AWS re:Invent 2025 - Automate any business process using Amazon Quick Suite (BIZ224)

All right, welcome everyone.

Uh, we're gonna start a little bit early since everyone's on time and we got a packed house.

Um, everyone is gonna need to put on your headphones.

I'm, uh, shouting a little bit just so that those who don't have their headphones on can hear me but um, hopefully once you get your headphones on everyone can hear and the mic's coming through clearly and we got a few head nods. It's, it's looking good.

Um, it's a little bright, uh, but I appreciate you coming out, uh, day 4 of reinvent.

I would guess a few of you were out late last night so it's good that it's 11 a.m.,

um, gave you a little bit of time to, to find us so I appreciate you spending, um, 50 minutes with us or so today.

Uh, we're excited to talk to you about Agentic automation.

Uh, my name is John Brock. I lead the Agente Automation products in a new service that we recently launched called the Amazon Quick Suite.

Uh, we'll talk about that a little bit and then we'll go into some deep dives about quick flows and also Quick Amate, really spanning the spectrum of business automation.

Um, up here with me is, is Nupur.

Hi, I'm Nupur, product manager for Quick Flows.

Great, yeah, I'll be talking shortly. We'll get

John to start us, yeah, yeah, thanks, Snooper. So, um,

additionally, the, the thing we're really excited about is we have Brian Merlin Brink joining us today from Jable. Jable

is a great partner of AWS but also a great customer.

Um, he'll tell you more about Jable, but they're automating some really key critical business processes using Quick Automate.

And so we're gonna spend a lot of the time today doing a deep dive on their customer use case and how they're using Quick Automate.

So this is kind of an overview. I, I gave a quick preview of this, but we're going to talk about Quick Suite a little bit, Quick flows, Quick Automate.

Newer also has some customer case studies and some deep dives about how Amazon is actually using Quick flows to automate a lot of our business processes by business users. So these are business users without technical experience, without AI experience, but they're still using Agente AI to automate their key routine business processes. And then we'll spend quite a bit of

processes. And then we'll spend quite a bit of time talking about Jable and how they're automating a really interesting use case internally and then we've got a few QR codes at the end.

So if you're worried about how do I learn more, it'll be here at the end.

That's kind of the teaser so that you stay all the way through, um, but then we'll we'll wrap up and we'll we'll go on to our next sessions.

So before we start, who, who has heard of QuickSweet or knows um remotely anything about it? OK,

about it? OK, so, so maybe 20-30%, which is good, um, we're, we're trying to get the word out.

We launched in October. We're gonna talk about that a little bit more. Uh, many of you are probably more familiar with QuickSight, uh, for BIN analytics. Um,

if you have Quicksight today, many of your instances are already enabled on QuickSuite as well.

So we always at AWS start working backwards from the customer and trying to figure out what products we build, what problems we want to solve, what are the challenges that you need help with, right? We probably talked to many of you in this room about what your expectations are, what you're hoping to get out of AI, how to use agents and how to apply them.

Um, the slide is interesting and it, it's kind of pointing in a direction that I, I think is, is, is something to dive in on. So we have 82% of organizations believe that they're gonna integrate agents into key business processes within 1 to 3 years.

But at the same time, within that same time period, we only expect 15% of business decisions to be made by those agents.

And so to me that number seems low on the decision making aspect.

And you know pretty optimistic on the number of agents integrated into business processes because otherwise that still means that there's humans that are maybe the bottleneck or having to do routine monotonous work or not really transforming how people work.

And so when we saw these stats and these numbers, we really wanted to figure out how do we help employees work with AI and work with agents more effectively. And one of the ways that we do that is through automation and hence why automation is such a key part of Quick sweep.

And so when we're talking about automation, it not only spans the business processes in the back office, but also how business users work, how knowledge workers work, how many of us work.

Many of us start our day with a task or a challenge, a problem, something that we need to improve about our business or improve or solve for our customers.

This process kind of. And we've divided into three pillars.

The first is gathering information from all different sources, internal sources. Maybe there's external

internal sources. Maybe there's external sources.

Maybe there's research we need to do.

We have to gather all that information and then oftentimes we need to consult with researchers or specialized data analysts or a BI engineer to get a dashboard built. This whole process

built. This whole process takes a lot of time um and we have to do all this before we can actually make a decision or take an action.

Once we've gathered all this, that's really when the human aspect comes into play, right? We're we're all very smart,

play, right? We're we're all very smart, very intelligent people, um, and so we wanna use that data, make a decision and improve the outcome for our business, um, but that takes a lot of time and so we get through that process. It's, it's

challenging, it's hard, it's not really fun, uh, but then we come back tomorrow.

And the next day and we do very similar things day by day and so we wanted QuickSuite to really help a business user through this process and allow agents to help the business users get through their day more effectively and more efficiently.

And so what QuickSuite is is a brand new way for the for the AI enterprise if you will. It's a unified workspace that

you will. It's a unified workspace that allows humans to collaborate with AI agents over a secure view of all of your customer data.

So all of your customer data means all of your structured data coming from sources like Aurora or Redshift or Dynamo or um your, your, your data lakes and your data warehouses, even if they don't live in AWS.

It includes your unstructured data coming from S3 and OneDrive and SharePoint and Google Drive and your docs and your wikis and your Quip and your all these different productivity solutions, and you have important business documents that are important to get insights from and information from.

