Muriel Médard / Optimum
Optimum CEO on why data propagation is blockchain's real scaling limit
In this episode
We sit down with Muriel Médard, Co-Founder & CEO of Optimum, to break down one of blockchain’s most overlooked scaling bottlenecks: data propagation. While most discussions focus on execution, consensus, or data availability, Muriel explains why the propagation layer is actually the ceiling for everything above it — and what happens when it starts to struggle under real-world load. From redundant transmission and wasted bandwidth to uneven geographic performance, we explore why traditional gossip-style P2P networking may be limiting Web3’s growth. Optimum is introducing Random Linear Network Coding (RLNC) and a universal bandwidth fabric designed to materially improve propagation speed and efficiency across chains. We dive into what’s live today, how validators can integrate, key performance metrics to track, and what Flexnodes enable in a permissionless mesh network. If propagation becomes reliably fast and predictable, what new classes of low-latency applications could finally become possible? Watch to understand the next scaling frontier in blockchain infrastructure.
@get_optimum on X · Watch on Refinitiv
- Data propagation is a fundamental scaling bottleneck in blockchain that limits performance of consensus and execution layers above it.
- Decentralized systems can achieve equal or better efficiency than centralized systems for data propagation without requiring centralized state management.
- Random Linear Network Coding enables more efficient data representation and transmission by leveraging mathematical flexibility of numbers versus physical constraints.
- Validators depend critically on propagation networks to communicate and reach consensus, making propagation infrastructure foundational to blockchain functionality.
- Optimizing propagation through better traffic engineering of data flows may be more practical than building entirely new infrastructure capacity.
Chapters
Transcript
Read the full transcript
I'm Ashen Addison from the Crypto Coin Show and today on blockchain interviews of Muriel Madard, co-founder and CEO of Optimum here to talk about blockchain data propagation, why propagation is the scaling ceiling for everything above it and the road map for blockchain adoption and what comes next. Muriel, welcome to the show and thanks for taking the time.
Ashton, it's a pleasure. Thanks for the invitation. Yeah, blockchain adoption is growing faster than ever, especially with institutions and enterprises getting involved as well. Them waking up to AI and now realizing the synergies between AI and blockchain and some blockchains might not have that scalability ready for this mainstream adoption and all of the other
factors. And I know that you've been studying this for a long time. I'd first love to hear a little bit on your background and how you got involved in this aspect of blockchain and adoption. then we can dive into Optimum.
Yeah, thank you. I'd love to do that. So as you mentioned, I'm CEO co-founder of Optimum. And I have spent my career until then at MIT.
I'm a professor there in electrical engineering, computer science. I hold the NEC chair in, software science and engineering for the school of engineering. So obviously have been extremely technical my whole life. And I've been leading the network coding and reliable communications group. And you know, the network coding and reliability,
you know, we're going to go back to that during our conversation because that's really key
in what optimum is doing. And for pretty much my entire academic career, I've been looking at what was at first just a deep almost philosophical curiosity. But I think this will resonate a lot with our listeners who who are you
know blog blockchain either enthusiasts or at least curious.
And the question is the following. Can you do as well in a decentralized system as in a centralized system? Yes or no? Right? I mean that's like a really fundamental question. And you know it's not about what you would prefer. You know you might philosophically be more comfortable with
something decentralized or not. Right? It's just you know a very key question. And for the aspect of propagating data of actually communicating data from you know a set of apps, sources whatever you want to call it to you know a set of machines, users whatever it may be who
want that data. The rather surprising question is the decentralized system does as well.
and in many ways actually does better for the following reason that in a decentralized system you don't need to keep track of state. Right. I mean, if you're talking about centralization, you're talking about having one really powerful entity that's then figuring out
exactly who does what and sending out instructions. And that doesn't work. That doesn't scale.
and so it has been, you know, a key interest. You can see why it's super interesting. You know, again, it's a fundamental question. And then you know one answer is just to say no actually you can do you don't have to go for centralization in order to have
optimality that's actually in our name optima but on top of that it can be done really easily right so it's one thing to say well in theory they should be the same you know that's maybe interesting but okay it stays as a curiosity it's another thing to say actually not only can you do as well or effectively better but it's it's very
practical to do so.
and basically that's my background.
