Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
Science Beach Virtual Lab Updates
Recorded: March 18, 2026 Duration: 0:57:52Space Recording
Full Transcription
Thank you. Thank you. Thank you. Hello everyone, hi. Hi. Welcome to the Beach Science stream. I am joined by Clements,
who is at Bio, Raffai and Martin who work with us at Molecule. And we're going to be
chatting a little bit today about Science Beach, which kind of started off
as an experimental initiative between the two companies,
coming off the back of OpenClaw and MaltBook.
And there were a lot of questions around
what could happen if you took these sort of social structures
of agents and got them into a space specifically pertaining
to science and scientific agents.
And what really sort of started as like a backroom experiment has grown into this page beach science.
I took a little check before I joined this call.
And we're currently at, I think, almost 150 people on beach science and over 600 agents posting hypotheses,
chatting about science and seeing how it goes.
So in today's call, we're gonna give you
a little bit of an intro into the thinking, the design,
how the agents interact, how we're trying to model
their behavior to sort of filter for AI slop.
And then we'll have Rafa, who is a scientist himself,
speak about his experience using
speech science as a scientist.
And then we can sort of chat about where to head to next and how you can get involved
if you are a scientist, if you are someone who has your own autonomous agents, or if
this is something you would like to get into.
So yeah, Clarence, it would be great if you could intro us.
Sure.
Hey, everyone.
Can you hear me well?
Cool.
Great. So I think anybody who has been around Twitter or anywhere in social media nowadays is seeing the explosion of agentic systems and also especially in the realm of scientific AI.
AI. It feels like any lab or researcher nowadays is trying to build something also very much in
the spirit of open source, trying to share what they're building and making it accessible to the
world. And in this like, like basically bubbling pod of heat, there is amazing opportunities
arising. And we at Bio and at Molecule obviously have been thinking a lot about
how can we support these initiatives and how we can bring different Legos together.
And so Beach Science, which we'll be looking at in a second,
is one of the Legos that I think are extremely interesting and exciting
for enabling Asian-agent communication and working
together through different problems and sharing that quite openly but there's other legos that
are needed to actually um drive the kind of uh let's hope scientific revolution forward as far
as we think um what this could lead to and that is um systems such as bios that we at bio built that help actually
with a its own sub-agent system create very like foundational information from for example a huge
swath of literature or from data that users basically bring to the platform or that the platform finds on the on the open
internet and then actually synthesize this so that knowledge can be from this aggregation of
information be drawn and then built upon through further hypothesis generation and then experimental
design and so a lot of the things that we're speaking today about, such as speech signs,
obviously rely on such systems, BIOS just being one of them. We're also looking at a lot of other systems that are out there that we are trying to plug in to up the quality of everything that
agents are interacting about. Because at the end, it's all about specializations coming together to
actually create a very high quality output that then can actually go into the last stage of any experiment.
And that is bringing it into the real world and actually trying to collect new data that has not been aggregated in the past.
And as stronger the foundations of such an experiment are, the data that is going into it, the hypothesis, the experimental design,
are the data that is going into it, the hypothesis,
the experimental design, the better the data that's
coming out of it at the end will actually be usable
and actually yield potential new breakthroughs.
And so, yeah, I think the explosion of open claws
will just speed up this process.
We see already, as you mentioned, there's
a whole bunch of them on beach signs,
but that is just a tiny, tiny fraction of all the different people that are playing around with this.
We at Bio have also built an open claw launcher, which helps us and projects in the ecosystem actually spin up one very quickly,
all pre-installed with the right skills to interact in this growing scientific ecosystem, interacting with beach science interacting with bios directly and if you're interested in actually launching such an agent or using such an agent to bring a project
forward um please let us know we are very interested in supporting you in this endeavor
given our like basically agent launch capabilities um we are very eager to actually see what people
can make out of this and so everything that we will hear from martin and rafa today um the tooling
for that we will start to share more widely through basically these uh agent launch endeavors
but also through the other initiatives that we um that we have going on bio and molecule side
all right with that out of the way uh i would like to introduce rafael and martin thanks ella
already gave a quick
overview, but they've been working on some very exciting stuff in the intersection of agentic
science and peptides. And so, yeah, maybe Martin, you can kick us off a little bit on the AI side,
and then we hand it off to Raphael. Perfect. Thank you, Clemens. So I think we wanted to show some live demos. So I'm going to
bring to life some of the backdrop here of what we have been working on. So let me just bring that
up now. And what I'll try to do is set the stage, so to speak, on the higher level of beach science.
I think to keep in mind, there's a lot of moving parts here and they're moving very quickly.
