WEBVTT

NOTE Sentence-level transcript of https://www.youtube.com/watch?v=adfw3xJBx1o

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/adfw3xJBx1o.html. A cue ends where the next begins, or 2 s after its last word.

s1
00:00:01.309 --> 00:00:03.309
[music]

s2
00:00:12.840 --> 00:00:15.480
Uh, we are Well, that was loud.

s3
00:00:15.480 --> 00:00:17.960
Uh, we are Felipe and Francesco.

s4
00:00:17.960 --> 00:00:25.960
Uh, we come to talk to you about the compound value of building Monzo sync house agents using LLMs.

s5
00:00:25.960 --> 00:00:28.800
Um, so what is Monzo?

s6
00:00:28.800 --> 00:00:30.400
We are a bank.

s7
00:00:30.400 --> 00:00:33.400
Digital bank trying to make money work for everyone.

s8
00:00:33.400 --> 00:00:45.680
Uh, to give you a size of the problem for coding agents, uh, we have our teams of engineers maintaining over 3,000 back-end services, around 13,000 um, data warehouse models,

s9
00:00:45.680 --> 00:00:49.360
and dozens and dozens of ML uh, predictors.

s10
00:00:49.360 --> 00:00:59.240
So, we do all sorts of things like deciding on credit, deciding on financial crime, and doing very complex use cases both for the app, the back-end, and data.

s11
00:00:59.240 --> 00:01:11.400
Um So, I'm pretty sure by even by today, but outside of the conference, you have heard a lot of uh, success cases of companies adopting AI agents for coding.

s12
00:01:11.400 --> 00:01:22.040
Uh, in fact, uh, there was a conference a couple of weeks ago here in New York, and this article came out talking about the state the state of the tech industry in 2026.

s13
00:01:22.040 --> 00:01:32.920
Uh, and there is this hopefully jokingly quote from X saying that if you're not building your own AI coding harness, are you even a serious tech company?

s14
00:01:32.920 --> 00:01:35.360
So, some food for thought, right?

s15
00:01:35.360 --> 00:01:42.600
Uh, then inside of Monzo in January 20 26, this a Slack channel went viral.

s16
00:01:42.600 --> 00:01:43.840
Agent chip.

s17
00:01:43.840 --> 00:01:46.720
And hopefully I can Oh, no, what happened?

s18
00:01:46.720 --> 00:01:47.600
You cannot do this.

s19
00:01:47.600 --> 00:01:48.960
It's all right.

s20
00:01:48.960 --> 00:01:51.960
So, hopefully I can bring your attention to this little emoji here.

s21
00:01:51.960 --> 00:01:52.920
Really cute.

s22
00:01:52.920 --> 00:01:53.640
We love it.

s23
00:01:53.640 --> 00:01:57.920
Uh, this became really, really viral inside of the company for a really good reason.

s24
00:01:57.920 --> 00:02:08.960
You can give it uh, any prompt directly by just reacting with the little emoji to the message and it will pick it up as a, you know, prompt.

s25
00:02:08.960 --> 00:02:17.360
Uh, so that evolved, we double down, and that became an always available teammate inside our Slack and whole environment.

s26
00:02:17.360 --> 00:02:21.080
Uh, to give you a few examples, it does feel like a coworker.

s27
00:02:21.080 --> 00:02:21.920
Uh, it's just there.

s28
00:02:21.920 --> 00:02:33.000
You can always at Agent Chip and get it to do something for you, answer a question, find out something about some architecture, or just have a chat with you, but I don't I don't think he likes it so much.

s29
00:02:33.000 --> 00:02:36.080
Uh, it can also do code reviews.

s30
00:02:36.080 --> 00:02:43.680
So, like a very senior, uh, engineer, you will just tell you it looks good to me and go on your merry way.

s31
00:02:43.680 --> 00:02:46.880
And you can also assign, uh, linear tasks, as well.

s32
00:02:46.880 --> 00:02:50.320
So, you can give it tickets and it will just get into it.

s33
00:02:50.320 --> 00:02:54.600
Um, so all of this is good and well.

s34
00:02:54.600 --> 00:03:10.239
Uh, forward into today, Agent Chip authors over 10% of all merge PRs across tech, so that's data, uh, machine learning, that's, uh, back in engineering, app, etc. Also runs more than 1,800,

s35
00:03:10.239 --> 00:03:14.519
uh, tasks per day, which is a lot.

