YT YouTube2Text

YouTube2Text → Transcripts → AWS re:Invent 2025 - Kiro: Your agentic IDE for spec-driven development (DVT209)

AWS re:Invent 2025 - Kiro: Your agentic IDE for spec-driven development (DVT209) — Transcript

by AWS Events · 9,565 words · 1,404 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:01All right, two one. Woo! All right, we
0:03can finally start. But even though you
0:05guys have been here for like 15 minutes,
0:07it feels like already. So, you must be
0:09really excited about this session. But
0:11if you've already been using Curo, I
0:14don't know how much of this is
0:15necessarily going to be net new
0:16information because well, if you're a
0:18practitioner and you're playing with it
0:20daily, you're going to get a lot of
0:21information already. Also, I bought
0:24these shoes today and if I fall, I want
0:26my seesat scores to be like super high.
0:28Okay, I need the sympathy vote.
0:33All right, so uh my name is Rory
0:36Richardson. I am joined by Vitir from
0:39Delta Airlines and Sam who's a solution
0:42architect focused on Kira. And what
0:46we're going to do is we're going to talk
0:47a little bit about like the preface of
0:49the where how we got here in the
0:51evolution of agentic AI. Then we're
0:54going to talk about the importance of
0:55spec driven development in relationship
0:57to vibe coding. Um we're going to do a
1:00quick introduction to Curo and then
1:02we're going to get some real world
1:04tactical uh experience by talking about
1:06how Deva Airlines has been incorporating
1:10Kuro into their software development
1:12life cycle and eventually their AI
1:15development life cycle. Um and then
1:19lastly we're going to wrap up with a
1:20demo. Uh the demo is uh coming in pretty
1:24hot. As you know, we've made a lot of
1:27announcements yesterday morning in
1:28Garmin's keynote. Did you all see
1:30Garmin's keynote? Raise your hand. Yeah.
1:33Was it good?
1:34>> Yeah. Um, one of my favorite parts, um,
1:37is the autonomous agents that we're
1:40introducing because it greatly expands
1:42the context and the ability of, um,
1:47combining multiple agents together in
1:50order to do more and more complex tests.
1:51It's it's it's it's a very foundational
1:54concept that I think is going to
1:56continue to grow over the next six
1:57months, year. I think more and more
2:00folks are going to do more and more
2:02complex things uh with the agents
2:05themselves and combining them to for the
2:08autonomous agents to be able to
2:10orchestrate, monitor and even verify the
2:13work of other agents to build them
2:14together.
2:16So let's talk a little bit about the um
2:18agentic mindset because um there are a
2:22few things that we've learned like over
2:25probably the last two years or 18 months
2:27when I would talk with a lot of
2:28customers um usually they would come
2:32with an idea or a business problem that
2:33they wanted to solve but unfortunately a
2:36lot of folks started with like really
2:37really hard problems and things that
2:40were customerf facing and things that
2:42were about the core competency of the
2:44business like great we're going get AI
2:46to do the very thing that we sell to our
2:48customers.
2:50Uhhuh. The problem is that's taking on
2:54the most risk, right? And especially if
2:56you haven't done one before, why take on
2:58the most amount of risk? Instead, the
3:01companies that we've seen to be the most
3:03agentic in nature are starting with the
3:06muck. Um, I was talking to this business
3:09consultant a year and a half ago from
3:12Boston Consulting Group. He said, "Rory,
3:14generative AI is fabulous. It can make
3:16music. The problem is nobody wants it
3:19to. What they want is for it to go do
3:21the dishes so that you can go off and
3:23make music." So, where we're seeing the
3:26most exponential value out of um the
3:30application of dive AI and agentic AI is
3:34removing toil and muck to free humans up
3:37to be human. No, it means a lot of
3:40developers are like, "Oh my god, AI is
3:43coming for my job." How many thinks AI
3:44is coming for your job? Raise your
3:45hands,
3:47right? Yes, you
3:51I don't uh Gartner said it best when
3:54they they said that
3:58AI is probably not coming for your
4:00developers, but a developer using AI is
4:03probably coming for a developer that is
4:05not using AI. And I think that's kind of
4:08a true thing because a lot of agents I
4:11kind of view as um various ages of
4:14children, right? You know, they have
4:16really great ideas like, "Oh, let me
4:17help you. Let me help you. Let me help
4:18you. Let me help you." And it's like,
4:19"No, I don't trust you to touch the
4:20dishes. You're [snorts] too young." So,
4:22how we are dealing with the agents
4:24depends on the level of sophistication,
4:26how much we trust the information that
4:27we're getting back from them before we
4:29incorporate it into what we're doing.
4:31because you're not going to fire AI, but
4:33I will fire a developer if they, you
4:35know, write bad code or malicious code
4:37or introduce risk into the application.
4:40Yeah, these are all things. The humans
4:42are still responsible, y'all. This is
4:44super important.
4:46But can a developer be augmented? I
4:49mean, think about all the the muck that
4:51you do in your software development life
4:53cycle, right? Who loves Jira? Raise your
4:56hand. Yeah.
5:01>> [sighs and gasps]
5:03>> Jira is something you
5:06but being able to um search through Jira
5:09automate you know Jira tickets and being
5:11able to optimize that experiment and
5:13will give everybody else here a better
5:16day because it's not when you're messing
5:18with Jira you're not creating that new
5:20application because you know when you
5:22first started being a developer you
5:23wanted to do you wanted to make cool
5:25stuff right so everything that is not
5:28helping you make cool stuff muck.
5:34So, quick history lesson. I know you're
5:37are all on this journey uh with us, but
5:40um we started off with plugins into the
5:42existing IDE, VS Code, you know, Jet
5:44Brains, Intelligj, and that seemed like
5:46a really great idea. And then we got
5:47this idea like, wouldn't it be cool if
5:49we made agents that go off and do
5:50bespoke tasks within the software
5:53development life cycle like slreview or
5:55slashdoc. This was great except we were
5:57teaching a bunch of people how to work
5:59with agents and not teaching the agents
6:02to work with the people.
6:04I mean, how many of you have a um a
6:06smartphone like an iPhone or something?
6:08Yeah. And do you remember that you uh
6:11you know were told specifically you had
6:13to use that device, how to optimize it,
6:15and then you had to take a you know an
6:17enablement session to fully use it?