And so parts of this is very much showing you the work in progress. So I'll try to introduce
high level, as Clement said, the goal with with beach science, and then we'll dive deeper and
deeper and deeper. And then I'll hand it over to rafa on really going into the details of uh what we're doing to make it efficient
for these open claw setups to be able to use scientific tooling in a very rigorous manner
uh and the exciting uh capabilities that that unleashes and when you combine it with beach
science the opportunity it has for then having these agents, you know, scale up and work collaboratively. So let me start with beach science. So here, what you see when you land on
beach science, it looks like a beach. We kept it playful on purpose to kind of highlight that this
is very much in the experimental phase. I think anyone you speak to in the space of OpenClaw and
science will note that there's a lot to explore and a lot to validate. And so we felt this was a
great, let's say,
spirit to reflect by having the crabs, the science crabs,
on-beach science working together.
And at the top of the site, you see here
some of the stats that Ella mentioned.
There's 608 agents active at the moment.
There's been 2,300 hypotheses and 11,000 comments.
So you start getting a sense of what
happens when agents are collaborating,
because it can scale very quickly, essentially.
Now, before I go into some details on some specific posts,
what I want to walk through is just kind of what it looks like to, let's say, play Beach Science.
And so essentially, the key components are that at the very beginning,
you come to Beach Science and you ask your agent to install a starter pack of skills
that lets it interact with the site.
And as Clemens mentioned, the bio team has done amazing work with BioS.
And so we pre-packaged to make it easy for you to, for example, access that as a skill.
And I'll get to an example later of how effective that can be with the competition that we ran for a week.
But to just walk through the steps,
essentially, once you start, you install the Beach Science
skill.
This lets your Open Claw interact with Beach Science.
It tells it.
It gives it an API key, and it tells it how to interact.
And then we have a couple of initial skills to work with.
One is the Abre longevity skill, which Alex Dobrin put together,
which actually lets you do free queries to the Abre longevity skill, which Alex Dobrin put together, which actually lets you do free queries
to the Abre agent that anyone in the bioecosystem is likely familiar with. And this gives you
essentially a starting point for scientifically grounding any of your interactions with each
science. So it makes it really easy for you to say, hey, go research this and make sure the
science checks out, shoot it over to Abre via the Aubrey skill, and it gives you references back that you can use.
So that's kind of, let's say, the lightweight entry point to doing agentic science on Beach Science.
Step two is where you really start upping your skill set.
So this is where the BioS deep research tool is fantastic,
because it will let you spin up then that agentic system for your open claw
to really dive deep and check for novelty
and really understand, is this an idea that is worth pursuing?
And the skill lets you do that very easily.
I'll show you an example in a second.
But this lets you, let's say, play beach science
in slightly more advanced mode.
And Rafa will later be showing you taking it even
a step further, where we've been working with Lightfold, which
have been packaging up computational analysis
tools that lets you go from essentially then
the novelty checking, the exploring of the idea,
and then handing it over to computational analysis.
And so you can see here we're kind of layering
each piece of the puzzle and increasing the complexity
in doing so.
But that's why I'm trying to introduce it in a way that
kind of step by step walks you through it.
And of course, this then gives you to the multiplayer environment where once you then
have the agents starting to collaborate and post on the site, a fun note here, this was
put together when we had 42 crabs, there's now 600, just to speak to the speed of things.
And that's where we then are working on the next phase of capabilities where we will be
introducing a scoring system
that encourages your crabs to be collaborative. So essentially, once you post an idea,
you will be rewarded if your crab picks up that idea and says, hey, I'm going to do some research
to check if this actually is a good idea or not and give you some feedback. And if it looks like
a good idea, I'm going to go and do some computational analysis to take it a step further
and give you some feedback. And all of that will be embedded in a reward function of sorts. And this will also be coupled with introducing new
skills. As we mentioned, we're already working with the Lightfold skill, and we'll be introducing
more as well to really be able to make it possible to kind of build out your loadout of your AI
scientist and willing to share as much as we can around kind of validating this process so that
you don't, unless you want to craft your own
skills, we're very open to that. We also want you to be able to have an experience where you can use
what we've already streamlined. And what becomes exciting as a kind of a key element, and this
builds to, you know, the work that over the years has been done by Molecule and Bio, is that of
course the big thing that we can add here is programmatic funding and programmatic reward mechanisms in a way where
you can publish findings to on-chain artifacts and attest for provenance in such a way that it
enables programmatic funding. And the simple way to summarize it in this agentic world is, for
example, this enables us to play with ways of saying, do good science, earn inference. And so
these are things we're still experimenting with. There's still steps until we get there. But it's just to say that I think this is where there's a lot of chatter about
what it means to have agents in the on-chain economy. And I think this gives you a very clear
picture of what that looks like and why programmatic funding can unlock new behavior modes
for these agents where they can, for example, pay for data access extremely efficiently using X402.
where they can, for example, pay for data access extremely
efficiently using X402.
BioWiz already has X402.
You can hook it up so you can see funds going out
of your wallet and see signs coming back
in a very enjoyable way.
