Livestream: Opening keynote and Product direction, at In The Loop 2026 (Oct 14)

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Livestream: Opening keynote and Product direction, at In The Loop 2026 (Oct 14) · Channel: n8n · 1:02:24 · Watch on YouTube

Topics: AI agents, Task and workflow automation, AI evaluation and benchmarks, AI developer tools, AI governance and regulation

Summary

At n8n’s first In The Loop conference in Berlin, CEO Yan Obußer describes the company’s ambition: make AI more capable while helping people stay informed, in control, and responsible for the work it performs. He frames n8n’s direction as becoming an operating system for AI-powered work, combining agents, deterministic workflows, automation, and human ownership. VP of Product David Roberts reviews recent product developments and explains how n8n is approaching building, auditing, deployment, monitoring, and governance as AI use expands. The keynote also previews work in progress, including a new workflow engine, cross-instance observability through Manifest, and improvements to evaluations and security.

Key points

n8n’s guiding principle is that more capable AI should create more capable people, not leave them understanding less or surrendering control.
Visibility into inputs, outputs, permissions, and responsibility helps people understand, control, and trust automated work.
The company aims to become an operating system for AI-powered work, combining agents with deterministic workflows and human oversight.
The keynote distinguishes individual work, where assistants can help, from shared organizational work, which requires orchestration and accountability.
Recent product developments include standalone agents, an AI assistant for building workflows, MCP-based workflow creation, and n8n Connect for selected credentials.
Agents and workflows have different strengths: agents handle variable tasks, while deterministic logic is suited to repeatable tasks where reliability matters.
As AI speeds up building, auditing becomes a major bottleneck; n8n is exploring easier evaluations and stresses the continuing importance of inspecting workflows.
A new workflow engine is being developed with features such as saving progress after each step, safe retries, and isolated resources.
Manifest is being trialed as an observability layer across multiple n8n instances.
Governance plans include agent identities, access controls, policies, usage limits, lifecycle management, testing, and security measures.

Chapters

Start	Chapter	Summary
00:04	Welcome to In The Loop	The event host welcomes attendees in Berlin and introduces the conference program and n8n CEO Yan Obußer.
03:17	The vision: more capable people	Yan introduces the idea that increasingly capable AI should help people understand more and stay in control.
05:20	n8n’s growth and community	Yan shares company and community milestones, then emphasizes feedback asking n8n to preserve the visual canvas while making it more powerful.
08:24	Visibility, trust, and agency	Yan argues that visibility into work enables understanding and control, and that critical processes need appropriate oversight.
11:29	Orchestrating work across organizations	The keynote distinguishes individual assistance from shared organizational work and describes the need to coordinate agents, software, automation, and people.
14:38	The next chapter and evolving brand	Yan presents n8n’s ambition to become an operating system for AI-powered work and explains its agent-first, workflow-second approach before introducing the updated brand.
20:20	Company culture and n8n Labs	Yan outlines investment in product, customer-facing teams, and Labs, while describing the culture and principles behind experimental work.
24:34	Product recap and recent developments	David Roberts reviews the new standalone agents, AI assistant, MCP workflow building, and n8n Connect.
28:10	How AI changes automation and trust	David discusses improving models, the changing balance between workflows and agents, approval fatigue, and how trust may increase through spending limits.
33:56	The automation lifecycle and building	David lays out the lifecycle of building, auditing, deploying, monitoring, and governing automations, then compares agents with deterministic workflows.
39:06	Building with agents and auditing work	David describes agents as an entry point to automation, explains the growing importance of auditing, and confirms that the visual canvas will remain.
46:22	Evaluations and the new workflow engine	David discusses the role of evaluations in improving agent reliability and previews a redesigned engine focused on reliable execution and safe retries.
53:08	Monitoring and governance	The keynote introduces Manifest for multi-instance observability and outlines governance needs, including access, guardrails, ownership, quality, and security.
59:19	Roadmap and closing	David recaps planned product improvements, previews a new n8n Labs product, and thanks the audience.