Yeah, definitely it does get technical especially when people are just getting involved in crypto bitcoin and ethereum. Maybe they've heard of propagation from websites and traditional coding but once you are rebuilding the entire financial system and monetary system into tech
it's like oh god we got to get this right. That's right. Yeah, that's right. And it's very much like something like urban planning, right? You could say, "Okay, I need more housing. I'm going to put all these buildings, but then, you know, if you don't have the infrastructure and the roads, you know, you build these buildings and then people are saying, well, I don't want to
live there because I'm stuck [laughter] on the on the road for hours trying to get to work. you know, this is, you know, what's the point in adding more adding more apartments or whatever it is if, you know, I don't have the public transport or the roads or whatever it is to actually use them.
Definitely, that's a great analogy. And
with the current case of some of these mainstream blockchains that most people are familiar with, it seems like the roads, they're there, but maybe they're not super highways. Yeah, they're not super highways. And basically, you know, there's there's also an issue of are you even using your roads as well as you could,
right?
so,
is it that you really would need to put in a ton of money and build lots and lots of new roads? Or is it that you're basically, you know, the timing in your red lights is completely off and you're creating traffic jams all over the place for no good reason, right? Like and it's basically the difference between, you know, going in your pocket and potentially having
to put out a lot of money and then, you know, having to maintain those roads. You know, it's it's really expensive. Or saying, "Well, why don't you start with doing the traffic engineering correctly?" you know,
and what we're doing is we're doing, if you will, the traffic engineering. The difference between what we do and physical roads
is that
we're talking about data, right? So, we're talking about numbers and you can take numbers and you can manipulate them. You can do different representations of these numbers. you know, you can't do that with physical cars or trucks or whatever, right? You can't go, I'll take a sum of this truck and that car and see what happens, right? So it's basically saying you
know why don't we do things that intelligently leverage the fact that you're using numbers and that numbers are very flexible you know wonderful entities that are going to allow us to represent data in a in a much more efficient fashion. Definitely. I'd love to dive into the use cases for blockchain with this technology. Are
is this primarily just for money transfers as we've seen like sending Bitcoin from one person to another or tokenization of stocks or decentralized applications? Does this really cover everything or are there specific use cases for having the propagation setup?
That's a really excellent question. Thanks Ashton. So the it's extremely versatile as as you've just hinted
you know this comes in so many ways in the same way that transportation in our life comes in so many ways right and there the ways when we experience it ourselves you know like I get in the car and I'm trying to get somewhere that would be like the user trying to basically you know do a transaction on Ethereum for instance and then there's also the
indirect ways that we experience it. You know, the transportation is basically behind the whole supply chain. So, you know, I may see that my goods are more expensive or that I go to the shop and I don't have what I need when I want it or, you know, it takes forever to get something delivered. That's because you know I may not experience it directly but you know everything that I
do is effectively reliant on a supply chain that lives on top of the transportation network. It's the same thing with blockchain, right? So, our first go to market has been going more to the people that actually make the blockchains work, you know, the real engine behind the blockchains. And that's the validators, right? So the
validators are not entities that users are interacting with directly, but they're entities that users are crucially dependent on. [snorts] so they're the ones that are act that are doing the work of validating as the name indicates. And they have of course a really
crucial dependence on propagation because they have to talk to each other right I mean all of consensus relies on the fact that to reach consensus you need to talk to people you need to send data to people you need to hear back from people
and that's happening through this propagation network. So in a way it's if you think you know if the transportation network was you know a
complete disaster probably the first place you would see it is in how you could access the goods you need. Mhm.
you would of course see it in your own, you know, personal experience with the transportation network, you know, getting to from work, school, whatever. But you would experience it probably in even a more radical fashion with respect
to, hey, you know, I went to the shop and there were no life, no vegetables, you know, because
they couldn't get there on time. So that's that's that's been our go to market but basically everything relies on having good propagation and now if you're talking about blockchain the propagation is decentralized right by its very nature you have a system
which is as not dependent on geography as possible. Right.
and you know one approach that people have taken sometimes is to try to decentralize everything. But then you can ask well why are you doing blockchain right? If you're just putting all the machines in one room how is that different from having a single machine?
you know would you really call that blockchain? What is the reliability aspects there? security, trust, you know, you're basically just pretending to talk to different machines. And while they may be physically different machines, effectively they're not. They're all in the same room under the same control.