And when we bring all this together, we allow humans to collaborate with the AI agents.

They're able to simplify and accelerate that process from idea or hypothesis through the information gathering stage, through the insights gathering stage, through the research stage to accelerate you to the point that you make a decision and take an action.

And that's also the other thing that we wanted to really emphasize with QuickSweet is that it's not just a chat interface, right? It actually helps you make a decision

right? It actually helps you make a decision and then take an action.

Right, take an action means update data in another system, file a customer support case, resolve a customer issue, and so you're not just asking a question and getting an answer, but you're actually taking action and moving your business forward.

And so the overview of Amazon QuickSuite is again a unified workspace where AI teammates help you perform research, find business insights, and automate key business processes.

So we have built-in agents that we'll talk about in just a second to help you complete these tasks and really you can see on the right we're really about enabling employees to make better decisions faster.

Take action on those and ensure that QuickSuite delivers security governance compliance visibility logging monitoring everything that enterprise organizations need to feel confident deploying and rolling out Agente AI.

Across the board, right, we really believe QuickSuite is transformative for every employee in your company regardless of their technical skill, regardless of their experience with AI, and so we, we know that enterprise reliability and security is front and center to that mission.

So a quick architecture slide, usually everyone loves this slide, so this is kind of the one to get your phone out, if you will, but Quick Suite, if you think about it, is kind of this top layer of this slide, the built-in agents that we have to help you and the business users perform those tasks. That we just talked about starting on the left to gather insights we have your normal chat agents. There's

a built-in agent there's also the ability for you to define your own custom agents and ground them in very specific contextual knowledge for your task or your project or your your team.

We also have a brand new research agent.

That performs comprehensive research across not only your internal data but also your internet sources and recently launched last week trusted third party providers like IDC, Factsat, PubMed, US patents, and many more.

This research agent runs for 20 to 30 minutes. It's

a recursive agent that's verifying, validating, checking stats, and it's going to output a 20 to 30 page comprehensive research report regardless of the domain that you're working in. It's

pretty incredible. I really encourage you to try that out. In the middle is BI

out. In the middle is BI and analytics, so we've improved and adjusted and enhanced our Quicksight capabilities to enable you to ask questions about your structured data and have those insights with the structured data combined with all the insights from your unstructured data.

What this means in practice is that when I open up a dashboard now within QuickSuite, I can ask a question about that dashboard, and if the data is on the dashboard, of course it works, but also it's doing more of an agentic Q&A process over the underlying data and the data sets that are backing that dashboard. And

that dashboard. And so my, my user permissions are respected, right, so I'm not going to see data that I shouldn't see.

But even if that BI engineer didn't add the one column I need to the dashboard, I can still get insights because now this Q&A agent can access and do dynamic SQL and dynamic queries to find the information and insights that I need.

Additionally, the agent has been enhanced with code skills and a code interpreter, so as we know, agents and models are not very good at math, and so the agent actually creates just in time code to perform the numerical calculations and generate real, provable, mathematically correct insights.

And then the category that we're all here for is the automation agents.

Uh we'll dive into this a little bit more on the next slide.

And then these experiences within QuickSuite are embeddable as well, so you can embed your dashboards. I was just talking to

your dashboards. I was just talking to a few customers up in the front and at the EBC earlier this week.

They put a lot of their dashboards into their internal sites and portals so that their users start their day. And in in their day in their current process, but they're still getting the benefits of of QuickSuite in their existing applications.

This also extends to chat. So if you need a chat assistant or a chat agent on your wiki, you can now embed those chat agents on your portals.

And then in the middle, if you look, think of this and really reconcile this is almost that agent core layer.

So everything within QuickSuite is built on top of Agent Core using all the new agent core technology that you've learned about this week.

But really key to QuickSuite is ensuring that we deliver governance data compliance.

Security, responsible AI and access controls, making sure that all of your data stays safe, secure, and it remains yours. The data is never shared with model

yours. The data is never shared with model providers.

We don't see it for service improvement. Your

queries, your responses remain yours, and your enterprise data remains yours at your company.

All of this is built on a broad set of data connectors so we can ingest data from 50+ data sources with built-in connectors, but we also have connectivity to over 1000 different sources using Open API, API, and now MCP servers.

A2A will also be coming soon, so be watching for that and probably Q1.

Additionally, access to your databases and data warehouses, and now users can also upload their own files through a concept called Spaces so that they can curate a curated, grounded set of data for their agents and their chat experiences.

All this is built on Bedrock.

QuickSuite uses a variety of models for the task at hand.

We Optimize that we make the experience for business users great out of the box with high accuracy.

Different products within the suite allow you different levers of control on how much model selection and model choice you have, and it just kind of varies based on the use case and the audience.

Additionally, action is really key that you can update your data in any of these different systems. So now talking about the automation agents within QuickSuite a little bit more is really the key of this session. Is

starting with flows is really tailored as a business user facing solution to automate routine business tasks for team and individual productivity.

Newer is going to go into a little bit more detail here in just a second. And then on the other end of the spectrum we have Quick Automate. This

enables technical teams to automate your most complex workflows with multiple agents.

You're able to interleave those multiple agents with deterministic. Steps like pulling

with deterministic. Steps like pulling files from S3 operating on Excel. You'll see that in Brian's

on Excel. You'll see that in Brian's demo, and it is essentially the fabric so that you can stitch multiple different agents together.

