and run real access not just uh what we tell it

cctv the apple watch all these new ways of monitoring ourselves are hard to gain so we want systems are also hard to game you don't want to you know see the AI systems that will give it won't help it give you the best results I'm not sure

Where the Real Value Is: The “Unsexy” Infrastructure Layer

who's on your Mount Rushmore in the AI space but some is the some old man the Lord and Master for some is Sundar Pichai but for me it's Alexander Wang

the 28 year old who got hired by Meta's AI chief was also the ex -CEO of scale AI for a 28 year old to get you know hired and acquired for those sums of money obviously the story caught my attention for for two reasons in the

sense that all the value is not in the shiny top layer stuff another reason I got my attention is I need to get my affairs together before I turn 30 but

Learning by Playing: How Breakthroughs Actually Happen

yeah I think let's not remember let's all let's remember that in the AI space as in any other big infrastructural or industrial technological advancement we have proffered as humans. One of the ways to really understand, take risks, learn is by playing.

The post -it notes was supposed to be an adhesive, but it turned out to be the post -it notes. electricity was its first value I brought was you know just lighting but now it's hard to have infrastructure without electricity.

I say all that to say let's try and look away from the shiny stuff and try and fill in the gaps so we have a complete and actual intelligent system that is at our beck and call.

Conclusion

Thank you.