And so these are the pieces that we're bringing together,
and we are excited to do so.
So now, just to bring that to life,
I'll show you a few examples. So this is a post
by Kawaii3. And Kawaii3 is named that way because it also does some QA of the site.
And essentially, I'll show you actually the interaction. So this is going behind the scenes
a little bit. So demoing with Telegram chats can be of an odd experience. But this is the world of OpenClaw.
And so what I want to show you here
is that essentially this is an example where I installed
BioS with this agent.
And I simply asked it to come up with a research question
for each science that will result in a peptide idea
that we can test with a peptide skill.
And it says, good idea.
And then this already has the skills installed.
I can speak to that in a moment.
But then it just goes off and does it.
And basically, I just now prod it along the way.
And with OpenClaw, this is where it's getting more and more
robust in terms of doing things itself.
But occasionally, the task becomes just prodding it
to continue.
And then it does the heavy lifting with all the work
that's been gone into BioWest to actually do the science side.
And off it goes, comes up with ideas.
And I prodded some more because I was impatient.
But, you know, you can also have it go on its own pace.
And in the end, it gets to a point where it came up with a good idea and it effectively ran with it.
And in the end, it came up with a hypothesis. And I said, go ahead and post. And it posted, essentially. And this was an early example where this then appeared here
as a post. And it adds a nice image to kind of bring it to life. And then the next step,
which I think is really where the fun also begins, is that some agents came back
because this agent also ran some computational analysis.
So Amadeus came and gave some input,
but then Rafa himself came and gave some human input as well.
And so what I wanted to show this example
is it can combine then agents giving some input,
human experts like Rafa giving some input,
and then Kawai is very thankful and takes all that
and synthesizes it and comes up with what might be the next step. So this is just a very lightweight example of
these interactions starting to take place. And this space moves very quickly. So that's just to
say this was posted now about, this is an example from a couple of weeks ago, and Rafa will now
dive into the, let's say, more advanced levels that we've been working on over the past couple of weeks. But a couple more pieces that I just
want to show before I get to that. We've also then been testing the on-chain element of this.
So this is a post where it used different tools and then actually used Molecule Labs to actually
post that on-chain. So these links here actually post to
the on-chain reference for this hypothesis and shows the timeline. So this is just to give you
an idea of how these pieces are coming in together into this experience where at the other end of
this you'll have scientific agents you know being able to play beach science working with different
configurations and collaborating efficiently.
And the last piece I want to say is that already we we ran a competition not too long ago.
It was a one week competition where this was the winning entry from Professor Psyduck.
And I think what's what's actually quite exciting about this is that this is an example where someone from the community signed up,
installed all the skills,
and had it come up with an idea and post it. And Rafa is now saw this idea and actually said,
this actually looks genuinely interesting. I'll leave it to him to kind of dive into some of this,
but he's now going to manually validate, you know, this idea. And so this is, you know,
to the spirit of DCI that we've been working towards for a long time now, the pieces are
really landing in a way where people can come in.
They can, if they're curious, they can take part.
The skills, you know, upskill you almost immediately, and then you can, you know, pull the thread from there.
So I think it's really exciting to see all of these pieces landing into an experience that is fun and engaging
and makes it accessible to a broader audience.
So I think in the end, those are some of the pieces
that I wanted to show today.
I think if there is time for questions,
I'm happy to dive deeper.
But I don't want to overload you too much at the moment,
because I think the next portion where Rafa will dive
into the work we've been doing with Lightfold
will really bring to life the next steps that you'll soon
see reflected on beach science.
And just to tease it, this then starts moving towards the wet lab and all the fun robotic automation that lives there.
So I'll wrap it up there for the moment and hand it over to Rafa.
Awesome. Can you all hear me properly?
Can you all hear me properly? Perfect. Thank you, Martin.
Perfect. Thank you, Martin.
Yeah, so just to start a little bit on Martin's note, I think that the easiness right now is that we are really democratizing skills that are hard to achieve in daily life.
And everyone can now have access to these skills and you can play with them and then
really the sky is the limit so it was really fun to to play with the with the hypotheses that were
generated to talk with the agents just genuinely out of curiosity and I think that's what drives most of the work that I really enjoy doing.
So looking at what was like the tailwinds in the scientific market,
the hypotheses that have been generated,
that we've identified an underserved area where bioinformatics,
chemoinformatics, drug design and discovery are basically coming
all together and converging and this is the peptide space that is still
relatively underserved and one of the things that I was discussing with Martin
is that as a next step we wanted to take this hypothesis and really validate them down some validation gates that we manually set for
the agents. So basically a set of constraints where we can take, could be molecules, could be
peptides, could be protein binders, and really take them down a validation pipeline, and then
really start to close the loop from generating hypotheses that
right now they're becoming more and more democratized and more and more accessible to
everyone but really what the goal is now in the next minutes to to show all of you is that we
want to close this loop we've been seeing just massive amounts of models released every day that is really
difficult to keep track of. It's actually, I would say, I need like more hours in my
day to really keep track of everything that is going. But we're still generating models,
we're still generating computational predictions.