Some of those agents are built in and available out of the box from Quick Automate.

Some may come from Agent Core with you and your developers are building.

Some may be agents from third party providers that you pull into those automations to help you complete your task.

So with that, I'm going to hand over to Ner and she's going to give you a little bit more deep dive on flows.

OK. Thank you, John, for that.

Let's take a step back.

I'm sure a lot of us have used all these amazing automation tools out there, and yet these tools rely on us in some way. We the people

way. We the people are the glue holding these processes together.

We are the ones jumping in between apps.

We are the ones copy pasting data from one system to another.

Uh, we are the ones sending executive summaries and follow up meeting notes.

Uh, and somewhere in all of this busy work, real productivity gets lost.

And that's exactly where quick flows comes in. It's our answer

comes in. It's our answer to make automations work for every business user, not just the tech savvy ones.

Now, before I get carried away and start listing all the capabilities of quick flows, you must ask me, what's the proof that this works?

Fair enough.

So let's dive into the proof.

A few months ago we launched Quick and we did the launch internally even a little earlier, and since then we've had tens of thousands of users at Amazon triggering close to half a million floor runs.

Now, I will share 3 stories today from across Amazon where individuals in different parts of the company have actually discovered flaws as an answer to automate their routine, time-consuming, repetitive tasks.

And Once they saw success, they broadly shared it with their teams. So let's start closer to home, within AWS.

Arus is in the worldwide Specialist Organization. I think I see him at the back and

Organization. I think I see him at the back and he looks nothing like this, so if you want to catch him, don't go by this picture.

Um, so he and his team of enterprise specialists, account managers, solution architects face this impossible equation where they have these complex business proposals, back to back customer meetings, frequent updates to CRMs, um, crazy days.

Now at Amazon scale, this challenge is amplified exponentially.

thousands of customer obsessed specialists must maintain expertise across the AWS portfolio of services, but at the same time they must also understand customer industry trends.

Now, on one hand you have this rapidly growing AWS portfolio of services, and on the other hand you have this constantly shifting, evolving industry from the customer side.

And this puts a lot of pressure.

And these teams are constantly drowning in research and preparation before they can meaningfully contribute to business outcomes for you. So

for you. So what did Aus do?

He created a meeting preparation flow that would pull data from AWS's product details, documentation about services, and at the same time bring context from the web about a customer.

We'll see a very sped up version of that run just now where I start by entering some information about the customer.

I'll pick a customer and you'll see, and then I get detailed notes on what I can talk about.

OK, so the customer will be putting up my email. The customer I wanna

email. The customer I wanna meet is Ring. They're in the home security space. I'm meeting their innovation

space. I'm meeting their innovation team, and they want to know how AWS AI solutions can help them.

Now you can see on the left hand side there is a longish workflow with multiple steps which is going through, you know.

Value proposition across various AWS services in the AI space and at the end of it, it's giving me an HTML formatted email that I can, you know, send to myself to somebody else to prepare for that meeting. Now this

that meeting. Now this is the email you can see I sent it to myself in this case, but it has a nice little table with all the details relevant for the customer.

It has links to resources that I can follow up on. I can

send it to anybody else on my team who's also preparing for that meeting.

In essence, Flows like these, in fact, this specific flow is being run 1,000s of times every week, and each run is saving these teams 2 hours.

This team is on track to save tens of thousands of hours annually with just this one flow.

If you don't believe me, Arus is at the back. You

can catch him after this.

Now, this isn't just about time savings.

It's about the fact that this was done with zero coding.

Not a single line of code written.

It has reimagined how account management can be done at scale.

Now you may ask me, Nupur, this is great.

You talked about sales, account management, but what about more complex, highly regulated areas?

Fair enough. Let's dive into global regulatory compliance.

Donna's on that team for us at Amazon, and she tells me it's like working like a detective, but instead of solving one crime at a time, they're looking at 23 different crimes in different languages with different rules, with different evidences.

And oh by the way, this is high risk, so you don't want to get this one wrong.

So for our compliance team, this was not a hypothetical scenario, this is their reality.

Every single policy review was taking them 80 hours.

I let that sink in. 80 hours

for one review is 2 full workweeks of effort.

Imagine instead of being that overwhelmed detective.

You are now assisted by an AI powered research assistant who can process a tremendous amount of information, analyze regulations, find patterns across 23 jurisdictions.

Basically, it's doing all the heavy lifting for you, and that's exactly what Donna's flow does.

And now while a great deal of heavy lifting is done by, you know, the floor, this team is still doing the critical work of, you know, decision making and critiquing that policy.

So from 80 hours they've come down to 15 to 20 hours of effort on every single policy analysis.

That's 75% savings.

Think about your biggest operational bottleneck and shrink it by 75%.

What would that mean for your business?

Now cannot share Donna's flow in entirety, but I'm going to talk about one neat little step in her flow.

So the policy team, when they're analyzing, you know, the various policies, they also need examples of how these policies would apply to products.

So the UI agents step in, uh, flows, this capability helps them do that. The UI agent can essentially navigate websites, perform browser tasks, take actions.

So in this case it's actually going on the Amazon website, finding matches for I think lithium ion batteries to which the policy may apply, and it can bring back results.

It can do things like submit forms, etc. too so.

In her workflow, if she identified a risk with a certain type of products, she has a step to give her examples. Again,

this can run on its own, and she doesn't have to, you know, go in between apps to find those examples.