Quotes

More capable AI should create more capable people. — 04:19

Please don’t remove the canvas. — 07:20

Visibility creates understanding. Understanding enables control and control preserves agency. — 08:54

trust is essentially the limit on how much we hand over to these models. — 30:16

The canvas is not going anywhere. — 45:21

Concepts and entities

n8n, Jan Oberhauser, David Roberts, SAP, Agent, workflow, MCP, Claude, Codex, n8n Connect, evals, Manifest, n8n Labs, Opus 4.5, Slack

Full transcript (original language)

00:04 [cheering] I didn’t even say something yet. Don’t worry, the real show is about to start. I’m just a forerunner and it’s a my great pleasure and my great joy to welcome all of you here in Berlin for Eden’s first in the loop. [applause] >> I do have to say though, I am a little sad. You know this company runs on basically two things which is efficiency and punctuality.

00:37 Now efficient I think we’re quite good you know I I mean there’s a lot of people on the side but punctuality guys you know you’re queuing I mean you’re here but some people are I think still queuing so well but I think Yan will forgive us because a lot of you traveled from very far away yesterday. Hey, we had a little pre- gathering. People from Brazil, people from the US, people from I believe India, people from Germany. Okay. Well, um, and it’s really cool to see that.

01:08 Oh, yeah. No. Yeah. You know, Germans and applause is a little bit of a problem. We are not typically super proud of like, yeah, Germany is nice. Anyways, so and it’s not only you that are here today, but people globally are watching us on a live stream, right? Your mom might be watching and be like, “Yeah, my son, he’s at the end conference and he’s having a great time learning about AI and automation, whatever, whatever that might be.” Now, we have a very, very full program for you. We have four stages that you will learn about the

01:40 whole universe of NADN and what we built in the past couple of months, what we’re going to build into the future. And I’m very very happy that the first person on stage is also Sabian like me and that he likes his keshler with the onions crispy on top. But compared to that, I think the thing that’s more important about Yan Obouser who’s the obviously the CEO, the founder, the first coder of Nadn, is

02:10 that he really values that you don’t have to waste any minute of your life doing things that could be automated. There’s a little story that goes around in the Anaden office that you know you can take multiple routes to the office but one is the optimal one right one takes 30 seconds less than the other and you know just count like you have a company 365 days in a year you might not go every day right you can save a lot of time if you pick the optimal one right and that’s the kind of spirit that Jan pours into this company and into our company culture we exist because we want

02:41 you to not waste any of your time doing things that could be automated like an AI agent could do it for you, right? Uh deterministic workflow could do it for you. And now it’s my great honor to welcome Yan Uberhouser to this wonderful stage to the first keynote of the day and right after David our VP product who’s going to tell us all about the evolution of Venaden. Enjoy.

03:17 Thank you, Janice. Really great to be here and welcome to the loop today. Very excited to see so many of you here today. It’s really amazing just how many people came in like Yiannis said from all around the world. So, thank you all for joining. Um, I also want to thank our sponsors and exhibitors. Like many of you, you already work for many years already. And like an event like this is simply not possible with any of them. So make sure to visit them during the day and say hello. Yesterday we had our partners and our ambassadors over. Today we have

03:48 everyone, our community, our builders and people in the live stream. The it’s the first conference ever and we’re doing it in Berlin where everything started. It’s very humbling to see so many of you here today. In this talk is going about where is going. I’m not going to talk so much about features. David our VP of product and design will talk about it more later after me. We talk more about the direction and the reasons behind it. And I start I want to start with one idea which sits under everything I will say today.

04:19 More capable more capable AI should create more capable people. I think I think that’s I think there’s one question in the next few years that’s very important. If AI gets smarter every few months, what happens to people? I think there are kind of two paths. There’s like path A, like AI grows more capable, people understand less and they just hope it works. But there’s also path B, like people become more capable too. They understand more. They stay in control. And we are building for path B.