Great. And looking back at Ethereum over
the years, it they spent many years with one consensus mechanism and then switched it to the proof of stake which many said was sort of like the analogy was changing the engine of the plane while it's flying and hoping it doesn't crash. So, you know, how has that journey with setting up the consensus and the propagation properly with Optimum from the beginning been so
far and where is it at right now?
That's a great question. Thank you so much, Ashton. And I think that the you know that phrase from from the Ethereum experiences is perfect. You know, you don't want to be changing engines on a plane. It's really scary. So we don't touch the consensus part at all. you know that's really the core
Of these at once you know how they do the consensus what kind of proofs they rely on and we are simply trying to change the propagation and also we're trying to change it in a way which is supplementary not a you know we're not yanking everything out anything out
what we sometimes say is we take a hypocratic approach approach to
networking which is first do no harm. It's not like oh I'll do this but yeah it might not work out. No, you know I'm not going to do anything that leaves you worse off than before.
and I think that's that's key. So what we do is we have a system which we call mum P2P. So some of our listeners may be familiar with P2P but I'm sure to many it's a new terminology.
P2P stands for peer-to-peer and is actually the mechanism by which systems like Ethereum but by no means is Ethereum the only one to do this do the data propagation that we mentioned before. And what we're doing is we're using I mentioned at the beginning of our conversation that I lead the network coding and reliable
communications group at MIT. So we use network coding and in particular a decentralized way of doing network coding called RLNC.
Again putting a pointer to some of what we chatted about earlier on. I talked about the fact that you know we're not sending cars, we're sending data. Data is malleable. It's numbers. So what coding means is actually making
equations. And what we do is we make these equations in a decentralized way in the network.
so those equations basically propagate the data optimally and what we mean by optimally means you could not have done this faster, right? And it's not often that you can have mathematical proofs of doing better than anybody else. But in
this case, we do have those proofs which is you know kind of mind-blowing when you think about it that you can do that you know. So often we say optimize when we just mean improve and here what we say optimize it means actually you know like definition optimize [laughter] like you know
that's very cool and I was reading into through optimum beforehand and I saw you
know the major competitive advantage this RLNC random linear network coding u so that's really the secret sauce I'm not sure how secret it is but to the decentralized approach for getting things done as fast as possible.
That's correct. That is correct. And I think you know some it came out of my lab. I'm a co-inventor on the technology. When we first put it out
there I mean we got so much push back and the notion was you must have made a mistake in the math. There's no way you can do this in a decentralized fashion randomly and get the best throughput like there's no way you should be able to do better if you had some control and the answer's like well you know read the paper now to be fair the early papers
were pretty turgid my fault I get it they were pretty turid and so you know, even though they won all kinds of international awards because people are like, "Wow, this is amazing." I realize [laughter] many people didn't read it or didn't read past the first two, three pages, and I don't blame them. [laughter] but actually the concept by itself,
Is pretty simple, if you will, even though the reasoning of why it works may be counterintuitive and if you want to do it in a very precise way, it can be quite involved. It's basically, you know, let's go back to this data we talked about the numbers. Numbers are variables, right? If you knew what the numbers were, they wouldn't be variables. You wouldn't have
to tell anybody. Yeah. So, for me to send information to you means that there's some variables you want to know, Ashton, we're going to call the variables X, Y, and Z. And, you know, the information is out there in the network. And you know traditional systems like P2P that I mentioned what they do is they'll be sending these variables around and
here's the problem. If you want to have everything centralized, you might say, "Okay, Ashton, I have XYZ. I'm going to send X along this particular path to you and Y along this particular path and Z and if it doesn't get there on time. Somebody's going to tell me they didn't get there on time and then I'm going to transmit Z again and you know we'll
do some really complicated centralized optimal you know crazy schedule for you. Alternatively if you want to do this in a decentralized way what P2P does is it says well I'm going to send multiple copies of XYZ
and eventually you'll get enough copies that you'll get XYZ. Here's the problem. Because I don't know exactly, it's decentralized. I don't know exactly when
you're going to get X, when you're going to get Y, when you're going to get Z. I say, well, you know, let's say I'll send you three copies of each. All right, great.
and now you're sitting there and you get maybe two copies of X, two copies of Y. So you got four variables, but you still don't have Z and you really need Z. It's like this is, you know, as we
talked about, a lot of this is financial data. It's not like, oh, well, that's good enough. It's like, no, I like, you know, [laughter] you know, you're you're doing a transaction and you're like, was that a plus sign or a minus sign? You know, I really need the [laughter] variable to tell me, you know, I'm not going to just get started and hope for the best.