OK, for our last story, we're gonna raise hands.

How many of us have been there where to answer one single business question we've hopscotched between 5 different apps?

Show of hands, there are enough over there, OK.

So that scenario was very common on our global seller team. This is on the retail side of our business.

So if a seller walked into the office and said they wanted to expand internationally, sell the products in a different country, some account manager was spending easily between 2 to 3 hours to, you know, bring up some data, figure out what the best strategy and insights would be to guide them.

Now instead of that, Evan on that team actually completely flipped the script.

He built this selection opportunities navigator which saved the team hours for every such task.

So from 3 hours of work they went into, you know, insights in 10 minutes.

Every account manager on their team was saving 19 hours of work per month with just this one flow.

Across the team they saw 12% bandwidth savings.

But more importantly, they've stopped being these spreadsheet warriors, and they had time back to be what their sellers need them to be trusted advisors.

OK, I've talked a lot about the success stories, but I haven't really shown you what quick flows are, how they can be created, uh, how do they run, so we're going to take a sneak peek at that.

Here I'm going to describe in plain English what flow I want to create.

So I have an Asana board where I get my customer leads. I want to summarize the status

leads. I want to summarize the status for my leadership, and at the end of it I want you to send me an email every Monday so I can review it before I send it out.

So I'm just describing that in plain English, and it's gonna generate this flow for me. It's really that simple.

for me. It's really that simple.

Now, because I didn't specify when on Monday, it's nudging me to specify pick a day, which Monday do you want me to start, um, and here I'm specifying the time on Monday, and I'm going to activate the schedule. Of course I'm gonna give some inputs on

schedule. Of course I'm gonna give some inputs on which asana board it should look at and which email address uh it should uh send the email to as part of the flow.

Gonna make some more configurations saying I've already created a sales assistant.

Use that to, you know, summarize the findings from the sales leads, and there I am running the flow.

Now it's not Monday. This is me running it proactively, but you can see it gives me detailed inputs for what I need. And if you need something to be changed, it's basically another English language prompt back in the editor mode. And yet at the end of it

mode. And yet at the end of it I have this weekly email that is being compiled.

So as that comes up, uh, you'll see I'm not gonna show you another email, we already saw that. I'm gonna show you something different that you can do.

You can also download the output. You don't

always need to send that email.

Uh, you can export it as a PDF or Word document. We're gonna bring up more formats.

I know some of us uh want PowerPoints.

And what did I do?

OK, so this is just giving me acknowledgement that it's uh sent that email, and now I'm gonna hit the download button to get that output. In this

case, I'm only interested in the leadership report and I'm gonna save it to a space for those of us who are not familiar, think of space as a data boundary, a data scope you can define within quick.

You can keep, you know, for your sales teams, for your own team, documents in that specific area. This could be a combination of

specific area. This could be a combination of uh structured.

Structure data as John talked about earlier, so your dashboards, your PDF files, your actions, they can all be a part of that space, uh, and you can set permissions, uh, for users you want to share it with. So here I'm putting this, uh,

with. So here I'm putting this, uh, generated output which is the leadership report, uh, in that same space so I can share it with my team.

The cool thing about this is I can actually talk to the space.

So next week if I want to know what was the status, you know, for the past two weeks for a customer, all I need to do is talk to the space and say, how's this customer trending?

Did we close the lead or not?

OK, so we saw how you can create a flow, how you run it, uh. But

uh. But what is really the entirety of what you can do with flows?

Our main purpose is to enable every business user to be able to create, customize, and share these purpose-built flows on their own without needing any technical support. And they do

support. And they do this while giving IT teams peace of mind. So we know there's this been concern about, you know, shadow IT and at what cost are we democratizing use of AI.

This has all the admin controls, governance that your IT teams may need to manage access to certain features. In fact, if you want

certain features. In fact, if you want to put guardrails around how these flows are shared in the common library, they can put an approval review behind that too. There

too. There is one feature John talked about earlier, which is a more detailed, comprehensive report that you can create through the research agent.

Recently we announced the research agent is now in flows.

So what that means is in your workflow you could actually trigger a call to that research agent.

So let's take Donna's example from compliance team. If she wanted

compliance team. If she wanted to do a deep dive into a policy, she can just use this research agent now, and we know research takes somewhere between 20 to 40 minutes, so you can schedule these flows too.

OK, so what did we see today?

These powerful automations are for business users, and they're just not another automation tool.

They're productivity multipliers.

We saw how Aus's team is on track to save tens of thousands of hours.

We saw how Donna's team has reduced the time spent on research by 75%.

We saw how Evan in Global Selling is giving back 19 hours per month to each of his account managers.

And these are real Amazonians solving real problems right now. I

know many of the customers I met earlier this week have said, you know, you're pitching this to us, but what's the proof? What are you doing at Amazon?

the proof? What are you doing at Amazon?

This is that proof.

And the best part is they did it all on their own.

There was no coding required, these are just business users with a problem and a tool that empowered them to solve it.

With quick flows, we are looking to put the power of agentic AI directly into the hands of the people who know their work best, your teams. Thanks Snooper and congratulations on the launch of Quick Flow's, uh, amazing capability and QuickSuite.