6:20Of course not. It's just you just get
6:23started and you get to choose whatever
6:25device that you want. That's the
6:26developer experience, right? So instead
6:28of teaching um developers or anyone how
6:31to interact the technology, we should
6:33have the opposite approach. And one of
6:35the ways we're able to do that is to
6:36have a an agentic AI approach to
6:40software development. So now instead of
6:42having just plugins into the IDE, we've
6:45seen the rise of
6:47um agentic idees and this was great. I
6:50mean, vibe coding's cool, right? You
6:52just get started. How many's had to
6:53debug one of your colleagues uh vibe
6:55coding projects?
6:57Yeah, super fun, right?
7:00But it's also kind of an anti-attern. I
7:02mean, first of all, the democratization
7:05of people being able to create something
7:08just through natural natural language is
7:10it was the dream of low code, no code.
7:12It's always been the dream like how do I
7:14can I just prototype this? How do I just
7:16get started? It was great. But then how
7:19do I work with other developers to do
7:21more and more complex things? And that's
7:23combining spec driven development with
7:26bi coding to accelerate how are you
7:31getting something you know uh uh uh from
7:33prototype into production.
7:35The other change that we've seen um this
7:38year is the rise of protocols for
7:40agentto agent communication. How many of
7:42you people are using MCP servers? Right?
7:46Wasn't that way last year? I mean this
7:48is a radical change in how we're
7:50thinking about the very architecture of
7:53agents themselves. So earlier this year
7:57I was like well what is the definition
7:59of an agent? Because if you look on the
8:01internet no one uh could agree. Like you
8:04get you know three people together they
8:05would have all different definitions of
8:07what an agent was. So we're going to go
8:08with my definition because I'm the one
8:10with the mic.
8:12Um I started out with well it has to be
8:14able to perform tests asynchronously. It
8:16has to go off and come back. It has to
8:19have some degree of autonomy.
8:21It has to have agency. And I mean that
8:23in a very pride and prejudice sense,
8:25right? It has to be able to make some
8:26decisions on its own. But this last
8:30idea, this last idea of adding the word
8:32atomic is fairly new and the most
8:35controversial. And yeah, they all start
8:37with a. I did that. Um,
8:40atomic is
8:43meaning that you're able to make more
8:44bespoke agents and then combine these
8:47bespoke smaller agents in different ways
8:50to do more and more complex things. So
8:52instead of using a really really large
8:55model that has been trained on a really
8:58really large data set, what if you could
9:00do something that was more bespoke? So
9:03instead of being good at everything,
9:05what if you could be great at a few
9:07things? The idea of making atomic agents
9:11kind of brings me back to my
9:12microservices days and the idea that you
9:15can combine microservices together to do
9:18more and more complex applications and
9:19also reuse those agents. This was sort
9:23of the foundational concept bes behind
9:26um what we're trying to do with
9:27autonomous agents, being able to use the
9:31right agent for the job, but not
9:33necessarily having to teach you which
9:36agent that is and when to call it. These
9:38autonomous agents have the ability to
9:41help make those decisions for you. Now,
9:45this kind of feel I I love uh a lot of
9:47sci-fi. Uh, I'm a huge sci-fi fan and it
9:51feels almost like what we really want is
9:53Jarvis, right? No, don't go full Ultron,
9:56right? But in Iron Man 1, Jarvis was a
10:00natural language interface for IoT
10:02devices in Tony's house. By the time you
10:05get to Iron Man 3, Tony is able to go to
10:06the holiday and go directly into the
10:09most humanistic areas of innovation.
10:12He's able to just start to vibe code the
10:15next Iron Man suit. And then when he's
10:18done, he's able to look over at Jarvis
10:20and said, "All right, now go off and
10:21build the Iron Legion." Because Jarvis
10:23is handling all the supply chain
10:25management um of all of the resources in
10:28the house, um managing all the design
10:30specs, managing search within, um the
10:33libraries of Tony's work. I mean, can
10:36you imagine what the world is going to
10:38be like when your kids have access to
10:40that level of assistance for anything
10:44that they want to make? That's the world
10:46that we're living in. I mean, I have a
10:4915-year-old and I've been trying to
10:50teach this kid to code since he was
10:51four. Uh, I can't get those 10 years
10:54back because what is the role of Python
10:57by the time that kid hits the job
10:58market? I mean, the syntax is kind of
11:02taken care of. I'm not saying computer
11:04science is dead by any stretch of the
11:05imagination, but the world has
11:07dramatically changed for how our
11:09children are going to interface with
11:10solutions and innovation. And it's not
11:12about agents per se. It's about how fast
11:16and how hard can they dream because
11:18they're going to be able to create it
11:20with the assistance of these tools.
11:25So we created an agentic IDE like hero
11:29that married the idea of vibe coding
11:33with traditional specdriven development.
11:36This also changed a lot of the roles
11:38within our two pizza teams at AWS. It's
11:41been highly disruptive because now we
11:44move much faster when we embed our PMs
11:47into those teams because we're able to
11:49iterate on the specs much much faster.
11:52So engineering reaches a problem they
11:54were like hey you this isn't going to
11:55work PM takes it and goes yep let's
11:57rewrite you know the direction of what
11:59we're going entire process is hey it
12:01takes a matter of hours and not weeks
12:05this level of speed has created other
12:08problems because the number of code
12:10commits that we're able to do test
12:13wasn't ready for QA wasn't ready for
12:15there's downstream impacts to the speed
12:17that we're able to get things done now
12:21So having an AI ID IDE has greatly
12:25accelerated well everything that we
12:27touch at AWS because and at Amazon you
12:32can view Amazon as like our first best
12:34worst customer because they know where
12:36we sit um and they come and they give us
12:38direct feedback and we have to compete
12:40for their developer time as well um in
12:43order to figure out like what is the
12:45tools that we should be creating
12:46internally and so that's our own
12:49internal test bed in order to refine
12:51like some of our thinking and what we
12:53decided to build.
12:57Also, we found tremendous
13:01uh uptick of the command line interface.
13:04The command line interface we found a
13:07lot of people were using it or
13:10weren't necessarily having an IDE
13:12already installed on their computer.
13:15That means this is a different type of
13:17user that is using the command line
13:18interface to do different tasks but also
13:21these tasks are associated with the
13:23software development life cycle. So we
13:26incorporated the command line interface
13:28also into Kira and so now Kira is also
13:30available in the terminal.