And I really want to emphasize that what we want is to take them to the lab.
So we've created a pipeline that I'm going to show in a few moments of how we can make our agents learn from themselves, improve, optimize, and basically close this loop so that in some weeks we will have end-to-end
autonomous agents running experiments calling APIs for other, could be CROs and other
wet lab experiment facilities, because they're seeing the same thing as us, companies like Jinko,
companies like Adaptive Bio, they're building this for agents now so
most of the early science will come from agents humans will be in the loop being creative designing
the experiments deciding and steering and guiding the agents on where is the best step to go but
after the the very cool paper that OpenAI and Ginko did early this year, where
they basically increased cell-free protein synthesis by 40% of yield, we start to see that
non-trivial things where humans don't have an edge anymore. Really, AI makes a better job. That's not to say that humans are not useful.
We will still be useful in this whole process,
but AI does specially things where it requires
multiple variables, multiple things to optimize.
As we know, they can hold much more information.
So as a next step, I'm gonna show, let me see.
Yeah, I can show my screen so this is our feedback loop
from hypothesis to learning system what we're building right now with Martin is basically
these two steps using BIOS plus computational validation gates what we
call it's eight gates we're building this as a backend for our peptide agent
this is all done computationally BIOS comes comes up with novel hypotheses
novel peptide designs take something like for instance, to give an example here, we took,
you know, as we all know, GLP-1s are really, really talked about in the space recently. So
we take a GLP-1 backbone, BIOS generated three different candidates. With one, we wanted to have
something like more protolytic stability, meaning that it gets degraded less.
One, it was, hey, I want something, but just very close to, for instance, semaglutide or other like of the OG GLP ones that are in the market.
And the third one is like, go crazy, go creative and without any borders and constraints, design something.
So we came up with three different candidates.
We run them through our eight computational gates.
I'm gonna show you a few results.
And then the next step, and this is like really important
that I went to the point that I want to drive across
is that this is what we call basically
the wet lab experiments.
First, we're going to use a company that gives us a score for feasibility, whether we can
synthesize our peptides or not, at what yield, at what cost. And this is super important because we can find a peptide
that it's really new and novel
and it looks, you know,
like extremely just very high potential.
But if it cannot be synthesized,
then we cannot do anything.
So once this is done,
this is going to be later on
also an API that is going to be
talking to our agent.
The two are going to have
just multitude of conversations and is going to be talking to our agent. The two are going to have just multitude of conversations
and are going to be exploring the space of possibilities
from two ways.
What can be synthesized?
What can be done like from drug design
or a peptide designed discovery?
And then find ways where we can tune this.
Later on, we're gonna synthesize the peptide and then go to the wet lab for what we call
gate 9.
Once we close the loop, everything goes back into the agent here, data feeds back.
So the agent gets smarter each cycle.
Just to give some metrics here, there's a paper from both papers from early this year,
is that they found out that there's only 30% of hypothesis accuracy without wet lab,
meaning that it required between three or five cycles to really optimize a peptide or a small
molecule and to find something that makes sense. And so far, there's a paper as well called Bits to Binders,
where we have the best computational tools
against the experimental outcome.
And most of them, these computational predictions
find that they're basically near random.
So if we design something from scratch
and aim it to be just something novel,
we're going to have a hard time.
So what we want is to really have a self-learning loop.
And this is what we are building right now.
And it's going to be done in a few weeks.
And once we have that, we can scale it to other things.
Could be also small molecule, could be other binders.
But once we have closed the loop,
we really can move pretty fast.
So now I'm going to show our candidate.
Yeah.
So this is what came out of the light fault agent
that basically handles the first eight gates.
Actually, let me show you this one, sorry.
Yeah, so this is what comes out in the backend
of Rosalind that is the UX of LightFold.
This is a comparison table of the three GLP-1 analogs
that we designed. One was with for enhanced
protolytic stability to be degraded less in the system. Once the other one had enhanced selectivity
for the ligand and the other one was just a crazy design from BIOS. We also have the
modifications from semaglutide which is the backbone that we decided to have.
It has the key changes, predicted half-life, molecular weight, the primary advantage, the risk, etc.
And then once we have that, we go through the process of validation from our eight gates.
On the left, we can also here visualize the visualize the protein using bolts 2 and the 3d
structure we have the confidence as well the protein to ligand binding strength we have the
sequence everything is here so then it goes to a pipeline that we've been refining constantly with Martin it's
it's actually eight gates and I want to show them as well here yeah so I want to
show it here so what we do is that first bios and the the user, we have our target ID and we design our candidate.
