WEBVTT

NOTE Sentence-level transcript of https://www.youtube.com/watch?v=YiFqcu9YA38

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/YiFqcu9YA38.html. A cue ends where the next begins, or 2 s after its last word.

s1
00:00:12.960 --> 00:00:14.080
Hello everyone.

s2
00:00:14.080 --> 00:00:19.199
My name is Sarah and I have a confession to make.

s3
00:00:19.199 --> 00:00:22.960
So, I've been using Data Dog every day for the last six months.

s4
00:00:22.960 --> 00:00:27.279
And the other day my co-orker comes up to me and he's like, "Hey Sarah, can I look at this alert?"

s5
00:00:27.279 --> 00:00:28.960
I'm like, "Okay."

s6
00:00:28.960 --> 00:00:31.279
So, I open up Data Dog and I'm logged out.

s7
00:00:31.279 --> 00:00:40.079
It's kind of awkward, so I have to log in and he's like looking over my shoulder and I'm like overly conscious about everything that I'm doing on my computer.

s8
00:00:40.079 --> 00:00:45.680
And I open up the data the data dog dashboard and I just froze.

s9
00:00:45.680 --> 00:00:48.079
I had no idea where anything was.

s10
00:00:48.079 --> 00:00:50.559
And this is actually not a knock on data dog.

s11
00:00:50.559 --> 00:00:55.440
The problem was I had genuinely never opened the dashboard.

s12
00:00:55.440 --> 00:00:59.520
I've been using it every day without ever looking at it.

s13
00:00:59.520 --> 00:01:06.080
And so today, I want to make an argument that sounds kind of insane and then show you that it's obvious.

s14
00:01:06.080 --> 00:01:07.840
The dashboard is dead.

s15
00:01:07.840 --> 00:01:18.000
And I think that's great news for everyone in this room except probably for me because I am my team is responsible for the Composio dashboard.

s16
00:01:18.720 --> 00:01:20.240
Alas, how did we get here?

s17
00:01:20.240 --> 00:01:22.000
I believe a postmortem is in order.

s18
00:01:22.000 --> 00:01:23.840
Shall we?

s19
00:01:23.840 --> 00:01:26.000
The year is 2022.

s20
00:01:26.000 --> 00:01:28.400
It is the dark ages.

s21
00:01:28.400 --> 00:01:29.520
Someone pings me on Slack.

s22
00:01:29.520 --> 00:01:34.640
They say, "Hey, I get this error when I try and search for prod."

s23
00:01:34.640 --> 00:01:37.360
I would have instantly spawned five different windows.

s24
00:01:37.360 --> 00:01:38.560
Read the context in Slack.

s25
00:01:38.560 --> 00:01:39.840
Write a query in data dog.

s26
00:01:39.840 --> 00:01:41.759
Check post for this session.

s27
00:01:41.759 --> 00:01:44.079
Uh, fix the bug in VS Code.

s28
00:01:44.079 --> 00:01:45.759
Open the PR in GitHub.

s29
00:01:45.759 --> 00:01:51.200
That is five tools and five UIs I have to learn and relearn every time they ship a new redesign.

s30
00:01:51.200 --> 00:01:57.680
But frankly, redesigns are the least of my problems because every single tool has its own query language.

s31
00:01:57.680 --> 00:01:59.920
So, data dog has its own query syntax.

s32
00:01:59.920 --> 00:02:04.000
Jira has JQL, search, Slack has its search modifiers.

s33
00:02:04.000 --> 00:02:13.840
And it gets worse because even when these tools claim to speak the same language, let's say SQL, they don't even agree on SQL.

s34
00:02:15.200 --> 00:02:27.040
Every dashboard you've ever used, every weird obscure query language ever written was a translation device between you and your data because the machines on the other end could not understand what you actually wanted.

s35
00:02:27.040 --> 00:02:30.239
You never wanted a dashboard or it's cursed query language.

s36
00:02:30.239 --> 00:02:32.400
You wanted the answer.

s37
00:02:32.400 --> 00:02:36.080
And in 2023, everything changed.

s38
00:02:37.599 --> 00:02:40.239
The sparkle button.

s39
00:02:40.640 --> 00:02:42.560
The sparkle button was born.

s40
00:02:42.560 --> 00:02:48.800
And with a click of this magical button, an LLM would write a sometimes correct query to get you the answer you need.

s41
00:02:48.800 --> 00:02:54.080
That is if your question does not require more than two database joins.

s42
00:02:55.360 --> 00:02:58.400
And soon enough, these sparkling buttons were everywhere.

s43
00:02:58.400 --> 00:03:00.800
So dashboards became AI native.

s44
00:03:00.800 --> 00:03:02.640
Problem solved, right?

s45
00:03:02.640 --> 00:03:04.800
Well, not quite.

s46
00:03:04.800 --> 00:03:14.640
In November of 2024, Enthropic announced the MCP protocol with a promise to create an open standard for connecting AI systems with its data sources and that it did.

s47
00:03:14.640 --> 00:03:24.319
Instead of using a sparkle button, your own agent, Claude, whom you've already been using every day, could generate the query for you, execute it on your behalf, and give you the answer directly.