04:49 That’s not new. That’s what we always did. Uh before I started edit, I was working effects for movies. I was building tools and pipelines for people and the artists. And people spend a lot of time on repetitive work. They were copying around data doing the same task again and again. I was doing the same thing and one day I started nodn working on nights and evenings and weekends. Um and for me first I actually started to build it for myself and then literally for anybody else out there and

05:20 then I launched it in Berlin in 2019. We started with the same mission we have today to give everybody use the computer taking superpowers. What changed is that the superpowers got much stronger. Luckily, a lot of people took us up on that. But now we have 6.7 million monthly active workflows. Already 60% of the Fortune 500 companies are using us. In the last funding round, we were valed $5.2 billion with the investment from SAP. When I wrote the first line of

05:50 code, I would never have believed any of these numbers. And honestly, this party still seem crazy. But the numbers actually not very important. What’s behind them is our community is a real mix. Like we have people who live in code, some people have never wrote a single line of code and all building together and helping each other out. But now we have a community of 519,000 people. We have 13,000 workflows live templates mostly built by our community. There are 2,300

06:20 integrations and we have 340 ambassadors. And there’s now a community event almost any day of the week somewhere in the world. Um, we are also a top 30 GitHub project of all times. We have 200,000 27,000 GitHub stars by now and 400 active partners including strategic partners, system integrators and agencies. We work with companies like SAP, Accenture, Capgeemini, Do Telecom. Many of them are here today. I still remember like one of the first community meetups. there were like 60 70

06:50 people like pretty much like a room today but obviously very different scale today but the same thing is true people showing what they have built we spend a lot of time listening in forums at events in interviews even Reddit which honestly is not always the most helpful thing to do but it is what it is but don’t but what you are saying to us is don’t replace what we love make it more powerful AI on top of the canvas not instead of Also, let me build my way

07:20 like on canvas in code or for my coding agent. You won’t have all three working together and also help me run it in production. Like to make sure we have the confidence that it actually runs also on Tuesday night when nobody’s watching. One um community comment actually summed it up perfectly. Please don’t remove the canvas. Message received. Let’s let me share it how one user puts it. He said what I like most about Nitn is it flexibility. I can build workflows and agents visually and really

07:52 understand what is happening and that’s what we hear again and again flexibility and being able to understand is what is happening. Um and the second part like understanding is let me understand what’s happening and the second piece understanding piece is something I will talk a lot more today. So where do we go from here? Our view is actually quite simple. More intelligence expands what we can delegate. People can still define the problem and shape the results. Like with every new model, we can hand over more. The old question was pretty much can I automate this? And the

08:24 answer is more and more yes you can. But the new question is what should people continue to understand? That opens up the next chapter the threat through this talk. All right. Didn’t realize it’s quite messed up. Okay. Back. Okay. This is my not my fault. Okay, let me start with what end does today. The foundation for anything else, for everything else when I build on edit, when you build an edit end, you can see the work every step input and output. And that what matters to

08:54 understand the work like if you understand the work, I can shape it. I can trust it and I can and you can stay responsible and if you cannot see it, you can only hope which is obviously very bad. I think of it more as like a chain like visibility. Visibility creates understanding. Understanding enables control and control preserves agency. And agency is your your is your ability to decide, to act and to change course. If you remove the first link, literally everything falls apart. That’s

09:25 why visibility is not a nice to have. Like that’s literally the whole point. But what does it mean to make intelligence visible and and governable? We think about it in four parts. First, what goes in the data, the context and the permissions. What is it allowed to touch? So, what happens not piece together what um piece together after it broke and what comes out like the results you can check, trace and stand behind and who’s responsible like

09:56 a clear owner and um any and moments for people to really step in. Only if you can answer all four of them, you can really trust it. And if not, you should really not run your business on it. This is why I believe that UI is a trust layer. Many think that interface goes away with AI. You prompt and you’re done. I actually think the opposite is true. Like the more AI does, the more

10:28 you need to to see what actually is happening there. That is true for the builder, but also for everyone who depends on it. But not every task needs the same level of visibility. If you’re asking to solve my personal inbox, all I care about is the result. If you get something wrong, fine, I can fix it. But if my team shares a piece of a process, a share process, I want to see the key steps. And if it’s a business critical operation like paying an invoice or onboarding customers or anything regulated, I really want to see every