Um, I, I'm blown away. I, I

think the use cases you saw today, um, you could see on the table of contents on the, the left that these, these are multi-step, right? They're, they're multi-agent. They

right? They're, they're multi-agent. They

include reasoning. They include research. They include a UI agent, amazing capabilities that a non-technical user has been able to build, deploy, and productionize within the organization.

I don't know about you, but not many of even our technical teams have, have sort of reached that level of maturity with, with Agentic AI and rolling out AI to, to an entire organization, but it's pretty incredible. All this is fully managed. It's run for you. There's no

fully managed. It's run for you. There's no

scaling. There's no,

you know, turning up servers. There's no carrying a pager. All of this is fully managed

a pager. All of this is fully managed as well.

So those are pretty complex processes I have to say, but what about all of your back office processes that maybe are running at really high scale? So

think of customer support, ticket triage or claims processing or loan origination.

How can we help with those?

So many of you today, uh, you're not going to admit it, so I'm not going to ask for hands to be raised, but many of you probably have hundreds if not thousands of enterprise systems that probably aren't well integrated.

They're probably data silos.

Newer mentioned, you know, you're navigating between 5, 10 different sites or applications or portals to get the information you need.

We're all technical folks, mostly in this group and in this room, but you know we've spent years building integration pipelines.

We've spent years of our life trying to figure out ETL jobs, and you know some of it sort of works and. And

and. And we've also tried other solutions, right? We've

tried RPA, you know, we've built low code applications that have had varying degrees of success, but many of these traditional tools are just not able to automate a lot of these really key business processes that still require reasoning, human judgment, logic, and agentic reasoning.

Additionally, they're really expensive to build and maintain.

Um, RPA is, is almost a four letter word these days, um, you know, many of you have inherited RPA. You're maintaining it. You're looking

RPA. You're maintaining it. You're looking

for what's next.

Um, Quick Automate is an amazing solution that we launched, um, inside of QuickSuite that allows you to build, deploy, and manage multi-agent automations for your most complex business processes.

So we start with an agent that helps you build and maintain the automation.

So you come, you can describe your process if you have an SOP or a standard operating procedure you can upload that to our our automation agent and it will help you scaffold out that entire workflow.

It's um sort of iterative, so it's gonna ask you some questions. It's gonna have you review a high level before it gets into all the details. It's going to make suggestions

details. It's going to make suggestions based off of best practices and it's gonna help you build out that work. Flow without having to manually drag and drop, you know, 100 or more nodes onto a canvas.

Additionally, if you need to make a change, oftentimes changes in, in older tools or tools we may be using today require that you update all 50 of those nodes, right, because there's cascading impact.

Well, now within Quick Automate you can describe that change in natural language through chat, and we automatically update your workflow for you. But the most

for you. But the most important thing about Quick Automate is really being this orchestration fabric between agents.

So there's a few different types of agents that we have in Quick Automate.

Uh, the first is the UI agent. So it's the same UI agent that Newer showed in flows, uh, but it's an industry leading computer use agent that's been fine tuned, and we've used reinforcement learning on it to tailor it for business applications.

So just, just being honest, it's probably not the best computer use. Agent to book your hotels and your

computer use. Agent to book your hotels and your flights, but it's been trained and reinforced on things like ERP and forms over data applications and internal data systems that we need to automate tasks on that don't have APIs or can't be automated in traditional systems that we use today.

The second type of agent is that within the workflow itself you can define a custom process agent that's built and deployed using Agent Core.

That's a fully no code agent building capability.

Allows you to define an instruction, a prompt. You can select the tools

a prompt. You can select the tools that you want to give to the agent. You select your knowledge that the agent has access to, and then we take care of the heavy lifting of deploying it to agent cop, but you get the benefit of having all the capabilities, of things like memory and guard rails and, um, you know, upcoming soon the new eval, um, you know, experiences that were launched.

To create this really amazing process agent very specifically tailored for your task, we have one customer that's fully automated their customer support inbox triage.

They have an agent that evaluates every email that comes into that inbox.

It has a series of tools. It has a Jira. Ticket tool. It has a ServiceNow tool.

Jira. Ticket tool. It has a ServiceNow tool.

It has a customer response tool. It has an email tool. It has a human in the loop tool

tool. It has a human in the loop tool so it can escalate if there's a question or you know an escalation needed or human oversight needed, and it automatically reasons about the email that came in, came in and automates the next step. The third type of

step. The third type of agent is that you can integrate your own agents that you've built with Agent Core, Bedrock agents or pull in agents and third party systems using MCP.

So whether your agents are built in Quick Automate or somewhere else, you can include them and put them in part of your automation fabric, and then the benefit is that you get all the automation connective tissue, if you will, around this, right? So you get deployment, you get versioning,

right? So you get deployment, you get versioning, you get rollback abilities, you get scheduling, you have logging, monitoring, governance controls to make sure that these agents within your process stay on track, have visibility into what they're doing, and you can go back and review what's been happening within your organization.

Additionally, very differentiated we believe for our product is that it's a purely consumption-based pricing model like every AWS service or most AWS services I should say, and you only pay for the time that that automation is executing if you're using the agents built into Quick Automate, the usage of that, the token consumption is included. It's pretty amazing.

included. It's pretty amazing.

It really simplifies the pricing, the billing choices, and the procurement.

I've mentioned a few of the use cases. There's,

there's a few here on the screen, um, but really just think about your most complex business processes that really run your business, and we have customers building those today.