13:37This evolution of agentic AI has changed
13:42everything about what we're doing. Um,
13:46and it continues to accelerate exactly
13:50how we're creating software, not only
13:53from just a traditional application
13:56development, but also how we're creating
13:58the agents themselves. Now, I got my
14:01roots in microservices and breaking up
14:03monolithic architectures into
14:05microservices.
14:06I've been talking with customers lately
14:08that aren't doing that anymore because
14:10that's so 2022.
14:12Now they're thinking about breaking up
14:14monoliths into true agentic architecture
14:16and bypassing that step. So the
14:20foundation of these applications are
14:22agentic. And so the process of creating
14:24agents has fundamentally changed and
14:27accelerated. You'll hear a lot about the
14:29tools that we are developing to enable
14:32that whether it's a framework like
14:33strands SDK or an agentic platform like
14:36agent core that gives you the building
14:38blocks for you know runtime and memory
14:41but also choosing the right inference
14:44for your job.
14:47But the thing that I think is the most
14:49interesting is data. Because now if
14:52you're able to create an agent that is
14:54directly responsible
14:56for your data, the responses that you're
14:59going to get are more bespoke to your
15:01environment. So the ability to create
15:04these agents in combination with the
15:06full tool chain, well, I'm really
15:09excited to see what's next and what
15:11people are going to create. I've been
15:13working really closely with Delta
15:15because Delta uh has a mindset that is
15:19agentic architecture in nature. Um when
15:22they started sharing some of their
15:23architecture diagrams, I was super
15:25impressed with how they were thinking
15:27about those autonomous agents agents and
15:31what do you call them? Nano agents or
15:33sub aents?
15:35>> Sub aents.
15:38And it made me feel really old because
15:40in my architecture diagram there was a
15:42load balancer, there was EC2
15:45and theirs is different. So they're
15:48trying to push the envelope of what's
15:50possible in the safest possible way that
15:54scales within an enterprise
15:55organization. So with that, I'd like to
15:58uh invite Vid here to talk about their
16:00experience and how they've applied it
16:02directly to Delta Airlines. Come on up.
16:06You are not going to catch me. I swear
16:07we're not going to trust fall. I'm going
16:09to be fine.
16:10>> To my word,
16:12>> I'm here to catch you.
16:13>> No, it's going to be fine. No.
16:15>> Okay, cool.
16:18[laughter]
16:20>> Before I begin,
16:23how many of you flew on Delta to reach
16:25here this amazing conference?
16:29That's great. Thank you for your
16:31business genuinely.
16:34And for those who use the free Wi-Fi
16:37in our flights to finish the last minute
16:40work or prepare flights like me, you are
16:43welcome.
16:45Delta is the first airline in US that
16:49introduced free Wi-Fi across our
16:51flights.
16:53We are the pioneers of innovation and
16:56technology in entire airline industry.
16:59100 years of legacy and history that we
17:03created.
17:05Now decisions like this are made behind
17:08a philosophy that which I would like to
17:11share with you all uh is
17:14we'd like to create experience for our
17:17customers the best experience possible
17:19for our customers and make it as the new
17:22standard
17:24and execute it every day and that is the
17:27philosophy that we follow in every
17:30domain and every group in Delta even in
17:32developer experience which We lead and
17:36serve for our de del delta developers.
17:38We follow the same philosophy. We would
17:40like to provide the best tools and
17:42technology for our developers so that we
17:44can reduce their friction and improve
17:46their innovation every day. Rory shared
17:50great vision on the agent roadmap that
17:54AWS is getting into and it is ali it
17:57aligns very much with our vision and the
18:00road map that we are planning as well.
18:04How many of you had seen slides like
18:07this and started preparing mental
18:10mapping of how to adjust my road map,
18:13which things I need to prioritize and
18:15dep prioritize? I'll raise my both hands
18:18because every tech conference that I
18:20visited, every exec we reached out to,
18:23whenever they shared something,
18:24immediately we started mapping how to
18:27adjust, which changes we need to make.
18:30But here is what we learned that
18:34we can achieve this autonomy agentic
18:39architecture by building on trust.
18:42The trust from our developer community
18:46with the rollout of Q developer on the
18:48left hand side the coding assistance Q
18:51developer is fully adopted in entire
18:54delta. Almost every developer uses Q
18:56developer and they are giving great
18:58feedback. It laid that foundation in the
19:01middle. what we have Kirro IDE the
19:04agentic IDE which we are scaling now
19:07across the delta enterprise and we are
19:10marching towards the autonomous agentic
19:14systems which are going to be integrated
19:16into our developer workflows
19:19and with the help of Q developer so far
19:22what we gained was we're able to capture
19:25the major wins that Q developer provided
19:27and also helped us to understand what
19:30are the areas of improvement what are
19:31the automations that we can build for
19:33our developers.
19:37So to increase this adoption and get to
19:42the stage that we want, we created a
19:44model, right?
19:47One thing that we didn't tried was we
19:50didn't try to convince all of our
19:52thousands of developers in one session
19:54or at a go, right? It is not possible.
19:57At the same time we are not
20:01uh staffed up to reach out and talk to
20:02each and every developer. So what we did
20:04was with the support of our leadership
20:07we reached out to all of our domain
20:09leaders and asked them nominate some of
20:13you the champions who are technically
20:15and technologically very sound and very
20:19trustworthy and they can influence the
20:22developer community with without any
20:25title and authority. Right? with the
20:27trust.
20:29So with the help of these dev champions,
20:32we are able to
20:34scale this model. We went domain by
20:37domain. We trained each domain and also
20:40we built the train the trainer model
20:43because these days everyone knows AI
20:45right uh when Rory was sharing that the
20:49technology Jarvis like technology how
20:51it's going to help for children uh in
20:54future that immediately reminded me a
20:57personal experience at my home as well
21:00few weeks back I was I have a 7-year-old
21:02daughter so I asked her just casually
21:05two simple math problems
21:07multiplications. I gave her try these
21:10out. So she took okay I'll try. Five
21:13minutes back she came back and asked can
21:17I use chat GPT?
21:20I'm I'm not making this up. It really
21:23happened. I was I was really shocked. I
21:25was like
21:27I'm really surprised. Where did you
21:28heard about charge?
21:30I was expecting maybe you're going to
21:32ask me can I use calculator to do do
21:34this and she was like what is a
21:37calculator
21:39see the AI advancements are happening
21:42very fast that everyone knows how to use
21:44AI how to adopt them but the most
21:47important thing is inputs to this AI the
21:51inference that we're going to provide to
21:53this AI because
21:56everyone knows like I said everything AI
21:58is an amplifier
22:01So the more structure prompts, the more
22:03context we provide, the more guardrails
22:06we introduce, we're going to get the
22:09best output that everyone is looking
22:10for.