10:59 single step. The more critical the process, the more you need to see it. That’s the range N is built for. But seeing it is only at the start. Understanding should turn into agency. That means three things. First, respect what the what the what the the work it did, what happened and why. Second, in the win when it matters. Stop, change, approve before it goes wrong, not after. And third, improve what you delegate. Every run teaches you something. Delegate more. That’s the loop. This may

11:29 be why we call the convers. Now I talk more about where and end matters the most. Building becomes easier and easier. Anybody can build now an impressive day in an afternoon. But using AI for business where the business really depends on it like every single day is still very very hard. Like work work inside of an orc is different. It runs across people across systems and and responsibilities. If you take a sales process, you have a CRM, you have legal, you have you have finance, you

12:00 have three different teams. What it needs is coordination and someone accountable. The gap between it works in a demo and the business runs on it. That’s where ended in was always strong and where the next years will be decided. I think there are two kinds of work. First, there’s individual work like one person, one context, their own tools and data and their own responsibilities. That’s where assistants are great. There’s also organizational work, shared system and context, dependencies across teams and shared responsibilities.

12:32 That’s where you need orchestration. There’s also a shift. We also see it when what people are building with NN. In the past, they built nice chat bots for themselves. Now they build shared shared processes like shared work. They process invoices, do customer onboarding and if it breaks, the business really feels it. This leads us to a simple principle. The greater the consequence, the greater the requirement for control and accountability. Like if an assistant books the wrong restaurant for me, that’s annoying, but

13:03 not the end of the world. But if an agent sends the wrong number to a regulator, that’s a very different conversation for so for critical work. It has to be clear who builds it, who can change it, who governs, and who checks. Software alone cannot do that. Needs people who know the process. It’s our job to give them the right tools. Critical work needs more than just AI. It needs four things. First, agents, judgment, and language where the work really needs them. Second, deterministic software. One right answer every time.

13:34 You literally don’t want that. The AI gets very creative when it does your text calculations. And third, automation, repeatable, repeatable steps at any volume. And fourth, people and teams, ownership, context, and the final call. And to be clear, the hard part is not any of them. The hard track is bringing them all together and that’s N atn. Three principles guide us. Keep close to the problem. The people who know the work can build, inspect and remove it with technical depth available when needed. Then delegate that growth.

14:05 Delegation that grows more delegation through context data historical traces with control that fits the consequence. and ambitious and rigorous. Ambitious about what’s possible and rigorous and responsible about what’s ready. Like, and we really need both. Like, if you’re only ambitious, like, you cannot trust it. And if you’re only rigorous, we’re going to be moving way too slow and you’re going to leave. I like how Ollie put it. He said, “N is the thing that allows you to integrate AI into your work and your processes in

14:38 a safe and controlled way.” That’s it. That’s what we’re trying to be. But let me connect the dots. AI makes intelligence possible and it then makes intelligence operational. So we we turn what models can do into work that actually runs every day. Our next chapter expands on that. We want to be the operating system for AI powered work. That’s our ambition. Some of it is already here today and a lot of we are still building. It matters as it matters as the way work

15:08 starts is changing. Before it started with the steps. You do this then this and so on. Now people define the outcome. Start from what should be true not every step. AI AI helps you build a path. Agents are drafted. Workflows make it dependable. People shape the result. Inspect it, adjust it and stand behind what runs. So it moves from doing every single step to owning the outcome. Um that what I mean with creating more capable people.

15:38 I think two principles come out of that. First, agent first, workflow second. You start with the agent at workflows, deterministic logic when needed. And AI first, not AI only. Like 100% autonomy is like magic when it works, but too unpredictable, unpredictable for critical work. The most powerful systems actually combine AI, detistic software, automation, and people and teams. The question is not AI or not AI, it’s how well you can combine them.