Um, Brian's gonna come up on stage and tell you about Jable and what Jable is doing, um, but as you can see, we have customers doing merchant onboarding to verify that they're onboarding customers that are compliant to their banking policies, right? It is a quick automate flow.

right? It is a quick automate flow.

We have merchant due diligence for M&A acquisitions and uh due diligence processes. We have

know your customer uh processes. We have supply chain resilience

uh processes. We have supply chain resilience and rebalancing optimization uh problems being solved with Quick Automate.

It's a really amazing technology, but, um, you know, again I'm, I'm excited. Brian, welcome.

Um, we're excited to have you. Thank you for being a great partner and, uh, let's, let's learn a little bit more about Jable. Yeah, alright, thanks,

about Jable. Yeah, alright, thanks, John and thanks Nooper and.

Thanks everybody for coming out.

My name's Brian Merlenbrink.

I'm, uh, with Enterprise architecture with Jable. I'm leading our data

with Jable. I'm leading our data and AI technology strategies.

Today I wanna tell you a little bit about Jable and what we're doing with Quick Automate, um, and the Quick suite.

So first, Jabel is one of the world's largest contract manufacturers. It's

contract manufacturers. It's kind of a big secret. Many of you don't know about Jable. We've been around for

about Jable. We've been around for just about 60 years. In fact, next year will be our 60th birthday.

We were founded by James Golden and Bill Marian, uh, starting in Michigan, building circuit boards. So James and Bill, they put their

boards. So James and Bill, they put their names together. That's where we get J Bill from.

names together. That's where we get J Bill from.

So we did, we're just about a $30 billion company in revenue as of uh FY 25 which we closed up recently.

We have about 140,000 employees across 30 countries.

So we have a lot of diversity in what we do, and our vision is to be the world's most advanced and trusted manufacturing solutions provider.

So what do we do? So we

have 400 of the world's most known brands, so things that every, everyone in this, uh, audience uses.

We span across many different industries from automotive to, uh, warehouse automation to semiconductor to packaging and healthcare.

To cloud and data center infrastructure where we're building servers and racks that are powering a lot of the AI workloads and the AI tools just like Quick Automate and other AI tools that you guys are using.

And we do this across the entire life cycle, so we will co-design with our partners. We will

go from design into manufacturing, go through mass production, logistics, and then even service with the with our customers as we move forward.

And through all of this across all these industries across 400 customers, we manage a massive supply chain, so we have over 38,000 suppliers globally that we're operating with and we're our global spend with those suppliers is is over 25 billion annually.

So it's a big, it's a big piece of what we do is working with our customers.

To drive that forward.

And so when it comes to AI to manage that supply chain, to manage what we're doing for our customers, there's a lot that we put into it, and we have a few key imperatives that we drive all of our data and AI strategy for.

So the very first thing is business first. So we focus on

business first. So we focus on what is the value for the business, for the customer.

We're not using technology just for technology. We're looking at where technology

technology. We're looking at where technology can apply to provide more value to the customer.

And then we're taking that same technology and we're working internally to find new ways of working.

We're ensuring that we're continuously providing education to our employees to bring them up to speed with the new technologies and how to operate in new and efficient ways, and we're looking at our processes and how we can reimagine them.

In this age and we do all of that with what we call our integrated digital core. So this is where we work with partners

core. So this is where we work with partners like Amazon and with others to build out a core digital platform that manages our data at scale and it manages all of the AI workloads that were coming in that are coming in.

And throughout all of this we're constantly looking for ways to continuously improve our processes, improve our people, give them more capability, give us more efficient processes, and all of this is through a lens of responsible AI.

So we started our AI journey long before generative AI and chat GPT. We've

been doing this for 7+ years.

We started doing a lot of computer vision and machine learning.

And when we first started this, we stepped back to look at it with a strong governance and responsible AI as a model. So we

built out a cross-functional team, this council of leaders from across all of our functions from HR, finance, IT, across the board, and this council is responsible for setting the governance, setting the policies on how we move forward with technology to apply to our business to benefit our people and benefit the business.

And so with that foundation in place and those imperatives, this is where I wanted to give you a little bit of a background on some of the AI journey that we've gone through with AWS over the last few years now, so. We'll start

so. We'll start back when Gen AI really began to take off.

Amazon introduced Quick Q Business, so we started out there building two primary assistants to facilitate our operations group, our intelligence shop floor system, and our IT group.

So with these we were able to pull in knowledge sets, knowledge bases of in the operations case, machine manuals on our production lines, uh, operating procedures, policies, things like that in the IT assistant space, similarly, it's policies that are applied.

Standards on how to do architecture, how to do, uh, request services, all of that was built very easily with Q Business, and it has enabled great operational efficiency. And one

of the keys is, as I said, we have 140,000 employees across 30 countries.

Everyone is able to interact with these in their native language, in natural language.

So the manuals, the procedures, they may not all be translated into everyone's language, but these assistants are making it easy for them to get the information they need in their language.

And it's improving user satisfaction and getting people the information they need quicker.

From there we started moving into more of the agentic flows, and this is where we started leveraging bedrock agents in what we're calling our debug agent and in this case, what this is doing is on the manufacturing line when a when a product is being manufactured.