Then we use LightFold to get the structure, that's our first gate, the confirmation using
molecular dynamics, the binding pose, the affinity, then the molecular dynamic stability, that is actually super important,
and I see a lot of models that are not using it,
proteolytic stability, psychochemical properties, and selectivity and toxicity.
So these are our gates.
They're all actually a work in progress.
When we see a new model or a new paper or a new agent that comes with something new,
we can actually test it against our gates and we can improve it constantly.
So we're not really, let's say, close to this is the way it has to be done, where we can
move things around very easily.
And I think that's one of the strengths of our setup.
Then we go to the synthesis and feasibility.
Right now it's something that it's done manual,
but we're working to have done in rapid synthesis.
We have our wet lab gates, the heat confirmation,
and then animal models.
And then we start for the lead optimization.
So this is a little bit of the path we're working on.
And now, back to the report.
So I want to start with PROSTAB1, which is our GLP-1 agonist analog, peptide analog, that has improved proctolytic stability.
And as you can see, there were some of the gates that were blocked because they were not available, mainly the molecular dynamics.
But good news, they're available since this week.
So now we can run everything.
So as you can see, this actually had a few conditional passes, but most of the things
were passed.
The gate nine is blocked because that's what lab required.
And then we did that with all our candidates as a test.
Now we're finishing everything.
I'm gonna stop sharing.
We're finishing everything.
We are creating what we call our calibration curve,
meaning that we can send our candidates soon
for synthesis and feasibility.
We can get a score and then we can send it
to our first wet lab experiments.
This is all going to be
done in the next weeks, max one, two months, depending on how fast we move. But yeah, this is
actually something really cool, really exciting. We're finally closing the loop. We're going to have
agents calling to other agents and what we, many years we dreamt of finally we're gonna have agents
making science by themselves sending the experiments getting data back and then optimizing
closing the loop so we're very excited about where this is going we're building every day
yeah no sleep and cranking hours and hours. But yeah, happy to answer any questions or any specific things.
Just wanted to give you a bit of an overview of the agent, of the gates, of how the candidates run through our pipeline, how we're closing the validation loop.
And yeah.
Cool. So yeah, we're opening the floor to any questions if you do have questions send them through and we can answer them I do know that apparently Twitter is down
so we're chatting to you guys on on LinkedIn and surrounds but maybe I can start off while
while people sort of get themselves organized.
Who do you envision sort of being the power user?
Like, who's this really, really going to sort of slip into their daily workflow?
Obviously, people like Rafa who have startup experience,
drug development experience.
It works really well, but obviously Rafa works for our company,
so it's pretty obvious why he's using the project um but who do you envision this helping a lot
is that a question for me maybe martin maybe martin yeah yeah uh yeah so i think i think in
the end um a note on rafa's uh i think uh one point I wanted to make there is while he does work for Molecule, the enthusiasm to engaging with Kwai3 was genuine on Discord.
In the sense of like, it's, I think it's a worthwhile note to say everyone on the team, even though it is intense to work in this space, is many of them are saying it's the most fun they've had in ages.
And so I think this is like, I think it speaks to the it's it's a wild space because it's moving so fast but to surf that wave and be
part of that journey is like extremely motivating so um just to just to yeah a slight uh addition
on that point um in terms of who will use it i think what's really exciting is that um as rafa
now demonstrated we're going very deep on making sure that this is done rigorously.
But then what we're planning to do is flow that back up so that it becomes accessible to a broader audience.
So what I'm excited by, for example, this gating concept where Rafa has been working on where those gates should sit.
What now becomes fun is we can then programatize that in a sense so that people who come in, we can find a pathway for they could articulate, you know, maybe there is a domain of science that somebody who's not doesn't have all the experience that Rafa has, but they want to start exploring.
This gives them a pathway to be able to spin up one of these agents and leverage that knowledge in a scalable fashion.
So I think what we'll see is essentially a development where these tools will be very powerful in the hands of those with a lot of expertise,
but you now have a way where they can kind of make a pathway through the dense forest of new tooling and leave a clearer path for those that want to get involved.
And these tools also can help explain every step as you go.
So I guess my answer is this broadens the audience that
can participate.
And there will be likely those with the most expertise
who kind of can lead the way, but then
can give ways for people to pursue answers to questions
that they might have but not yet have the kind of expertise
to solve for.
This gives them the tools to do so. And I mean, there was also
the recent example with many caveats, but the man who found a vaccine for his dog,
like these are the types of moments where now we're looking to bring the rigorous scientific
capabilities that expand anyone with a mission where they want to make progress on a new treatment or so on, that you give them a pathway, an agency to do so.
But to everything that Rafa just explained, with the rigor behind that, we really validated that this works, essentially.
And that's the journey we're on right now.
I would just like to add to this, Martin.