s36
00:03:14.519 --> 00:03:19.519
Um, and it's being actively, uh, starting to be adopted by business users as well.

s37
00:03:19.519 --> 00:03:28.680
So, it's not just coding and tech, it's also trying to answer business questions, writing queries, or self-serving anything that a business user may need.

s38
00:03:29.200 --> 00:03:37.519
So, again, you've heard many success stories about companies adopting AI coding agents, but so what?

s39
00:03:37.519 --> 00:03:40.519
Why are we What are we trying to share today?

s40
00:03:40.519 --> 00:03:46.959
Are we saying that we want to brag that we built Claude Tag 6 months before Anthropic released it?

s41
00:03:46.959 --> 00:03:48.239
We could be.

s42
00:03:48.239 --> 00:03:51.200
That's not the focus of today's talk.

s43
00:03:51.200 --> 00:04:00.520
Um, we want to share the secret sauce of what makes Agent Ship a successful product and it's LLMs.

s44
00:04:00.520 --> 00:04:06.240
But actually LLMs in this case stands for lots of little multipliers.

s45
00:04:06.640 --> 00:04:15.680
We believe that what enabled us to build Agent Ship so quickly and so invaluable way is a very strong platform at Monzo.

s46
00:04:15.680 --> 00:04:22.360
So when we say that this went viral and you could write a prompt, react, and profit.

s47
00:04:22.360 --> 00:04:26.680
Actually this comes from a number of capabilities we already had.

s48
00:04:26.680 --> 00:04:41.720
So alerting, monitoring, integrations with third parties such as the ones Francesco mentioned, a staging environment that allows us to test easily, queuing abstractions, easy deployment, database abstractions, and a number of other things

s49
00:04:41.720 --> 00:04:43.840
I could be mentioning here.

s50
00:04:43.840 --> 00:04:51.560
So first and foremost what we have in our platform is what allowed Agent Ship to be successful.

s51
00:04:51.560 --> 00:04:53.520
But this does not stop there.

s52
00:04:53.520 --> 00:05:04.720
We saw even before Agent Ship as a remote agent existed, we started to invest in supporting AI coding agents even in local environments.

s53
00:05:04.720 --> 00:05:14.680
So the one of the first things we did was develop an MCP gateway that allowed easy access to a set of knowledge sources and internal tools and some of the

s54
00:05:14.680 --> 00:05:16.720
services that Francesco mentioned.

s55
00:05:16.720 --> 00:05:20.280
So this is roughly how the architecture looks like.

s56
00:05:20.280 --> 00:05:25.600
We have the things we want to connect as the internal services and knowledge sources.

s57
00:05:25.600 --> 00:05:38.000
And we put that behind the MCP server that we built to get the deterministic behavior choosing what are the things we want to expose to the agents and then we manage also authorization,

s58
00:05:38.000 --> 00:05:41.800
authentication for the number of different agents.

s59
00:05:41.800 --> 00:05:58.600
Again, this started in the world of local coding agents that wanted to connect and by by now probably everyone has heard in multiple sessions that the more context and the more data we are able to give safely to our agents the better.

s60
00:05:59.480 --> 00:06:10.320
After that, we built the agent orchestration system, which is exactly what powers Agent Ship behind the scenes and makes it work in a sandbox and with very strict tools and permissions.

s61
00:06:10.320 --> 00:06:23.960
So, if we go back to our diagram, the whole architecture for the MCP server basically condenses into this box and this is where we connect the rest of the Agent Ship

s62
00:06:23.960 --> 00:06:24.680
architecture.

s63
00:06:24.680 --> 00:06:37.120
So, we have remote agents running inside a sandbox in an isolated namespace and then the orchestration system triggers and then reads and gets this information from from that.

s64
00:06:37.120 --> 00:06:49.919
The main idea here is and again, this is coming across in a multiple of multiple providers and labs and so on is that we want to separate this from the rest of our infrastructure

s65
00:06:49.919 --> 00:06:50.800
in production.

s66
00:06:50.800 --> 00:07:00.720
Again, we are a bank, we live in a regulated environment, so we want to make sure that the non-determinism from the agents is controlled.

s67
00:07:00.720 --> 00:07:06.120
But at the same time, we want to have an environment where we can let it run loose and develop.

s68
00:07:06.120 --> 00:07:11.840
And again, if we explode this to give you a bit more under the hood visibility of what is going on.