We know from our manufacturing execution system exactly what's going down the line, what station there is, and if an issue happens, if something doesn't work, our machines know they're feeding into our MES what is happening. They

know this error code happened. They know what product is being developed. They know what components are going in at that

developed. They know what components are going in at that station. And we've built this

station. And we've built this debug agent that is able to key in on that data when an error occurs, pull up all the relevant information, the operating procedures, the manuals, everything relevant for that product, that station, and just in time give the instructions to the operator to debug that with the most likely steps needed to move that forward and keep our line

moving through. And this has helped us reduce

moving through. And this has helped us reduce scrap and improve first past yield significantly.

And that leads us to Quick Automate, which we're here to talk more about now.

So we started probably 6+ months ago or so somewhere around there, working with Amazon on. Uh, early previews

on. Uh, early previews of Quick Automate, we've looked into, so here I'm gonna tell you about two of the use cases that we're working on for engineering and for finance.

So one is, uh, for what we're doing for quoting and scheduling when our engineering teams get in RFQs for new products or, uh, changes to products and with our accounts receivable team.

And with these, we're already seeing great.

Improvements in productivity and quicker turnarounds, and we're going to talk a little bit more about those in detail in the next couple of slides here.

So first for the the quoting and scheduling agent for our engineering team.

So here the challenge is that engineers are making avoidable errors due to inconsistent access to lessons learned.

So we're developing product across all those industries we we talked about up front.

And there are consistencies and there are similarities in things that we are learning across those industries and as we build products we capture our lessons learned, we capture best practices on how to build those types of capabilities, how to do those types of manufacturing processes.

But when a new request quote comes in a request for a quote comes in.

It's hard for an engineer to know the vast universe of everything that's happening globally around the world. Where are all the knowledge base, all the knowledge bases with all the lessons learned.

So our idea was to build an intelligent agent that can get the contextual information from all of our lessons learned, all of our best practices, review through these requests for quotes when they come in, and match up what's most relevant relevant for them contextually and provide it to them just in time.

So the agent is keyed into the project management system. So as an engineer is assigned a task, the agent will automatically at that point in time, read through the RFQ, find relevant information, and provide it to them. So when,

as soon as the employee knows a task was assigned to them, they already have all the relevant information to begin that design.

And so with this we've been able to use Quick Automate to build this agent that gives them a conversational interface and embed into that workflow for that just in time insight, and we've been seeing great productivity gains, reduced errors, and less scrap.

And on the accounts receivable side for finance here, accounts receivable processes, everyone here has finance teams that are doing these.

These are manual labor intensive people don't really like doing them.

They're having to go through all of the open invoices.

Communicate with customers, receive back massive amounts of unstructured emails or attachments, parse through them, go back and update our systems. So our idea was an intelligent agent that can facilitate that email process back and forth. It can send emails out to the customer with lists of their open invoices, and it can receive them back.

Understand the emails, extract out the details and send it and uh update our back office systems or interact with our finance team.

And here we're averaging a little over 12,000 emails per month and we've seen great efficiency gains with this agent because previously that was all the accounts receivable team performing those actions and we've been able to optimize our headcount and have that accounts receivable team working on higher value work now.

And so in a second we'll take a look at a demo, but before we do that, I did want to just key in on a few quick learnings that we've, uh, captured as we've been building out these with Quick Automate and really for building agentic flows in general.

And the first one is to start small.

So think big, think what, what if you go back to our key imperatives, value first. What are the big values

value first. What are the big values that will move the needle for your business?

Look at those processes, but break them down, start small, iterate towards the end goal.

And with that, that's where I'll move on to the next one for reimagined processes.

Look at the processes you have that are well documented that you really understand, as John mentioned earlier, they have the agent that you can feed in your operating procedures and it will build out that scaffolding for you initially. So

where you have well defined, well understood processes, this can already be an accelerator.

But don't just automate what you're doing today.

There's new technology, new capability here.

Think how you can reimagine that process, make it more efficient.

And to do that you need to understand the capabilities of AI.

As we all know, they can make mistakes.

They're not 100% deterministic.

So as you're breaking down your process, look for the areas where accuracy is paramount, where you must have consistency and accuracy, and you must get complete answers and provide tools to your agents.

Fall back to traditional ML models that have higher confidence.

Use deterministic uh algorithms that you have, whatever it is, don't think that the agent can just do everything, understand its limitations and use your existing tool sets, um, just like the employees would let the agents have the same tools.

And finally, human in the loop, work with your enterprise, understand the risks of these tools, the risk of your process, and make sure you're setting proper guidelines for when humans should be in the loop on a process.

All right thanks Brian that was great great overview thank you so now I think what we're all here to see is we talked about the accounts receivable agent but Brian has also recorded that actual process and the automation that runs that accounts receivable process and we're gonna show it here today.

So some of the data is, you know, not, uh, not real, you know, so it's, it's not uh actual invoices that Brian's team is processing, um, but it's the actual workflow that's being run at Jable and saving all this time.

um The other thing that you'll see that just keep this in mind, right? I, I had no idea Jable is this huge company.

But they're getting 12,000 emails every single month to their accounts receivable team that basically has invoice information. Some of it's attached, some of it's written

information. Some of it's attached, some of it's written in text.

It's not programmatic at all, and someone was manually parsing all these 12,000 emails trying to figure out what's been paid, what's past due, who do I need to go collect from, what's my, it's just amazing, amazing, not in a good way, but amazing now that we have an agent process. So

let's jump in.