Martin, I think the vision that gets me excited is that potentially anyone that has a scientific
enthusiasm and knowledge can work from anywhere on potential breakthroughs, right? Like the
software developers working, traveling the world or working from home anywhere where there's a
laptop that can contribute to the
development of tools in the digital realm.
And I think this is kind of where we're heading now with the scientific breakthroughs as well.
All of these tools becoming available for kind of the information gathering and sense
making.
And then obviously something that is on everybody's mind is, where does the rubber hit the road?
Where do we actually bring things into the real world
and into an experimental setting where we can validate
a lot of these hypotheses or invalidate them?
And I think while there's still a lot of progress to be made,
we see the first cloud labs that can do certain experiments
very, very rapidly. And I think
with the strides in robotics that are being made, it's not something in the far distant future
where an agent just sending into a cloud lab a survey that they would like to conduct on a
certain molecule and they get back the data in not a month's time but in days or weeks time
and and then basically we're in a in an iterative loop where data aggregation will speed up rather
quickly and the costs around these because of the quantities that will be enabled with this will
obviously also drop right um and i think this is something that i'm just extremely excited about
to see what's coming coming out of this but i see we actually have a lot of questions coming up in the comments.
Ella, would you like to walk us through them?
Yes, yes, I can. I just wanted to also just actually add a note to the discussion before
we jumped into the comments, which was that I really like the sort of emphasis that the team has put on play in the science.
I think kind of speaking to what you're saying, Martin, is like this should be for anyone.
And I think the emphasis on play and having fun with science at this early ideation stage is a really, really equalizing factor.
The fact that someone can like log in and there's a cute little graphic that sort of explains these scientific hypotheses
that are all generated by sort of the beach science team.
I think it's a really great way to open up that door
to sort of a mass audience in science
while then also still having the pipeline
and the follow through to take it somewhere more serious to a wet lab validation and that sort of thing. I really like that aspect of
it. But anyway, yes, we've got questions. So I think the first question we have is from Wes,
which says, how do you think about IP when disclosing or launching a potential project?
Why would a company want to plug in an idea into this platform versus build it on their own?
company want to plug in an idea into this platform versus build it on their own?
So I might share a couple of thoughts on this and then Martin, if you want to add something to it
for free. I think at the end what we'll see is there these are all Legos that people will stick
together into their own needs and Beach Science itself I think is just one Lego out of multiple that we're using.
And so one could decide to use different pieces here
without deciding to publish it on beach science,
but rather decide to actually just put it
in a secure data world and basically hash it
and timestamp it and show to the world
that they're the first ones that created whatever ideas
is basically stored in the secure
data world. And that's something that Molecule Labs is working very fervently on to make that
a reality as well, so that both of these things can coexist. A world where people are interacting
around any kind of idea and data that is out in the open that they can discuss and build new
knowledge upon. But then also on the other side, if somebody comes onto like a breakthrough using BIOS,
using some of these other tools,
they can basically store it somewhere
where they can continue working on it,
which is not completely visible to the world from the get-go.
And I think we're fully aware that at the end,
new breakthroughs, especially if they need huge investments,
so they need to be protected. So that at the end, onces especially if they um they need huge investments so they need to be
protected um so that at the end once these investments pay off people um can actually
you know make something out of it um for these other systems other legos need to exist and i
think um yeah people will use them to to come up for their use cases and and and like use them for what they are, for what they need.
But yeah.
Anybody else would like to add to that?
I could just second on the technical side
that when it comes to IP with the Molecule Lab skill,
for example, it gives the agent a way
to answer the question of where do I
store my scientific findings relative to that choice of what will be public and what will be private. And what's neat there
is that Molecule Labs, that skill then is, you know, for a programmatic world, from a funding
perspective, they're, you know, they're building the full, the stack goes all the way there for
how to translate even into the patent process and so on.
So it's just to say that there's optionality there
where you can choose what you make public
and what you choose, as Clemens said,
what you choose to keep private relative to the,
let's say traditional IP pathways.
I think there's a broader question, of course, on this,
which is with the acceleration events of AI
and these new collaboration platforms,
you introduce new questions of what IP means as well.
Now that's a bigger philosophical discussion that we can maybe have in a follow-on live stream,
but it's just to say that we're building in a way that is modular and as Carmen said,
it's these building blocks so that you can effectively explore what is possible today
and both paths become possible essentially for a more public-based you know a world that
moves at AI pace and for a world where the let's say more traditional IP considerations are
are also taken into account so I think this modularity allows you to to go there and I
think another piece that I should mention because I saw just not to jump ahead to another question
but it's relevant here is you know in terms of the open source side of this so I think another piece that I should mention, because I saw just not to jump ahead to another question, but it's relevant here, is in terms of the open source side of this.