s48
00:03:24.319 --> 00:03:27.519
And you might be thinking, problem solved, right?

s49
00:03:27.519 --> 00:03:29.440
This is the end of the story.

s50
00:03:29.440 --> 00:03:32.879
Well, far from it, because MCP is a protocol.

s51
00:03:32.879 --> 00:03:35.840
It's a channel for communication between agents and your service.

s52
00:03:35.840 --> 00:03:40.799
And it's up to you, the service, to choose how to communicate with the agent.

s53
00:03:40.799 --> 00:03:43.599
And it turns out that makes all the difference.

s54
00:03:43.599 --> 00:03:52.000
If you've ever actually wired up a dozen MCP servers, you will know the reality is a mess for three reasons.

s55
00:03:52.000 --> 00:03:53.599
Agents don't learn.

s56
00:03:53.599 --> 00:03:55.040
Every conversation starts from zero.

s57
00:03:55.040 --> 00:04:01.840
It has no memory of how it fumbled formatting links properly in Slack yesterday, and so it's just going to fumble again today.

s58
00:04:01.840 --> 00:04:09.519
And we patched this with skills, but skills is just a band-aid because agents become dumber with the more context you provide.

s59
00:04:09.519 --> 00:04:12.879
And loading more tools, more skills takes more context.

s60
00:04:12.879 --> 00:04:17.600
And if you connect enough servers, you're dumping thousands of tool definitions straight into the context window.

s61
00:04:17.600 --> 00:04:22.240
I mean, our GitHub like toolkit alone has over 200 tools.

s62
00:04:22.240 --> 00:04:23.360
The model just drowns.

s63
00:04:23.360 --> 00:04:29.680
It grabs the wrong tool or struggles to resolve dependencies and it can't figure out which one it needs to call first.

s64
00:04:29.680 --> 00:04:32.800
And third, every app is isolated.

s65
00:04:32.800 --> 00:04:36.479
Each MCP server knows about itself and nothing else.

s66
00:04:36.479 --> 00:04:44.880
So the moment a task spans two apps and like they always do, it is your job to piece that together.

s67
00:04:44.880 --> 00:04:48.000
And so MCP gave agents a door into every app.

s68
00:04:48.000 --> 00:04:54.720
But it left them standing in thousands of separate rooms with no map and no memory of ever being there.

s69
00:04:54.800 --> 00:05:03.759
I'm going to show you a mockup demo here of how I would solve this bug report uh from the very beginning how I would solve it in 2026.

s70
00:05:03.759 --> 00:05:09.120
So imagine in this mockup I have my claude connected to the composio MCP.

s71
00:05:09.120 --> 00:05:19.280
What I would do is I would literally just copy and paste the link to the message from Slack and say please use century data dog find the root cause a draft PR make no mistakes.

s72
00:05:19.280 --> 00:05:24.400
And the first thing it does is it calls composio search to state the task it wants to accomplish.

s73
00:05:24.400 --> 00:05:26.880
So in this case it stated three.

s74
00:05:26.880 --> 00:05:28.400
It wants to fetch slack messages.

s75
00:05:28.400 --> 00:05:29.680
It wants to search sentry issues.

s76
00:05:29.680 --> 00:05:31.680
It wants to search data dog logs.

s77
00:05:31.680 --> 00:05:39.039
And for each of these three, Composeio returns not only the correct tools it needs, but actually a plan of how to use them.

s78
00:05:39.039 --> 00:05:46.080
So for example, for Slack, you actually have to find the Slack channel ID before you can start quering for messages.

s79
00:05:46.479 --> 00:05:50.160
Now that it's armed with all the context it needs, it has a plan.

s80
00:05:50.160 --> 00:05:54.000
It starts pulling the message from Slack to see what the user complaint was.

s81
00:05:54.000 --> 00:05:56.320
Then it begins pulling from data sources in parallel.

s82
00:05:56.320 --> 00:06:00.479
Data dog sentry and it begins scanning the codebase.

s83
00:06:00.479 --> 00:06:06.000
Once it identifies the root cause, PR is up with a fix in less than five minutes.

s84
00:06:06.000 --> 00:06:10.479
I didn't have to build a workflow or write a skill to teach it how to do this.

s85
00:06:10.479 --> 00:06:14.240
This is literally just claude using Composeio's MCP.

s86
00:06:14.240 --> 00:06:21.360
And it has become so much easier to do things this way that opening up Cloud has just become like muscle memory for me to do literally anything.

s87
00:06:21.360 --> 00:06:27.840
And the dashboards I used to open every day are becoming increasingly unfamiliar to me.

s88
00:06:27.840 --> 00:06:30.000
And the impact of designing for agents is measurable.

s89
00:06:30.000 --> 00:06:38.639
So these are some early unreleased uh results comparing Composio against each app's own native MCP that's listed in the clawed uh marketplace.

s90
00:06:38.639 --> 00:06:44.080
So it's the same tasks, it's the same model, and we see a clear difference.

s91
00:06:44.479 --> 00:06:45.680
But why?

s92
00:06:45.680 --> 00:06:49.840
Why is this experience so much better than native MCPs?