22:12At the same time, if we have chaos, it's
22:15going to amplify that chaos. So that's
22:17why we focused a lot on setting those
22:20inputs right, getting our community
22:23ready so that we can scale this model.
22:28Now these dev champions this is not
22:31their primary job. They have their day
22:33jobs. They need to work in their sprints
22:35and scrum models. So when we reached out
22:38and formed this community we clearly
22:40said we wanted to clearly set the
22:42expectations because we want to approach
22:46this outcome based right. So we clearly
22:48said that this is what we expect from
22:50you.
22:51We want you to participate in the
22:54initial session so that you can know
22:57learn how to train how to train your
23:00teams. We would we need you to
23:02participate in our office hours so that
23:04you can help others and also listen to
23:06their questions and we need you to
23:09contribute your prompts any prompts that
23:10you worked and you need we need you to
23:13bring along your entire team not just
23:15developers your product owners scrum
23:17masters how to use Q developer they
23:21helped us to reach almost every
23:23developer in entire delta IT which is
23:28the best part that any email campaign or
23:31marketing campaign could do because
23:33nobody reads emails. That's the fact,
23:36right? So using the dev champions, we
23:39were able to establish that community
23:41and able to get scale this model and
23:45increase the Q developer adoption.
23:48But
23:50we started seeing that developers are
23:52writing code. We started seeing that uh
23:55more we're writing more code much
23:57faster.
23:58But still we're not able to see the
24:01result. What is the end result? What is
24:03the outcome that is aligning with our
24:05business expectations?
24:07So this is the pivotal moment which is
24:09the very important part. The feedback
24:12that we got from our developers is
24:16it is very important to understand that
24:19moving from help me write this function
24:22to help me build this feature.
24:26We need AI that understands the business
24:28context, our governance, our compliance
24:31rules, our integrations.
24:33It needs to understand all of those.
24:36Then only we can get to the stage of
24:39delivering the feature level output.
24:43That is when KO entered into our story.
24:46So we partnered with our AWS team and we
24:50are one of the early pilot users of KO.
24:53We introduced it to our again our same
24:56champions community and also some power
24:58users of Q developer
25:01and we asked them go ahead and use Kira
25:03and start giving us your feedback
25:06because we already built the trust right
25:08when we rolled out the Q developer they
25:10we started to gain their trust and they
25:13started giving us responding the
25:15feedback in the form of feedback four
25:18main themes that uh evolved from from
25:20their feedback They clearly said that
25:24using Kirro versus any other tool, we're
25:27able to see that it can understand our
25:30complex code base, our business rules,
25:33our existing logic and extract it and
25:35build the context that it is expecting.
25:39At the same time, they clearly said that
25:41it was able to break down the given
25:44requirement into easily implemented
25:46design and executable task. And most
25:50important thing we started seeing that
25:53ko with the help of kro we are able to
25:55pro build production ready code I want
25:59to underline that production ready code
26:02no two enterprises will have the same
26:04set of governance and compliance rules
26:06to build their products. So it is very
26:10important a tool to understand establish
26:12the context and adjust and deliver the
26:15code that is production ready and
26:17readily to deploy into the higher
26:19environments which we are able to see
26:21using KO.
26:23We also heard feedback from our
26:25architects and some of the developers
26:27that they were able to use KO
26:31to be a critic in their architecture
26:33decisions. They use some of them use
26:35that as a soundboard to validate that is
26:38my design and approach is it correct or
26:40not. That is the most exciting part.
26:45So we started seeing that Kirro becoming
26:48one of the team member who has this
26:51beautiful memory of memorizing the
26:54thousand lines of code establishing the
26:57context at the same time giving you
26:59advices and suggestions which direction
27:02you need to go for your next product and
27:04next implementation.
27:06That is the most exciting part we
27:08started seeing and we started realizing
27:10with the help of Ko
27:15and the specdriven development is the
27:17most unique and innovative feature that
27:20we came across in KO
27:25simple [snorts]
27:25it's it's not that complex three simple
27:28step one lock your requirements you give
27:31your detailed requirements
27:34step two KO automatically breaks down
27:36your requirements and start building the
27:38design, generating the design
27:40and step three it will also give you the
27:43tasks giving power to the developers.
27:46That is the most important part. Human
27:48in the loop
27:50ultimately developers has the power
27:53whether they accept the design or
27:55they're going to execute those task or
27:57they're going to give refine or alter
27:59any any any direction that Kroll needs
28:02to implement those. So we started seeing
28:05that level of flexibility at the same
28:06time context awareness and the most
28:09important part
28:13the backlog grooming.
28:15How many of you familiar with this?
28:19I see a lot of nods. Everyone right?
28:21Backlog grooming. Several hours of
28:23backlog grooming, lot of debates about
28:26story points, definition of done,
28:29acceptance criteria and a lot of
28:31feelings get hurt and so much so much
28:35drama, [clears throat] right? So what we
28:39started to this is the thing that we
28:41never expected when we introduced Kira
28:43to our pilot users. They start they
28:46transformed this backlog grooming days
28:48into kospec design days. Kirro spec
28:52design sessions. So they used that time
28:55to refine the requirements. They started
28:57with Kirro instead of backlog grooming.
29:00They started with Kro with their
29:01requirements and gave Kirro the prompt
29:04in a way that this is the format we
29:06create our features and user stories and
29:09this is the this is our model. We would
29:12like you to create accordingly. They
29:14started seeing those results and those
29:17sessions because Kirro can remi remember
29:20the context and it can automatically
29:23adjust based on your codebase.
29:26The amount of time even to do the uh
29:29spec refinement is also getting shorter
29:31and shorter
29:33and
29:35surprisingly no one complained or
29:37petitioned to bring back those backlog
29:39rooming session.
29:41We'll take that win.
29:44So
29:46with the pilot, right, I'm I'm not
29:49talking about scaling the KO. We didn't
29:50scale the Ko. We get getting ready to
29:52scale the KO with just the pilot users,
29:55right? We're able to deliver
29:58the first ever most intelligent
30:00developer portal end to end production
30:03ready,
30:04meeting all of our compliance and
30:06governance rules.