s69
00:07:11.840 --> 00:07:20.240
So, in this isolated namespace, multiple pods come up with sandboxes that the agents are running on.

s70
00:07:20.240 --> 00:07:25.919
The agent image is where we define the exact tools the agents will have as as access.

s71
00:07:25.919 --> 00:07:35.560
So, things like the environment, the CLIs and the authentication and the authorization that again, we get from the gateway itself.

s72
00:07:35.560 --> 00:07:41.720
The proxy is what allows us to connect to both the exterior world.

s73
00:07:41.720 --> 00:07:48.080
So, that's how for instance Agent Chip is able to connect to our code, to our repositories, and work with that.

s74
00:07:48.080 --> 00:07:54.440
Um but, of course, we only get the egress that we choose to do choose to give it.

s75
00:07:54.440 --> 00:07:56.190
Um so, that is how we make sure that

s76
00:07:56.190 --> 00:07:56.440
[snorts]

s77
00:07:56.440 --> 00:08:02.600
it's very controlled and deliberately giving access to the things we want it to be able to see.

s78
00:08:02.600 --> 00:08:08.080
And finally, the little uh blue-green box there is where the agent itself is running.

s79
00:08:08.080 --> 00:08:17.080
So, that box inside the box inside the box is where the nondeterminism happens, and it's where the agent can perform everything that it wants.

s80
00:08:17.080 --> 00:08:31.280
So, this is how the whole diagram looks like when we are able to give our MCP server access to everything that we care about for the agent, we are able to safely give it access to our data,

s81
00:08:31.280 --> 00:08:37.240
and the agent orchestration system handles everything that's around it.

s82
00:08:37.240 --> 00:08:44.839
With these as building blocks, any use cases that come after it are much, much easier to be built on top of.

s83
00:08:44.839 --> 00:08:57.560
So, as Francesco mentioned, this we can connect Slack to, and then we can connect GitHub, and then we connect Linear, and then I just use a different color to show this is the more than 3,000

s84
00:08:57.560 --> 00:09:01.120
services that we can be connecting to to this as a building block.

s85
00:09:01.120 --> 00:09:09.960
So, in the same fashion that we had multiple things in our platform that enabled us to build Agent Chip and get value out of it.

s86
00:09:09.960 --> 00:09:19.040
Now that Agent Chip is also a building block, the multiple parts of our bank can also leverage it to create other use cases.

s87
00:09:19.040 --> 00:09:29.400
So, we start to see this building blocks enabled us to not only get more um product features internally, but to drive concrete business outcomes.

s88
00:09:29.400 --> 00:09:40.560
Um we observed a higher breadth and volume of tasks that engineers are able to run, as you you expect, because then we don't need people to be babysitting their agents all the time.

s89
00:09:40.560 --> 00:09:47.920
They can use their coding assistant for um specific tasks, but they can also delegate to remote agents.

s90
00:09:47.920 --> 00:09:50.520
We can automate routine processes much better.

s91
00:09:50.520 --> 00:09:57.760
So, that allows humans to focus on the higher value problems where the ambiguity lies particularly.

s92
00:09:57.760 --> 00:10:00.640
Some examples of these that we are able to do.

s93
00:10:00.640 --> 00:10:08.000
Um this is Agent Chip working in incident mode and leveraging the MCP to get very relevant data.

s94
00:10:08.000 --> 00:10:23.600
So, what happens here is whenever we identify and trigger a new incident, Agent Chip joins the channel and is able to start an investigation, understanding logs, understanding traces, being able to connect to that exact service, and of course, the code itself

s95
00:10:23.600 --> 00:10:27.880
to see what's going on and to help the human with a very high signal.

s96
00:10:27.880 --> 00:10:37.960
We don't want the agent to be saying a lot of things during an incident and testing hypotheses like we want it to maintain high signal, and we give the humans also control

s97
00:10:37.960 --> 00:10:41.960
of how much Agent Chip will be able to support there.

s98
00:10:41.960 --> 00:10:50.200
We can continue to use it on Slack, but not only by triggering it by mentioning it, uh but also in what we call ask channels.

s99
00:10:50.200 --> 00:10:54.040
It's basically squad-to-squad interaction inside Mondzo.

s100
00:10:54.040 --> 00:11:04.520
A squad needs to have some context on a back-end service, on a product development, on the road map, and Agent Chip can also, since it can connect to our knowledge sources,

s101
00:11:04.520 --> 00:11:10.640
it's able to draft a response for the human in the loop to get as reference and support there.