16:09 But let let me give you two examples. The first one was dispense. They chose NN as the global low code automation platform making it available across the company. They think in layers. Some use AI, others orchestrate it and the different group builds advanced solutions and then is orchestration layer. They run self-hosted and cloud agnostic. So the data and the control stays with them. But not only large enterprises or a small UK agency built in a system for this Rama group. The team sends WhatsApp

16:40 voice notes that updates the CRM and their systems. The result is that 97 less 97% less time spent on the task. So it went from literally 4 to 5 minutes to 10 to 20 seconds also from 62 to 44 days from onboarding to selling a property. I’m a founder shared by now 70% of his business is running on NN. So from car maker to small agency is the same idea intelligence turning into dependable work. So why is end the best place to deliver

17:11 on this? This accessible building the people closest to the problem can build. There’s technical depth there if you need it not if you don’t. Visibility every step is readable by the people responsible for it. And orchestration agent system and people coordinated. and obviously the ecosystem and that’s the hardest thing to copy because we didn’t build it alone. We built it with all of you. So for our next chapter, we’re looking to to grow from a product to an operating system for AI powered work.

17:42 But what does it really mean? This something that really produces outcomes for people. Getting there requires requires broader capabilities and over time additional products. So you may have noticed that some things look differently around here. Our brand is evolving. Our old logo kind of celebrated the notes. The new one is celebrating the space between them. The activity, the connection makes possible. And honestly, it even took myself some time to get used to it. I know change is hard. But now I really love it. But the

18:14 pink stays. The look changes what we’ve sent for does not. >> Hi, I’m Max or as some of you know me, the original flowmer. Much like any builder’s journey, N8 ends is a story of constant iteration. Ship [music] stuff, break stuff, learn, listen, and get better as you go. And honestly, that’s why it’s time for how the NAND brand feels to get a so when the NAND brand team showed me what they’ve been working on over the last few months, I was super excited to be

18:44 the one to share it with you. Now, they didn’t start from scratch. At the heart of Naden is the glorious community of builders and a platform centered on connection, control, and flexibility. The old logo connected the nodes. The new one is interested in what happens between them because that’s where it gets interesting. People jumping in, agents agenting, systems interacting behind the scenes, and decisions made in real time. And a tight face to match, too. precise, technical, it’s still

19:17 human. Now, not everything needed a revamp. Pink has always been part of N. Though, a brand only really matters when it makes life easier for the people using it across the product, community events, [music] and everything in between. And as more people discover the product, it still needs to look and sound like Nitn. [music] A brand people can wear, share, and make their own. Designed to bring ideas to life. Designed to put builders center stage.

19:49 Rebrand. [applause] Thank you, Max. I love his energy. I bought like had some of it as well to be honest but it’s not me [snorts] anyway how we build a company must support thank you [laughter] how we build a company must support the collective ambition as well if you believe AI makes people more capable that has to be true inside of as well

20:20 like our culture our capabilities and our ecosystem need to deliver on this ambition as well and to keep the agency that brought us here so to to get there we have to invest in three areas first product engineering, the broader platform our that our vision really needs. Second, teams that put Naden to work, people who help customers and partners to get real work into production and third and labs exploring what comes next. We also want to build the or of the future, multiplying the impact everyone

20:51 can have with AI and NN. We run NN on NN and do that a lot. Um and and all of that obviously rests on our culture code which you can see here which is we are builders, we are welcoming, we do the right thing even when it’s hard. We are early and we are multipliers. They’re important for how we work with each other, how we work with you and how you experience Naden. Yeah. And I think one line from our culture code really says it like and I

21:23 really like it. It’s like when something needs to be doing, you do it from the switch call to the dish in the sink. And obviously it’s also true for me. Maybe less today as we bought a dishwasher. But honestly, I’m still the person very often getting things out of it which still needs doing. But this becomes less than washing stuff in the sink. But what it means like nobody no task is too big or too small for anyone. This how we got here is how we keep on going. Monro thing how we will get there. We started ended in labs. It explor It explores new approaches and potential new products that support this ambition.

21:56 Labs works by a few principles. Speed and signal over polish. We want to move fast. It doesn’t have to look pretty. Build for the capability curve. Models get better very fast. What doesn’t work today or barely works probably going to work perfectly in six months. And they fall to things shutting things down. Like shutting down is not a failure. That’s a way of learning. It’s a very small team with literally just one job. increase the prob probability that ended n wins the AI native future.