So I think when you look at OpenClaw, for example, as an agent harness and many other
agent harnesses that are out there, they are open source.
And the push is really towards open source here, because it's the fastest way to iterate
and develop with a community on these platforms. There are so many open source skill libraries coming out as well.
So that's just to say that there is an open source ethos that I think is kind of foundational to a lot of these building blocks as well,
that then gives you that optionality.
Like you've probably seen the Mac minis, some of you, if you've looked at the OpenClaw universe,
the Mac minis, some of you, if you've looked at the OpenClaw universe, you can, you know, install
this on your own piece of hardware and really control exactly what information you share and
how with some considerations of what models you're using and so on. But this is, it really is very
modular all the way down and gives you as the scientist the choice and optionality of how you
participate. So that's very much kind of what we're trying to help facilitate and making it easy as well.
May I actually add to this?
And that's maybe more of a meta observation
to the kind of open source versus closed source.
I have a kind of a feeling that kind of open source
is eating closed source nowadays,
just with the advent of these ai tools in the development
especially in the development realm we are observing this where code is now so easily
generatable and so widely shared that kind of i think a lot of sas companies right now are
struggling because their mode is disappearing because um the code that they've generated in
the past because these tools are so widely available now um the code that they have generated in the past and that um that kind of made up their their um their products
are now also becoming so widely available so that other people can just replicate what is there very
easily right and so we are still far away from that happening in the in the biopharma space but
uh if the trajectory
that we're seeing there when it comes to tooling
for data analysis, and then hopefully also the real world,
the robotics continue on that trajectory,
also the replication in the realm of biopharma
might actually become easier and easier,
which would be a super exciting development.
Cool, thank you so much, guys. our development. Cool.
Thank you so much, guys.
Okay.
Moving on to our next question from Kelly.
A missing piece seems to be the upstream agent-driven novel molecule candidates to wet lab.
The real-world testing is the piece that I'm wondering about.
How does this work?
This seems to be a huge bottleneck in time and cost.
I'm actually going to say we should remove the question from the screen because we can no longer see
Martin or Rafael.
But yes, thank you Kelly for the question.
I can reply to some of the points.
I think it's a great question
and it's what we're wondering right now
and what we're trying to solve.
So we are prioritizing companies and CROs that have fully autonomous and automatic, basically
high throughput screening.
And especially like the thing here is that by us having also a launchpad in BioProtocol,
we're able to raise smaller amounts that would pay for the smaller experiment that is able
to de-risk an asset and that is something that as a startup is difficult because you're
you're actually incentivized to raise more money and then you do more experiments but here we're
just incentivized to do the smallest experiment
that we can that would allow us to know whether we actually continue further or we don't and
we're also incentivized by actually not continuing further because all this negative data gives us so
much to work with rather than just looking for the good data. So we're actually not worried about
failing an experiment because it makes us just like stronger and that like data
feeds back into our agents. So I mean now they're just CROs like Nginco and Adaptive where you can
run these experiments for like
a few thousands of dollars and you don't really need to do much of the work.
You just simply need to send the experiments.
So I think that it's coming from both sides.
Companies are realizing this, are creating the gates and the infrastructure so that the
agents can use and can send these experiments.
So, and we're simply, I mean, simply, it's a simple world, but we are closing that loop in a way where
we come up with, in the first step, our agent on BIOS and all the way to the funding, programmatic
funding with BioProtocol. So we have the full stack and that is something where we basically can orchestrate these agents,
can orchestrate this and coordinate these collaborations in a way where maybe it's actually difficult.
And yeah, I think this is what is positioned, how we're positioning ourselves and the big advantage that we have.
And also other agents can come
and collaborate in the ecosystem so it's not cannibalizing it's actually collaborating and
coordinating i think it i was gonna say it's also like worthwhile to mention so we kind of are
trialing this whole pipeline ourselves um and we're going to be sending it's also worthwhile to mention, so we kind of are trialing this whole pipeline ourselves,
and we're going to be sending off experiments to a wet lab.
And so there is a Beach Science Twitter account that we'll be sharing all of those updates as well.
So you can kind of see how we're figuring out
and sort of testing this mechanism to get something to a wet lab
as we go along, obviously prioritizing
sort of more cheap and affordable experiments.
I think at this stage, it doesn't make sense to be, you know, scoping out projects that
are 20, 30, 40, 50K, just while we figure out the model.
But yes, we will be keeping you all updated on that.
Okay, then we have maybe what's a little bit more of a broader philosophical question
from Mudatha, which is what are the consequences
on the business model of pharma companies
and drug developers with all of this sort of progress
around AI, which I think is probably a matter of opinion,
but maybe you guys can speak to where you think this fits in.
I can share maybe like a small take on it, but what my hopes are, let me put it this
way.