So basically what happened is, you know, you only needed three variables. You've used you know four units of data basically four variables right so you have already used whatever 33% more than what you needed in terms of bandwidth and you still don't have what you need
and then you know and then eventually you'll get Z but you know by that time
maybe it's like the fifth thing that you receive you're like okay this wasn't very efficient
and if you go well you know I'll just keep sending more and more copies, you just clog everything up, right?
so, you know, we go back to this example with the with the physical network and you go, well, you know, my deliveries are not really
very reliable. So, what I'm going to do is I'm going to go and order, you know, my coffee capsules, like, you know, one from Amazon, one from Target or whatever, and then I'll get the one that I received first. And then you got all these, well, coffee capsules at least keep, but imagine if it was your groceries, right?
Yeah.
Okay. I just wasted a lot of money just
because I really needed it tonight and I wasn't quite sure which one would come first, right? Not a good idea. So the idea is we then instead create equations. So now you get one equation with XYZ, another equation with XYZ, a third equation with XYZ, and you go, okay, I've got three unknowns, got three equations, done,
all set, you know,
and therefore you don't need to send
a lot of equations to try to make up for the inherent unreliability that you get in a system. you know, the approach that you could say is like, well, why don't you send them to me directly? But, you know, the world is huge, [laughter]
right? Yeah.
And, you know, so one of the things that I've I've done a lot is gone to like kindergarten classes and
try to teach about networking. And one of the first things I have I always tell the kids, okay, you know, you need to have some yarn, you know, make sure you have some yarn. And then, you know, I show them how networks work. And you know, first you have two kids and you connect them with one piece of yarn and you're like, "Okay, great." You know, so
whatever Bob is able to talk to Ahmed or whatever, you know, and then you add Suzie and then, okay, you know, you put three pieces of string. That's okay, everybody. And then you take more kids and, you know, the whole thing looks like a plate of spaghetti and they're like, "That's not going to work, right?" So that you know you can't have direct links from everybody to everybody
because you know there's a lot of people and it grows with the square of the number of people.
So you need to have a network you know it needs to be the sort of messy world that we have. It's again like if you were in roads and you go oh you know it's taking me a long time to get to work. I know what I'll make a road from me to my job directly like and
everybody's going to do that. [laughter] That would be great.
It'd be great. But you can see the cost. [laughter]
Yeah, definitely.
So, with this technology, you know, there's a huge network effect on Yeah.
what we've built so far with blockchains in general and you know, new technology keeps evolving. Is this something that the current blockchain
infrastructure can take advantage of or maybe compatibility to port things into the optimum network? You know, how can we build on what's already been built?
That's a really really good question. I think this gets back a little bit to the theme you were saying about trying not to change engines in flying airplanes, right?
and you know, the FAA wouldn't
approve and neither would the passengers. So what we do with respect to, you know, existing blockchains, we've started on Ethereum, but we're already starting to migrate to other blockchains. Is we work with the infrastructure that's there and then just put a network system that's additive to that, just improving on that.
but again, very much with respect
for what's already been built, what already works, and what people are relying on. For new blockchains you know can't can't disclose but we're in some conversations there of course you can say hey look
why don't you do this from the start you know it will be smoother and it will save us the trouble of having to you know ensure that backward
compatibility with a nonoptimal system and just adding to it
so that's has been our approach with respect to existing blockchains. Take what's out there and just improve it without breaking anything, without removing the plane engine, you know, maybe if you will adding a booster to it rather, you know, yanking it out.
but now you go, hey, look,
if you're going to start from scratch, you know, you should just do it right.
Yeah, definitely. I agree. Earlier you were talking about the validator network and the importance of it in that the decentralized nature of it through geography and you know making sure that it's optimized for speed and transactions as well. It's always a balance. I saw there's this thing called
flex nodes with optimum. Is that how can can people get involved like for example you've never mined bitcoin or you don't have a server knowledge you know is that something that more regular people can get involved in or how does that exactly work?