Yeah So let's take a quick look here. So

look here. So we'll start this off traditionally what's been happening is our accounts receivable team will go through and put together a list of all the open invoices that are approaching due dates or have um maybe already past due, so they'll pull together this list for all the for all the customers that have these, uh, this information.

And in this early demo here, they'll put it into a spreadsheet.

And what they'll do Is then the agent will review this list.

And begin processing through all of that information.

To oops, I think it might have jumped ahead a little bit there.

But it will begin processing through grouping those customers, collecting all of their invoices.

And Oop, sorry, it did jump ahead, sorry about that. This

that. This is, yeah, one, there we go, yeah, all right, um, so here it will send those emails out to the customer. It'll

collect the invoices for the customer, attach them so they have a single email per customer providing the list of all their open invoices.

And then they will be able to respond back to that, um, and we will have this agent that is able to here we can take a look at the, um.

The action that we set up for this agent, so it is able to go and we're telling it what its job is so it's an accounts receivable agent.

Its job is to read emails, read attachments, and extract out the information relevant for the invoices.

So maybe Brian let's just pause here um so what you're what you're kind of seeing behind this model is the quick automate canvas you can see all the steps and all the actions that Brian and his team have put um into that canvas and what's happening is the the workflow is a series of deterministic and agentic processes and steps and so they have one process that's reading the inbox

queue. They have conditional logic

queue. They have conditional logic that's processing the distillation of that email into what task is this email? What's

what's the category, right? Is it a payment?

Is it a past due? Is it a you know X Y Z?

It's going to categorize those emails and they use a bedrock agent today to perform that task and that's kind of that agentic prompt that you just saw.

But Brian and his team, um, because they were early, they were kind of before we added our inline process agents, so we're excited to kind of work with them to convert that bedrock agent into an inbuilt agent directly within the workflow.

If you saw the prompt, it's probably small on the screen, but it's very much an agent prompt. It'll make it easier for Brian and

prompt. It'll make it easier for Brian and his team to then attach tools directly to that agent rather than adding as many steps to the workflow.

But the beauty of this, you know, sort of visual canvas is that you get a representation of what's happening in your business process so you can see everything that's coming in, what are the inputs, what are the outputs, how is it correlated? You can actually transform the inputs to your agent so that they are structured and then you can transform the outputs into structured outputs as well, right? So most of the agents respond

right? So most of the agents respond in tech. That text is not

in tech. That text is not super useful to then make an API call or to save to S3 or to update a database and so you're able here to directly tell the agent and specify a schema even for your output so that that can then be strongly typed. It can be programmatic.

strongly typed. It can be programmatic.

It can be expected on the next step of what that output shape is going to look like really makes it easy to apply. Input

agents into these processes. Yes,

so the other thing Brian's team can do within this canvas is that you can directly test and debug your workflows.

So on the right we don't show this, um, but at the very top there's a debug button that'll pop up a little debug panel. You can run the agent, the UI agent, the

panel. You can run the agent, the UI agent, the full workflow, depending on your goal. Look at debug logs and then eventually publish that and deploy it to production.

So, uh, Brian, amazing demo.

Um, the thing that I think is also shocking is, uh, you were telling me before we came on stage, what, what's the actual impact of this accounts receivable agent to, to Jable and your, your business?

Yeah so if we take a look at these, these actually these first two agents that we built for the quoting and scheduling and the accounts receivable, we've been able to, um.

Across the two, improve our quote accuracies, we've seen a reduction in scrap for about 10% from the quoting and scheduling.

And we're estimating about a $400,000 savings for the for the year across the agents, so it's been a significant benefit to JBel, and these are just the first agents we've started building, as John said, we've already started using the new capabilities. We

were early on with some of this, so we're starting to look into the inline agents and multi-agent setups.

And we are already working on the next couple of use cases internally. Yeah, that's

amazing Brian.

Thank you so much for being a partner.

You and your team, I think, are, are gonna get the award for most bugs filed and logged, uh, but also for really influencing the product roadmap. We appreciate you being along the journey with us. So thank you. Thank you, really appreciate it. Thank you everybody.

appreciate it. Thank you everybody.

All right, so, so just to wrap up, um, a lot of exciting things that we talked about today we, uh, started with a quick overview of QuickS Suite, pun intended.

And then we deep dive into all of the automation capabilities within QuickSuite, starting with newer covering flows, enabling business users to self-serve and transform how they complete routine daily tasks. And

then we spent quite a bit of time with Brian and Jabel talking about how they're using Quick Automate to use to harness multiple agents to solve and automate some of their really important back-end business processes.

So it's been a fun session. I hope you've enjoyed it, learned something from it.

This QR code is a getting started guide.

Both Quick flows and Quick Automate are within QuickSuite, so you'll sign up. There's a free trial. You can start for free, 30 day free trial, um, easy to get started if you have an AWS account.

It's even easier. You can find QuickSuite from the AWS console and so excited to hear how you're using it, excited to learn your feedback.

There's more information in the expo as well. Um,

so you can find the QuickS Suite booth and the Quick Automate booth, um, in the expo. So

if you wanna get your hands on and actually, you know, type a prompt into QuickSuite and see what the, the agent's gonna give you back as an answer, um, go, go and check that out, um, but with that I wanted to thank everyone. Thank

you for spending some time with us today. Thank you for attending and thank you for coming to reinvent.

Really appreciate it. um, have a great afternoon and thank you.