30:10So I'm sure many of you are familiar
30:13whenever there is a business group with
30:17the product owners and there is a
30:18technology counterparts
30:20how [snorts] much our handoff happens
30:22how many our meetings we go through our
30:25business product owners they will create
30:28the mockups the requirements word
30:30document detailed word documents and god
30:33help us some excel sheets with all of
30:35the requirements
30:37Right?
30:38But still even after all of this back
30:41and forth there will be some gaps still
30:43whenever the end result on the end
30:45product is delivered.
30:48So what we wanted to try out was
30:50recently we trained our business product
30:54owners how to use Kira. These are the
30:58people who never coded a single line in
31:02their life. All they did was word
31:05documents mockups. So we trained them
31:07how to use Kirro.
31:09The end result they were able to provide
31:13or generate working prototypes aligning
31:17with the road map aligning with the new
31:20feature that they want and they were
31:22able to easily hand off that working
31:24prototype with their technology
31:27counterparts who are easily in ours
31:31actually integrate with their working
31:34application.
31:36Again, I'm not making this up. These
31:39business product owners said that
31:42the prototype or the design usually
31:46takes us a quarter, one increment to get
31:50that and we're able to complete it in a
31:54matter of days and weeks.
31:57It's not only about the time we are
31:59saying saving it is also about
32:03fixing those gaps right we started to
32:06see that the gap between the technology
32:09team and the business product teams
32:11getting disappeared because now everyone
32:13started to see and talk everything in
32:16the same language. No more assumptions,
32:18no more imaginations.
32:20product owners are able to give the
32:22working prototypes and technology
32:23counterparts able to immediately
32:26visualize and integrate the actual
32:28application
32:30and like I said production ready code is
32:33getting delivered every every time
32:37and I we sh I shared you what we
32:40delivered at the same time I would also
32:41like to share what our developers were
32:43saying right highlighting here three
32:46quotes from the feedback that we got
32:49from our pilot user users. One of the
32:51engineering leader,
32:53right, who who said that I forgot my
32:56coding uh practice several years, but I
32:59gave it a try using Kirro. He said that
33:02now I can give tasks to KO and I can go
33:05back to my meetings. In the background,
33:08KO is able to continue to executing
33:10those tasks and once I'm done with the
33:12meetings and I'm coming back and
33:13refining and reviewing the code that it
33:16generated.
33:17And one of the developers, she said that
33:20specdriven development stopped me going
33:23circles with wipe coding. Anyone heard
33:25about wipe coding? Anyone? No.
33:29So it is the I I think it's the official
33:32word of the year for 2025. Wipe coding.
33:35Isn't it lot of prompting, iteration,
33:38prompting, screaming,
33:41lot of tears. We we think that there is
33:44a progress being made but our commit
33:46history will tell the actual story right
33:49so
33:51using kro with the specdriven
33:53development
33:55what she shared was um
33:58it is remembering the context it is
34:01giving me the destination at the same
34:03time the path to get to that destination
34:08it is not
34:10building code out of vacuum The code is
34:13be getting generated based on the
34:15context and one of the senior developers
34:18said that breaking down technical design
34:20into stories with full repo context.
34:26This is very important right because
34:28there's a lot of there will be a lot of
34:30freedom in green field development. If
34:33you're starting from scratch sky is the
34:35limit blue skies we can make a lot of
34:37assumptions lot of imaginations
34:39prototyping all of that. But if we are
34:41implementing or doing a brownfield
34:43application, if you have a code that is
34:45written several years back and it is the
34:47most important application, it's very
34:50challenging to plan any changes. But
34:53with the help of KO, they are able to
34:55create the work that needs to be done
34:58aligning with the context of the code
35:00and 94% of the pilot participants
35:03reported four and greater than four out
35:05of five. The feedback
35:10ids are like rifles to the marines to
35:14the developers. There's so much
35:16attached. It is not a simple task asking
35:19developers to try out this new IDE. But
35:22yet all of our pilot participants were
35:25because of the trust we built. They were
35:27very open and flexible and used this and
35:30started giving this great feedback.
35:36And I also shared that Q developer the
35:38success right that we achieved and this
35:42is what the trust looks like.
35:45So we launched Q developer end of March
35:49in the first week of April and
35:53in 6 months we're able to achieve,948%
35:59growth. That's not a typo. That's a real
36:02percentage. And I really wish my stock
36:05portfolio gave me this kind of returns
36:06in six months. Because of the trust that
36:09we built, the developers em are
36:12embracing the any technology that we are
36:14building. And the Q developer laid that
36:18foundations on those foundations were
36:21getting ready to scale Kirao across the
36:24enterprise.
36:26Of
36:28course you need to measure any of your
36:31any roll out any technology enhancements
36:34that you're doing. Always metrics are
36:36very important. So we created our own
36:39dashboard which helped us to uh measure
36:42what is the user adoption is looking
36:44like what is the engagement is looking
36:46like at the same time what is the Q
36:48developer impact which is clearly showed
36:51us that we're able to produce more code
36:53in in the less amount of time.
36:57And again like I said the developers
36:59what they're saying was
37:01primarily it's coming down to they're
37:03seeing a fundamental shift in our ways
37:06of working. They're no longer bogged
37:09down or tied to the implementation
37:10details because they were able to give
37:12those implementation details to Kira and
37:15let it work and they are primarily
37:17focusing and a lot of focus on
37:19architecture and design thinking and
37:22also they're focusing now to get rid of
37:25technical debt
37:28right everyone's favorite word isn't
37:31that so they're taking advantage of ko
37:33and getting rid of technical debt and
37:35lot of use cases we we implemented
37:38within this short span using pilot uh
37:41pilot group.
37:45So
37:47the takeaway that I we I would like to
37:49share with you all is from this pilot
37:52experience and from the adoption of Q
37:54developer with this agentic AI tools
37:57developer tools we are able to say yes
38:01more. Now
38:04I know many of people leaders here you
38:06might have read the book saying no is
38:08very important but now these asentic
38:10tools are helping us to say yes more
38:13years worth of backlog which got
38:15competed and deprioritized because of
38:18competing priorities. Now we are able to
38:20say yes and start solving those
38:22problems. We are able to bridge the gap
38:24between our business counterparts and
38:26technology counterparts and everyone is
38:28trying to now is able to see the same
38:31speak the same language and see the same
38:33outcome.