22:26 So when you put it all together, that’s why we’re the best place to deliver. Accessible building. The people closest to the problem can build technical depth. Don’t hide the complexity. Visibility. Every step readable by the people responsible for it. Orchestration agent systems and people coordinated. And the ecosystem a community partners integrations around every build. And as I said earlier, the ecosystem nobody can copy because if we build it with all of you

22:57 with all of these changes very clear on what never changes. We obviously keep our community edition. It will stay free and self-hosted forever. The model choice, it’s your models, your providers and it’s your call. The community and ecosystem, our programs, our partners and how we operate free and pay free and paying users will be always treated alike. and we the current nit we will have 50% of the IC’s keep on working on the core this is a community we never we never going to stop caring about and we restrict never access to it ever

23:29 so clapping close where I started more capable AI should create more capable people that’s the reason why ended started that’s why we bu what we are building towards and if I’m honest it’s only possible because of the people in this room and watching online so How does it become real in a product? David Roberts, our VP of product design, will show you what is available today, what is next, what we are still exploring. So, thank you for have a nice day and over to David. [applause]

24:02 [music] >> [music] >> Good morning, Berlin. How are you doing? [cheering and applause] >> That was pretty good. You guys are probably not from here. I’m guessing. [laughter] >> Okay. So, I’m here to talk about um N’s next chapter. Um and

24:34 uh there’s four steps to this, four stages. First of all, a lightning recap of some of the highlights of the last 12 months. Secondly, I’m going to talk about some strategic themes that we spend a lot of time thinking about inside NA10 and then how these themes have shaped our thinking about the role we play and and what comes in the future. And then finally, what we do have planned, what’s coming next. Okay, so what have we been up to? N is a

25:05 very different company to its early days. Um, and one thing we pay a lot of attention to is moving with the times because the times are changing so fast. Nense started as a workflow company obviously and while that’s not going away, there’s so much more to automation these days. One example of that is agents. Um, and while agents were always available inside workflows, we’ve taken agents out

25:35 of the workflows and put them on the same level. Um, because they are just such an important part of the future. And this was rolled out fully last month. Uh, they’re easier to create, they support skills, they support MCP more easily, and they also have richer integrations with tools like Slack. Now, these agents are about 20% more likely to get to production compared to the old style of agents. Um, so they are

26:05 definitely a step in the right direction and there’s going to be lots more content on this throughout the day. Okay. There’s also the assistant. So, we’ve had uh AIS help in building workflows for a while, but it’s really come into its own in the last 6 months or so. In fact, it’s become the default way of building workflows in NA10. Now, uh the assistant’s had over 1.6 million

26:35 conversations. It’s built over 800,000 workflows and eight out of 10 users that we surveyed rate it good or very good. And it’s available everywhere. It’s available on cloud. It’s available and self-hosted and it’s coming to enterprise very soon. It’s not the only way of building an NAM though with AI. Um while we’ve supported MCP for a long time, this year saw us add the ability to build workflows with it. So you can now hook up Claude or Codeex or your favorite coding agent um

27:08 to N and build workflows that way. And this is a real power user feature because people who use this feature are about 3.4 times more likely to get to success with NA10 and they create three times more workflows than the average user. So word on the street is it just works which I think is kind of dangerous to say about any AI feature to be honest but um it it’s definitely a strong addition uh to NAS. And then finally, we’ve got NS Connect.

27:39 Um, setting up credentials [snorts] is perhaps the most painful thing about automation. And NASA Connect removes that pain by doing it for you for selected credentials. It launched recently, but it’s already being used by a quarter of our users and it makes them about 25% more likely to get to success. That’s just a little sample of some of the things we’ve been up to. There’s much more in our change log on our docs

28:10 website if you’re interested. But now let’s move a little bit into how we’re thinking about the world and I’ve got five themes here for you. So first of all, we all know models are getting better and better and what the best ones can do today would have felt like magic just 12 months ago and we’re getting more and more models. Open source has stayed consistently on the tail of the major labs so far and we believe that LLMs are expensive today

28:41 basically because of limited supply. You know that’s going to change especially if you’re okay with something that’s a tiny bit behind the frontier. Uh and this has lots of implications. I think one of the ones we think about a lot is designing for the exponential. So when we’re thinking about prioritizing features, we ask ourselves the question, if models get 10 times better, will this feature also get 10 times better? Like can we harness this uh this trend that we’re all riding? And today you have to