Absolutely. Yes. Yes. So that's that's another great question. So let's go back to this notion of these equations
right. So basically the idea is that what we do is we're doing this data acceleration network by the transmission of equations. Anybody can help in sending equations right you're taking equations from the network and you're pushing equations in the network to help other nodes. From the point of view of say let's start with the cost of the
machine it's very low like you know you know my laptop right now is more than enough it takes you know
about half a CPU does the trick and
you know unless you're doing something pretty intense on your on your computer half of your CPU usually is idle right
from the point of view of bandwidth with, you know, whatever you send is helpful. So, it doesn't have a
gating factor where you say, you know, you need to send me at least this much otherwise I can't use it.
it's really something that's, you know, anything that you can do is helpful.
And then the other aspect is the uptime. So you know as you said you know if you're going to mine bitcoin or if you can if you're thinking of saying running a validator on ethereum or salon or
whatever it's a big investment we mentioned the compute you know you need to have very stringent requirements on your machine of course you need to stake
and some chains require you to put a lot of a lot of investment in the staking but you also need uptime
and that uptime can be really challenging particularly in certain geographies where you I don't have
that kind of uptime you know I was talking to somebody who runs validator nodes for Ethereum you know from home and you know he was telling me he has three ISP connections and then he has to kind of like you know he one of them is basically just there for backup and then he does some clever you know merging of the two first ones to get the full bandwidth. I mean, it's it's it's a lot
of work.
and it's a, you know, and then of course you have the risk of slashing, etc. right? And then, okay, you can get slashing insurance or, you know, pool the risk with other folks, but you know, again, it's just it's a little scary.
Yeah, a lot of factors. [laughter]
No, there's a lot of factors. you know, I think I have always deep deep
admiration for folks who do it. And now with a flex node, you go, look, as long as I don't do anything wrong, like, you know, give you bad equations and of course we have, you know, as you can imagine, fantastic, you know, math to deal with if somebody does. So, I won't go into that unless you want to. But, unless I don't do anything
wrong, I'm just getting paid for what I did. You know, if I gave you equations because, hey, you know, you needed them, I had the ability to give them to you and they were helpful, great. If I close my laptop and I was doing something else, hey, that's okay, too. I just, you know, I don't do anything. I don't get paid.
Fair enough, right?
so it's kind of like again more like
real life. You should be getting paid and rewarded for the useful things you did. You know, not sort of punished for the things you didn't do because you were not in a position to do them. Yeah,
definitely. And I love that you know, there's been many examples in automotive and physical AI that about how inefficient, you know, your car is
sitting at your house 90% of the time
and you've paid for it. And most people, they know that, you know, they're even when they're using their computer, the CPU usage, most of it's not being used, never mind overnight. And looking for opportunities to utilize that CPU power that they've paid for. It's at their house, but it's just not in use.
That's right. And you paid for your
cable at home, right? So, you paid for your bandwidth. You're sleeping or you know you know unless it's when you know in my household when you know my husband and I are doing things and our son is also you [laughter] know also playing whatever he is you know if it's you know most of the time you have spare bandwidth also and even if you feel
like your Wi-Fi is a little a little clogged up you know chances are that your you know your cable or your fiber is not
and therefore you know why are you paying for it and not getting money for it and also helping other folks right so you know you're getting money because you're actually doing useful work that is you know not honorous
And it's understandable that you know if you lose your cable or whatever it's it's okay you know
okay you don't get paid during that you didn't expect to.
Yeah. At least at least you don't have to have all those hard setup costs like they that's why I haven't got into Bitcoin mining or these other validator programs because the upfront investment
is just for that. You know, it's like you're going to have a computer anyway to do your daily work. But there's a lot of upfront investment. And talking about earning, can you dive into how exactly that works? you know, if you just contribute a little bit, you is it is there an optimum token that you get paid in or other crypto or fiat?