38:36Now the deliverables are no longer tied
38:38to the size of the team
38:42that they are not no longer
38:43proportionate. Smaller teams are
38:45starting to deliver the biggest outcome
38:48with the leveraging tools like Kira. And
38:51what's next for Delta? Like I said,
38:53we're ready. We're getting ready to
38:55scale KO across our enterprise and lot
38:58of exciting announcements, KO agents
39:00that were announced and we're going to
39:03integrate all of the Ko agents into our
39:05developer workflows and life cycle.
39:10See with in all of this with all this
39:12journey one thing that our personal
39:15observation is the companies in in near
39:19future like a decade or so they're going
39:22to win not just with AI tools but when
39:26these AI tools are people are woven
39:29together starting from the build design
39:33deployment from product owners all the
39:35way to production
39:37right We need to
39:41focus on introducing AI across the
39:43product development life cycle which we
39:46also call as PDLC because we're delta we
39:50love our acronyms. Um so
39:54introducing AI tools and automation
39:56across entire your product development
39:58life cycle is very important. Q
40:00developer laid that foundation and
40:02helped to build the trust among our
40:05developer community. Now we are taking
40:08the advantage and rolling out KO for our
40:11future.
40:13Now I'm going to ask Sam to come and
40:15show some magic how the Kira is going to
40:17look like in the ID.
40:20>> Thanks the dear. It was really good to
40:22hear.
40:22>> You're supposed to clap his bosses here.
40:24[laughter]
40:26[applause]
40:31>> Nice to hear a practical uh experience
40:33with the tool here. So I'm Sam. I'm
40:35going to show you a quick tour of Curo
40:38here uh with a demo that I put together
40:40as a video because it uh there's a lot
40:42to show uh and it uh will go pretty
40:46quick here and so there'll be some uh
40:48time lapse that I'll include. Um so Kira
40:51is uh a fullfeatured IDE. Uh so you'll
40:54see a lot of the stuff that you're kind
40:56of used to seeing in your IDE, but it's
40:57purpose-built for AI agent development.
41:01And as you open it up, you'll see the uh
41:04spec versus vibe kind of options here.
41:07Um, which is kind of new. And I'll talk
41:09a little bit more about the spec uh
41:12version of this or uh features. Uh, we
41:16have some tools included in here to help
41:17you manage your context, like being able
41:19to load files if you want. Um, grabbing
41:23output from your terminal as you're
41:24debugging, uh, pointing to particular
41:27URLs if you need that. Uh you can even
41:29include uh images um so that if you have
41:33uh UI designs or sequence diagrams or
41:36something like that you want to include
41:38you can add those uh and uh you can
41:41select the different agent models you
41:43want uh including the latest um like
41:46sonet 45 the we have opus um and there's
41:51this option for auto mode uh the auto
41:54mode will let ko automatically pick
41:58which model it's going to use optimal
42:00for that and it allows you to save a
42:02little bit on the credits as you're
42:04doing it that way. So it'll pick maybe a
42:06model that might be a little bit cheaper
42:07to use for the particular use case that
42:10it's going after. Uh on this panel on
42:12the left over here you'll see all the
42:14hero tools and these are the things that
42:16are going to really help you with your
42:18uh project manage the AI agents. Uh
42:22particularly uh we'll start with this uh
42:25the steering files component here. So
42:27steering files are a way to give rules
42:30and guidance to the AI. Uh you can have
42:33prompts that you share amongst your
42:34teams that you might uh put in there. Um
42:38coding standards, those kinds of things.
42:40Uh and it'll give Curo a map of your
42:43codebase as well. So you can click this
42:45button here and Curo will just
42:47automatically examine your uh project if
42:50you have a project that you've already
42:51built and kind of creates these three
42:54like foundational steering files for it.
42:57One that's just a a product file that
43:00describes what the product is supposed
43:02to do uh the features and functions.
43:04It'll create a tech file that uh
43:06outlines what all the um the tech stack
43:09that you're using uh and uh a structure
43:13file. So the structure gives it kind of
43:15a map of where files are, what kind of
43:17data you have, um, anything that the
43:21agent might need to know. So it doesn't
43:22have to load the whole entire codebase
43:24all the time. It'll use this to navigate
43:26through uh, your environment.
43:29It also has this feature with a a file
43:31match pattern that you can give it. So
43:33this particular steering file I have
43:35here is for my UX guidelines, and I'm
43:38telling it to only load this when it's
43:40working on HTML or CSS files. And that
43:42way this doesn't get into my context all
43:44the time. Um because too much context
43:46can confuse uh the AI agent here. Um
43:54these are just MD files. So you can
43:57source them with your project. You can
43:59share them. Uh of course with Kira we
44:02can manage uh MCP. Um but if you go to
44:06um kirao.dev dev, you'll find that we
44:09have a directory of MCPS that we
44:12recommend that are easy to just with a
44:14click you get them added to your
44:15project. And we'll take that a step
44:18further uh with something that um I
44:20think today is probably the first we've
44:22ever really talked about this yet is uh
44:25powers. So Kira powers uh is a little
44:27bit they're like MCP on steroids if you
44:31will. Uh you get a little bit more than
44:33just an MCP when you're trying to like
44:34you have an MCP file, you have several
44:36of them. you're not sure when you need
44:38them, when you don't need them. You'll
44:41maybe want to augment that with some
44:43rules and steering files. But with
44:45powers, it combines those all into uh
44:48basically one. So, you'll get uh and we
44:52have some recommended ones here as well,
44:53but you can create your own. You'll see
44:55a bigger list in the actual released
44:58version that uh will come out. Uh I
45:00think it actually was available this
45:01morning. Um, but for this particular
45:04project, uh, I'm going to build, uh,
45:06something that I'm going to use strands
45:07for. So, I can just go to the directory,
45:10include, uh, the strands power, and now
45:13I've enabled Curo with the information
45:15it needs in order to help me build this
45:19strands, uh,
45:22driven power. So, I'm going to build a
45:24project here. I'm going to show you this
45:26is Kiraosan. This is my uh, fake retail
45:29site that I built for my demo. Uh, and I
45:32want to add a um customer support agent
45:37to this using bedrock and strands. So,
45:42uh, I'll just jump into Curo here. I'll
45:44give it a prompt. In the past, I might
45:46have to be very very careful about the
45:48prompt that I'm giving uh my AI agent.
45:51But here, I'm going to do fairly simple
45:55uh because I'll get a chance to refine
45:57this and work through this.
46:01And I wish my play dude would disappear,
46:04but okay. Um
46:07so Kira will go off uh examine this
46:10prompt and create a u requirements uh
46:15document for me. This is an MD file.