s93
00:06:49.840 --> 00:06:56.319
It's because at Composeio, we are developing a brand new interface specifically designed for agents.

s94
00:06:56.319 --> 00:07:05.759
We translate messy, sparsely documented, everchanging APIs from all of the apps that you live in every day into something that agents really love to use.

s95
00:07:05.759 --> 00:07:19.199
So, we build on top of MCP CLI and native tools to deliver a cohesive, unified, and excellent experience for your agents that allows them to perform complex operations across these apps seamlessly.

s96
00:07:19.199 --> 00:07:23.919
I'm going to show you a more classic example of like what I usually use it for every day.

s97
00:07:23.919 --> 00:07:26.319
So, let's say I want to dig into some user data.

s98
00:07:26.319 --> 00:07:30.639
I want to see the distribution of what vertical my users select during onboarding.

s99
00:07:30.639 --> 00:07:32.639
I can simply just ask Claude.

s100
00:07:32.639 --> 00:07:35.199
Claude will again run Composio search.

s101
00:07:35.199 --> 00:07:38.880
It will write a query in Post Hog, execute it nice and simple.

s102
00:07:38.880 --> 00:07:40.720
There's my results.

s103
00:07:40.720 --> 00:07:43.840
But what if I wanted to take this up a notch?

s104
00:07:43.840 --> 00:07:46.560
So I want to look at that e-commerce section right there.

s105
00:07:46.560 --> 00:07:50.639
And I want to see what tool kits they like to use.

s106
00:07:50.639 --> 00:07:54.240
Toolkits is another word for apps we like to use at Composio.

s107
00:07:54.240 --> 00:08:02.960
So let's see if cloud can figure out how it can pull this data from metabase uh given the user ids in post hog.

s108
00:08:02.960 --> 00:08:06.319
So once again it uses composio search.

s109
00:08:06.319 --> 00:08:16.000
It queries the user ids of those who selected e-commerce from post hog and it actually saves it without loading the full results into its context.

s110
00:08:16.000 --> 00:08:22.479
Then it makes some queries uh in metabase to get the database schema sample some data get a feel for how things are structured.

s111
00:08:22.479 --> 00:08:34.800
And once it's confident, it actually uses a very cool tool called Composio Remote Workbench to dynamically generate an SQL query with a reax string to search for all those user IDs

s112
00:08:34.800 --> 00:08:39.360
again without ever loading the entire thing into its context window.

s113
00:08:39.360 --> 00:08:41.919
And there we go.

s114
00:08:42.479 --> 00:08:48.880
So we can see the most popular toolkits amongst this user persona in less than a few minutes.

s115
00:08:48.880 --> 00:08:51.519
Uh and that's data pulled from both sources.

s116
00:08:51.519 --> 00:08:54.560
If you want to play with the live version, come visit us at the Composio booth.

s117
00:08:54.560 --> 00:08:58.320
Uh, I'd love to have you play with this live.

s118
00:08:59.839 --> 00:09:01.760
And so, the dashboard has died.

s119
00:09:01.760 --> 00:09:04.720
And I'm thinking more and more about what this means for the future.

s120
00:09:04.720 --> 00:09:11.120
Many startups have begun making their landing pages and their websites AI friendly for GEO purposes.

s121
00:09:11.120 --> 00:09:16.080
But so few have really prepared their applications to be used by agents.

s122
00:09:16.080 --> 00:09:19.200
Compose began as a tool to help developers.

s123
00:09:19.200 --> 00:09:28.000
We were taking these popular apps that users like and turn them into these tools that agents could use.

s124
00:09:28.080 --> 00:09:33.519
But the humans are tired of dashboards and their agents are tired of poorly designed MCP servers.

s125
00:09:33.519 --> 00:09:41.200
And now we're getting requests from startups saying their clients are begging for a way to use their services through their agents.

s126
00:09:41.440 --> 00:09:44.160
So this is a lesson for everyone building anything right now.

s127
00:09:44.160 --> 00:09:47.120
You are now serving a new species of user.

s128
00:09:47.120 --> 00:09:48.399
They don't have eyes.

s129
00:09:48.399 --> 00:09:51.600
They are not going to click your sparkle button.

s130
00:09:51.600 --> 00:09:59.600
It shows up with a goal and a set of tools and it judges you on exactly one thing, whether it can get the job done.

s131
00:09:59.600 --> 00:10:05.360
For a decade, we've built pro uh we've built products to be easy for humans to use.

s132
00:10:05.360 --> 00:10:11.120
And the next era belongs to those that are easy for agents to use.

s133
00:10:11.120 --> 00:10:12.160
So, picture me again.

s134
00:10:12.160 --> 00:10:14.079
I'm frozen in front of that data dog dashboard.

s135
00:10:14.079 --> 00:10:17.839
And I was thinking to myself in that moment that I'm falling behind.

s136
00:10:17.839 --> 00:10:22.160
Uh, but now I think of that moment as a preview of what's to come.

s137
00:10:22.160 --> 00:10:24.399
Thank you.

s138
00:10:29.462 --> 00:10:31.462
[music]

s139
00:10:37.977 --> 00:10:39.977
[music]