s102
00:11:10.640 --> 00:11:17.240
Maybe the response will be enough, and Agent Chip can just solve that ask ticket by itself.

s103
00:11:17.240 --> 00:11:25.320
Sometimes it will ask for more context, and the human in the loop can be tagged um and come in to support more.

s104
00:11:25.320 --> 00:11:36.720
Um something that we're doing very recently that's been uh creating a lot of value is also triaging data quality alerts and checking for back-end changes that can create problems.

s105
00:11:36.720 --> 00:11:45.120
So, basically we call it Agent Chip on Call and it observes the failures that we have and triggers an investigation automatically.

s106
00:11:45.120 --> 00:11:47.880
This is the reactive version of it.

s107
00:11:47.880 --> 00:11:51.280
But, what if the errors never happened in the first place?

s108
00:11:51.280 --> 00:12:07.320
We created based on this the proactive version that monitors back-end changes and identifies which are the potential outcomes that that can have for our data even before the data is processed in the next batch execution.

s109
00:12:07.320 --> 00:12:20.040
So, again, all of these are examples of how our technology team is using it, but even better than that, we've observed a lower barrier for non-specialists in the company to contribute to our systems.

s110
00:12:20.040 --> 00:12:32.320
With a very quick entry point for Agent Chip on Slack, non-specialists can, based on product feedback, based on ideas, based on discussions they are having, just call Agent Chip,

s111
00:12:32.320 --> 00:12:43.000
get a draft that works, that is already tested in staging, and then, and only then, involve an engineer to make sure we get the changes reviewed and deployed properly.

s112
00:12:43.000 --> 00:12:57.240
So, this doesn't change the rigor to which we develop software at Monzo, but this does allow a number of non-specialists to also start contributing more closely to the team.

s113
00:12:57.600 --> 00:13:01.720
You can be thinking now, the market is moving fast.

s114
00:13:01.720 --> 00:13:13.000
We are hearing a lot of exciting developments and new products and new features coming from multiple companies here in the conference and otherwise.

s115
00:13:13.000 --> 00:13:20.480
What makes us believe that Agent Chip continues to be a valuable thing at this point in 2026 and moving forward?

s116
00:13:20.480 --> 00:13:31.600
Um As I said and I like to to reiterate, Agent Chip being composable, it is now a new building block that we have in our platform.

s117
00:13:31.600 --> 00:13:38.560
So, it very quickly enables new use cases for agentic implementation at Monzo.

s118
00:13:38.560 --> 00:13:46.400
Now that we can connect it very safely with all our security requirements and all our back end and platform requirements at the same time.

s119
00:13:46.400 --> 00:13:50.000
Our platform is plat provider agnostic by definition.

s120
00:13:50.000 --> 00:13:59.600
So, if we come back to the diagram, the little blue/green agent box there can be very easily swapped by any for any model.

s121
00:13:59.600 --> 00:14:04.480
We can dynamically choose which model best fits the purposes that we have.

s122
00:14:04.480 --> 00:14:16.120
So, we are not specifically tied to any provider by definition and that gives us flexibility to understand which ones will fit Monzo's needs the best.

s123
00:14:16.120 --> 00:14:21.360
Um and finally, we are continued to move to cover a bigger development window.

s124
00:14:21.360 --> 00:14:33.760
So, if we even if we stop development today, this would already be valuable, but we can cover a much broader sense in terms of we draft changes in the back end, we draft changes in the app.

s125
00:14:33.760 --> 00:14:38.360
We can test these changes in a staging environment, cover the data tests.

s126
00:14:38.360 --> 00:14:38.880
Who knows?

s127
00:14:38.880 --> 00:14:41.320
Maybe even predictors on the other side.

s128
00:14:41.320 --> 00:14:49.000
So, when remote agents are able to do this end-to-end, we can understand and integrate our systems much better.

s129
00:14:49.000 --> 00:15:07.360
So again, the market is moving, doesn't mean that we are necessarily married to agent chip until until the end, but these and other aspects of it make us strongly believe that is it is a very strong part of our internal platform now.

s130
00:15:07.360 --> 00:15:21.560
And the key takeaway that I want to I want to you all to take from is how can you benefit from LLMs, but not the traditional LLMs, but the lots of data multipliers that you and your team can use to leverage value.

s131
00:15:21.560 --> 00:15:23.560
Thank you.

s132
00:15:23.560 --> 00:15:25.560
[applause]