46:18I'll be able to share this with my team.
46:21So a chance to review this with
46:23stakeholders uh go back and forth a
46:25little bit and iterate on the
46:27requirements for this and then uh if I
46:30need to make some updates to it, I can
46:32do that. Um, and so [clears throat] I'm
46:35reviewing these requirements and just
46:37kind of for the sake of demo, I'm
46:39noticing something missing. Uh, I want
46:41to make the window movable and
46:43resizable. So I I could do this with
46:46just adding the text in there. But in
46:47this case, I'm going to ask Kira to add
46:50this uh to the requirements so that I
46:53can then go ahead and uh when it builds
46:55it, the window will be movable and
46:57resizable. And when I'm happy with those
46:59requirements, I can just click this
47:01button here and we're going to move on
47:03to the design phase. So Kira is going to
47:06analyze what it saw there. It's kind of
47:08figuring out that uh I wanted to do
47:11something with strands. Uh and so it's
47:14going to do a little research in the
47:16background. And here it activated the
47:18Curo power. And what the the cool thing
47:20about these powers, if you're using an
47:22MCP server, it might get loaded all the
47:23time. But here this is on on when it's
47:27needed it will load this power because
47:29it knows from what it's working on that
47:31context that it's going to do something
47:33with strands.
47:35[clears throat] It'll deliver this
47:36requirement uh design document for me.
47:39I'll get some diagrams. I'll get pseudo
47:41code in here. Um and then it it does
47:44this other neat thing that's these um
47:46correctness properties. So this is a a
47:49set of properties that it uh derived
47:52from the requirements to kind of
47:55validate and tie everything together to
47:57help us get back to make sure that the
47:59requirements that we're building uh
48:01match uh the what we're what we asked it
48:04to. So it's another way for it to do
48:06some testing and reasoning to make sure
48:08that it's getting things right. And it's
48:10also really useful for you when you're
48:13reviewing the design to make sure that
48:15it covered everything. Uh, and I noticed
48:17that um when I was covering this that I
48:19would like a little bit more um uh
48:23robust I talked about input validation,
48:25but I wanted some more robust uh
48:28validation to make sure that there was
48:30no toxic content getting into my uh
48:33agent. So I I went ahead and asked uh
48:36Kira to add that. And then once I'm
48:40comfortable with the design, we move to
48:43the next phase which is uh creating the
48:45tasks for the project.
48:48And it gave me this option for if I want
48:51those tasks to be uh all included or do
48:54I want just some of them to be optional
48:56or not. And if I I chose optional for uh
48:59making the demo a little quicker. Um but
49:02you can see these things that are gray
49:04uh are optional tasks. And uh I will get
49:07an I can turn them on if I want to. I
49:10can review this and decide okay some of
49:12these I want uh and some of them I don't
49:16um [clears throat]
49:18for the demo that I'm going to do here
49:20too. I also have all these tasks and I
49:22want to make sure that they get synced
49:24up in Jira. Um I know not everybody
49:27loves Jira but uh it's a good way to
49:29illustrate what we're doing.
49:32It's a good way to illustrate the work
49:34that we're going through here. So, and
49:35also uh the power of the MCP which I
49:38think a lot of you are already familiar
49:39with but within the IDE here I can add
49:41this uh Jira MCP server and uh we have
49:46you know OOTH authentication so it logs
49:48into my credentials and now I'm wired up
49:51to my Jira account and I can go in here
49:54and just ask Curo to take all those
49:57stories and tasks that it created in
50:00those files and create stories and tasks
50:02for me in Jira. Um, so this is a spot
50:06where I definitely needed to speed up
50:07the video. It took a little bit, but we
50:09can go through and it created uh eight
50:10user stories uh and 19 subtasks for me
50:16all in Jira. And so now I can I can
50:18track these. I can distribute this
50:20throughout my team if I wanted to. Uh
50:22this is a pretty small example for this
50:25demo. Um, but if you had a bigger
50:27project and you maybe wanted to
50:28parallelize some of this work uh
50:30throughout your team, you can because
50:33here you can see everything in Jira with
50:35the descriptions and the details and all
50:37that of what the project was doing.
50:42Um, [clears throat] but I might want to
50:45sync this up a little bit with the work
50:46that I'm doing locally. Um, so Kuro
50:50[clears throat]
50:50has agent hooks. And so agent hooks are
50:55uh a way for Cur to respond to events
50:58that happen in your environment and then
51:00call the agent and take some action on
51:02your behalf. Uh so for what I've done
51:04for the demo here is I'm going to set up
51:06a hook that will uh if anything changes
51:10in my uh task file to check that and
51:14then do the right thing to update uh my
51:18Jira for me. Um so I can I can describe
51:20this in natural language. I don't have
51:22like a whole bunch of configuration
51:23clickies that I have to do. Um, and it
51:25figured out that it needs to monitor the
51:28taskm file. It kind of broke down to
51:31make sure that I'm clear. I can iterate
51:32back and forth with it about what I
51:34want. And I wanted to like if I mark a
51:36task done, make check to see is that the
51:39last task in the story, can I mark the
51:41story done? Um, and then it kind of
51:44updated and kept track of that for me in
51:46my task file. And now that I'm ready to
51:48get working, uh, these MD files, uh, are
51:52embedded in Kirao in a way that I can
51:55just click on that little button there
51:57that says start the task. Gives me a
52:00little indicator that tells me the
52:01progress of the task. I can watch it uh,
52:04in the chat dialogue and see what it's
52:06doing. Uh, if it needs permissions for
52:08something, it might ask me.
52:10[clears throat] Uh, it kind of I
52:13probably noticed there was a little red
52:14there. found it. It kind ran into an
52:16error, but it figured it out and moved
52:18on. Uh, and kind of built that first
52:21task and gave me a summary of what it
52:23did. Now, as I mentioned, there's 19
52:25tasks in here. Um, and I don't want to
52:28handhold them all. I want to go get
52:29coffee. I just told it to go ahead and
52:31power through the rest of them. Uh, and
52:34I did. I went and got a coffee and came
52:36back about 15 20 minutes later. And here
52:39we go. We got all our tasks completed uh
52:42with a summary, a wrap-up of them. It
52:44went to Jira. It updated uh the status
52:47of those. Uh and
52:51just for fun, I'm like, let me check the
52:53burndown chart. There isn't granularity
52:55in Jira that can show you what actually
52:57happened because I got all those tasks
52:59done in less than an hour. And that's
53:01the smallest granularity uh in their
53:03burndown chart.