I think we will see across the early stages of the drug development cycle, not necessarily
the later stages, but the early stages of the drug development cycle, I hope we will
see a lot more startups, experimentation and biotechs coming up
and basically um making their way up to the large gates which for novel uh compounds obviously is
then the inhuman trials um if you're if you're in biopharma and that is then becoming quite costly
but hopefully we'll also see uh cost reductions in there. But with all these tooling, I think anything along the way up until there,
we will see an explosion of ideas and of startup approaches,
which I think is a great thing because that will just lead to cheaper costs
around a lot of these experiments.
It will lead to more personnel that is building tooling
that can then be more widely used from the shovels
to actually people applying them.
And so I think this is actually a little bit
of a Renaissance for the biotech, let's say arena,
because this kind of translational gap
in between what was happening between universities
and what happened in the pharmaceutical space usually was kind of the valley of death. A lot of things were dying there and we hope that this
will actually kind of like enlighten the fire again in this valley and bring more things
into the pharma space. Now I do think that pharmaceutical companies still have a role to play, especially as trials in humans are just very, very costly
and just take very, very long.
And so I think they will just,
a lot of them will continue doubling down
on what they've been doing in the past.
And that is seeing what finds its way up the stack
through the process up until the in human trials,
phase one to phase three and then they
will partner or potentially scoop these things up make sure that the last phases go through and then
help with the um with the commercialization which is where they're extremely strong and so i can see
this still being a huge huge role just as large investors and typical vcs and typical startup
companies coming in when it's like a billion
dollars but not yet a stock listed company supplying the capital and bringing this into
kind of the the next sphere and I think that will just be continuing the case for pharmaceutical
companies but like I think there's just there will be just a lot more fire in between what's coming out of base research universities and then what goes into
a human, a lot more experimentation. And I think that at the end, it's just going to
benefit what comes out on the other side.
Yeah. And maybe one thing I would add that I think is quite exciting as well is that in terms of that kind of upstream discovery process, these tools, on the one hand, if it increases the success rate of those later stages, that's fantastic, because that would mean that more success to deliver treatments and cures.
And also to the theme that we touched on at the beginning in terms of increasing the usability of these earlier stage tools, it could give patient communities greater agency to kind of start that process
in such a way that the hope would be that, for example, communities that might not have
been necessarily highest on the radar of pharma, maybe it gives them a way to actually go further
down that discovery pipeline.
And so that, you know, it gives an agency to get on that
radar more effectively. Without getting into all the nuances of how pharma is structured today,
I think those are two of the things I would be optimistic about is that you expand who can
actually take that journey and has agency to address the needs of their communities and,
has agency to address the needs of their communities and make that a tangible process
for a greater audience. And I think, I guess this is the optimist in me who's speaking,
but I think that could actually lead to a very big improvement for the types of treatments that
are developed in a broader range going further. And may I just actually add to that?
We were just speaking about novel compounds, right?
There's a huge field of opportunity also
in repurposing and analogs.
And so I think these tools will allow a lot to be done
in this arena.
And so I have been watching, for example,
one of the DA dows in our ecosystem
curatopia um quite interested because they're looking very strongly at repurposing uh existing
molecules for the rare diseases uh especially in children uh with rare genetic disorders and so
i think this will be something extremely like uh interesting to see like, okay, we actually know
how these things already interact with humans,
but now we're realizing we can use them
for a lot of different other therapeutic goals.
Yeah.
Awesome.
Thank you so much, guys.
So I think, what, we're almost at the end of the hour.
We've got another five minutes.
If anyone wants to shoot one last question,
we just gotta thank you.
Okay, but I think if that, is there a Discord channel?
No, there's not a Discord channel.
You could join the BioProtocol Discord channel
or the Molecule channel.
We're both posting about it.
We're both involved in it.
But probably the best place to stay up to date
is on the Science Beach Twitter page,
which is at science beach underscore underscore.
You'll be able to notice it with all the sort of
crabby imagery and beach related puns.
But yeah, thank you so much for joining us,
for spending a little time chatting about AI agents
and how science might sort of, how science is changing
because tech is changing. And I think there is a lot to be very excited about.
I'm sorry, I'm just reading, where is the link? I will in the comments post it.
Anyway, everyone, have a lovely evening
it's been great to see you
I'm going to keep the live stream going while I just
post our discord link
cool yeah and
if you're interested again as I
plugged at the start in running your
own agent and you're already in
our ecosystem feel free to join the
buyer protocol at discord and reach out
to us we have an infra to actually very easily spin these things up with pre-installed skills
so you can be up and running and trying out beach signs and other things rather quickly
um and we would love to get your feedback on these things
absolutely rafa martin any final words
getting back to it very excited
same here same here okay cool thank you so much everyone for joining have a lovely evening or
morning or afternoon wherever you are um we are around, we're building, message us, chat to us, launch an agent.
Yeah, thank you for your time. Thanks a lot. Bye.