Yeah. Yeah. Yeah. So, we don't have a
token currently and you know, I mean, we do plan on having one at some point, but you know, I don't want to give time frames or anything. So, I'm going to talk more in a general way of the mechanism, right? So imagine that your validator Ashton you know you finally decided to buy the bulleted moon so you went and said okay this is
really expensive and my opex and my capex okay I'll do it okay and then you say okay oh gosh you know I'm about to miss my attestation you know hey mural you're right there come on you know give me a few equations like oh yeah Hey, I gave you a question. Phw, that was great. Thanks a lot. You know, here I'll reward you, right? And then and then, you know, the
idea is you did you did something which was valuable to you and I was able to help, you know, we figured something out, [laughter] right? Because inherently there's true value. I did something that helped you and maybe you were an Ethereum validator. Maybe you were Salana validator, maybe you were a BNB validator. I mean, you know, that's that's kind of like that's
your business. And I just happened to be able to pull equations and give you equations at the right time for what you needed. So, you know, there was real value and there's going to be, you know, fair exchange of value between us. You know, this also means that now if you were in a in a geography, you go, I'm not quite sure that I can do the uptime.
Maybe you'll be less hesitant, right? Because you go, hey, look, you know,
my connectivity has suddenly gotten better because people in my neighborhood are running flex nodes
and they're helping my neighborhood. This is great, you know. So definitely when we do our tests one of the things we see is that you know we do tests both with a kind of no
distributions that are more typical currently which are generally sort of North America
a certain parts of Asia Europe right those tend to be the more typical arrangements of nodes but we do a lot of tests which are massively geodispersed setting you know not just the geographies I just mentioned but you
know Africa other parts of Asia, South America and it really really it's you know Australia you know it really really is stark how much better how much better you do. So it also allows all of a sudden a much more inclusive geography in terms of people being able to operate. And of course you know the
more people can participate the more these chains can scale
and the better it is for everybody. And you know going back to the user remember we were talking about the user and the roads and so on. You know, one of the things that's great for a user is that they may not notice again, you know, the roads that made the lettuce appear on the shelf, but all of
a sudden it's like, I've got letters there all the time and it's cheap. This is great. [laughter] So, it's like, oh, my transaction, you know, it happened quickly and I didn't have to pay a lot of gas. I'm happy. So these are the kinds of things you know that people notice even though they may not interact directly with some of this infra propagation they do see
the effect and the benefit.
Definitely. What are some of the major milestones coming up? I think 2026 is going to be a big year for optimum. What are the major things that we have coming up ahead? Yeah. So, the one that is coming up soonest and I think is one of the most exciting is in about a month we will be on Ethereum mainet.
so we have been Yeah, that's big. Thank you. Yes. Yes, we are excited. And you know we have been working with a host of just amazing validator partners. you know, I mentioned that was how we did our go to market. And they've really been fantastic. You know, just super involved, super positive, super helpful. And
you know, and people like Everstake, P2P.org, you know, Kill Block Demon, Nethermind, I mean the list goes on and on. And we then go to Maynet and you know then that's when you know the rubber hits the road. In order to have a business we think it's important going back to the example we gave before right where you know you
wanted something I gave you something that you wanted and you know you got a return from it is we need to provide a better return for the validators
because that's how they stay in business and if you know they're not in business then bye-bye blockchain [laughter] so I mean it's basically it
[gasps]
so it's really really important to
make it so that we improve their business and you know the main indicator for them is their annual percentage yield right their APY are they able to you know stake and get return from that staking
and for us it's about improving that for them. Yeah, definitely financial incentives are important to make people do things. That's for sure.
it's also because you know you can't expect people to put in that level of effort without you know the they it's it's too much.
Definitely. Yeah. I'm excited to see that EVM mainet launch. Would love to follow up again. And there was many there was many other things that we spoke about that it's like this deserves a whole another episode by
itself.
so
I would love that. I would love that question.
Looking forward to it. Thank you so much Muriel for all the insights into this how we can propagate optimally blockchains and we need that if we're going to drive adoption into nontechnical sectors. We need to be optimized before they adopt and during and not back testing it after. So
this technology is important. Congratulations so far on all the success with Optimum and let's follow up again in the near future.
Thank you so much Ashton. I would really like that and as I mentioned at the beginning really appreciate the invitations. This was a fun chat.
Related coverage
More interviews
How Birdeye bootstrapped to 5,000 enterprise clientsOct 7, 2026
How Anvil lets you use crypto as collateral without sellingOct 6, 2026
Pawel Mastalerz on how Daski helps AI agents navigate the webOct 6, 2026
Why banks are building for Web3 in silenceOct 5, 2026
How TrendTrader Pro raised $1M for a token buyback fundOct 2, 2026
Illia Polosukhin on NEAR's privacy features and AI agents in 2026Oct 1, 2026