53:07>> [clears throat]
53:07>> But I can go in and uh just like look in
53:10my terminal here and kind of show you uh
53:13that we can see the code here. It added
53:17uh initializing the bedrock agent, added
53:19the agent uh for us. Uh it 2,264
53:25lines of code and 18 over 1,800 lines of
53:28just documentation. So it also updated
53:32uh all my docs for me. And here's my
53:35app. and I got a new button on my app. I
53:38can ask it questions about
53:40[clears throat] my catalog. So, it knows
53:42about my the API for my catalog and is
53:45able to have searched that uh and found
53:48what I had, but it's not perfect.
53:51[clears throat]
53:51Uh the font is not in a color that I can
53:54really read here. It did a fairly decent
53:56job picking up my styles, but I don't
53:58think we had anything in there uh for
54:01this particular issue, which but let's
54:04just go back into Kira and see what we
54:06can do to fix it. Um, and this is where
54:08like we can just vibe code this one
54:10here. We can click the vibe code button.
54:12I'll describe the problem that I ran
54:14into. Also, the output from the LOM
54:17there uh at the end, that big block was
54:20just a big block of text, but it's it's
54:22marked down. So I'm asking it, can you
54:24please format the markdown so that we
54:26can get bullets and headers and all that
54:28kind of stuff.
54:29[clears throat]
54:30So it runs off and does that for me. Uh
54:34we can just refresh the app here real
54:36quick and see what it did. I don't even
54:38have to reload it. And there we go. Now
54:40I can read it. I got bullets uh and
54:44[clears throat] headers uh and it looks
54:46pretty good. But you probably noticed I
54:48was talking over it. It was very slow.
54:51Um, uh, that was not right. Um, but that
54:55gives me a chance to show you the CLI.
54:57Uh, the CLI is, you know, just in the
55:00terminal. There's a ton of tools in it.
55:02It's a little bit more handsheld manual
55:04kind of environment if you want to be a
55:06little bit more interactive and
55:07involved. Uh, you know, different tools
55:09for different folks. Uh, I like driving
55:11a stick shift. So, for me, this is
55:13really the world I like to be in. I
55:15really like I use this daily uh for
55:20pretty much everything I do. You can see
55:22the list of features and a lot of
55:24experimental stuff that's coming out. Uh
55:26if you're using this, you should be
55:27checking the release notes pretty much
55:29every other day or so. They're they're
55:31using Curo to build Curo. So it it's
55:34just it's just fast. My job is to
55:36describe Curo to you and I'm learning
55:38things from people because it's just so
55:40fast that it's coming out.
55:42>> [clears throat]
55:43>> One of the features is this ability to
55:44do agents and you can kind of bundle um
55:49MCP files and steering files and stuff
55:52like that and you can switch those back
55:53and forth depending on what you're
55:55working on. So maybe you you can say
55:57like you can bounce back and forth.
55:59We're doing some kind of agent core
56:00related stuff here. So I'm using my
56:02agent core. Um, and I told it what I was
56:05experiencing and it noticed that I it
56:09did I didn't we didn't use the streaming
56:11interface that were blocking on the
56:13calls from Bedrock. Uh, so it went ahead
56:16and um is making the update. It's asking
56:18me if I want them. It's showing me the
56:20coding uh changes that it's about to do.
56:23Um, and I just went ahead and approved
56:25them for the sake of this. And then
56:26we'll come back in here and I'm going to
56:28ask it a question again about my catalog
56:31if it can go out there. And you can see
56:33this time it was a whole lot faster.
56:36So there, that's a lot of stuff. I went
56:38through a lot really quick and I could
56:40really could talk about this all day.
56:42There's a lot of really cool features uh
56:44within Kuro that you'll see if you play
56:47with it and experiment with it. Um just
56:50neat little stuff like being able to
56:52automatically generate um your comments
56:55for a a commit. uh you click the button
56:58here, it knows what the context is, what
56:59the chat has been, and can create that
57:02message for you. Um we've recently added
57:05uh checkpointing as well, so that if the
57:08agent just did something that you kind
57:10of don't like and want to go back, you
57:12can revert that checkpoint uh and take
57:15it back. Um
57:18[clears throat] and things, like I said,
57:19things are coming out pretty much on a
57:21daily basis. So, uh, to get a real deep
57:24dive into this, there's some other
57:26sessions at reinvent that I would
57:27recommend. Just go find those. Come see
57:29us at at the kiosk, but more
57:31importantly, download it. Um, play with
57:34it. Um, let me know what you think. Uh,
57:37and I'm going to give this back to Rory
57:39here. She's going to close it out with
57:41some more information of how you can
57:42learn more about Kira and get started.
57:45>> Thanks. [applause]
57:52How do I switch it back to prisoner
57:53mode? This one. Oh, nice. All right. So,
57:58getting started with Curo is pretty darn
58:00easy as far as, you know, downloading
58:02it. There was an Easter egg in that demo
58:04that I don't know if you are paying
58:06attention, but it was Curo Powers. Curo
58:09Powers is available today. It was an
58:12embargoed session. So, if you want to
58:14learn more about hero powers, I would
58:17recommend going to Nil's session DVT228
58:22um tomorrow to learn more how to get
58:25what did you call it? MCP
58:27uh servers on steroids. Yeah. Well, I
58:30don't think that's our official
58:31marketing line, but um it is a way that
58:34we can accelerate the sort of connective
58:37tissue, especially with other AWS
58:39services like agent core. Um so thank
58:43you very much for attending this
58:45session. Uh we would love to hear more
58:49about what you like, what you don't like
58:51because like I said our development
58:53cycle has been greatly disrupted. Take
58:55Kilo Curo Powers. I don't think we had
58:59um anything planned for Curo Powers a
59:02month ago. Um and we went from inception
59:05all the way into production to launching
59:06it um at reinvent in under four weeks.
59:09And that's actually pretty slow for us
59:11now. So uh [laughter] these technologies
59:14are massively disruptive not only to our
59:19own internal processes but I hope they
59:21get to disrupt your processes as well
59:24and give you the opportunity to do more
59:27faster and put more human in your
59:29projects. Thank you.
59:32[applause]

About this transcript

This page contains the full transcript of AWS re:Invent 2025 - Kiro: Your agentic IDE for spec-driven development (DVT209) by AWS Events, generated from the public captions YouTube serves with the video. The transcript has 9,565 words across 1,404 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.