29:13 build scaffolding around models, especially weak ones need a lot of help in order to do a job. As models get better, that scaffolding has to be removed because if it isn’t, you’re actually making the model’s life harder and you’re not realizing its true potential. And of course, when thinking is cheap, we do more of it. There are going to be many more agents than humans in the businesses of the future. And the questions are going to shift from can we do it with an agent to are

29:45 we allowed to do it with the correct result. Now most automation today is workflow first. And so what that means is a person or an AI decides the steps in advance and they map them out. Now some of these steps could be agents but they be within a workflow and this is the majority of what runs in production today. Now a smaller but faster growing share of that is agent first of course and there you’re giving the system a goal

30:16 and it’s free to decide how to reach that goal and it’s growing because models keep getting better and they keep getting cheaper but workflows are not going to disappear and I’ll [snorts] say a bit more on that later what is very clear is that the balance is going to change and as agents become more capable we move from can I do it to do I trust it to trust is essentially the limit on how much we hand over to these models.

30:47 So if you have a capable agent but you don’t have the trust you’re just going to use it for drafts and for summaries but a trusted agent can do real work. There is a downside though to all of this this human involvement. Um and that’s one of in one example is approval fatigue. You know if you’re asked to approve the same thing a 100 times you are just going to it’s very hard to not just click the button without really thinking about what you’re approving. Um [snorts]

31:17 and so the the real question here is where is a human’s time worth spending? And the answer to this is constantly changing as the technological landscape evolves. And trust gets built in steps. Money is one way to measure those steps. And so here in step one, an agent suggests and the human does. There’s no spending authority with the agent. And most AI usage is here today. In step two, the agent does the work and

31:50 the human approves any spend. So the agent can act, but it can’t commit funds. And then in step three, the human sets the budget and then walks away. And the agent is now an economic actor in its own right. [snorts] It can commit resources on its own within a limit that someone has set for it. And so the change here is trusting the agent more with each step. And the human job changes from doing to approving to setting limits and then reading a

32:21 report. And by the way, budget is a potential answer to this approval fatigue thing that I just talked about because instead of requiring a 100 decisions, you can just set a threshold that the the agent can’t exceed. It’s not the answer to all approval fatigue, unfortunately, but it is for for some things. Okay. Then the final the final point I wanted to make is that everything I’ve said so far applies within a single company. You have one set of rules. You have one owner. You have one person to

32:53 call when it goes wrong. And that could end pretty quickly when my agent can negotiate with your agent or my agent can place orders with my supplier or my agent can book in your calendar because then you have a company border. And at a company border, you have two sets of rules meeting each other. And the questions get harder. you know, whose limits apply? You know, this companies, this companies, is it the the stricter of the two? Um, how do you prove that an

33:25 agent is allowed to act for me? And who keeps the records? Who’s [snorts] liable when something goes wrong? Today, we answer those questions with humans and with contracts. But when agents move at machine speed, we need [snorts] something that’s built into the system. uh and we’re going to see new institutions evolving I think um to satisfy these needs. Okay. So what does this all mean for

33:56 NASA? This is my favorite framework for thinking about automation. Uh it’s the life cycle that an automation goes through and the engineers in the room will find this very familiar because it’s basically the software development life cycle too. To start off with, you have to build your automation. I mean, that much is obvious. Uh, then you’ve got to get confidence that it works properly, that it’s going to do what you want it to do. That’s the auditing step. Then you’ve got to put it somewhere and run it, which is deployment. And while it’s

34:27 running, you’ve got to keep an eye on it. You’ve got to check that it’s working properly. You’ve got to investigate any failures, and that’s the monitoring. And then finally, there is the whole human or organizational side of running anything in a company. And that’s where management uh and governance come in. So let’s talk about building. So 12 months ago, if you think back, we were on the verge of Opus 4.5 and the explosion of clawed code. And I

35:00 think it’s fair to say the world has changed beyond recognition since then, especially if you’re an engineer. And when these things work well, they really do feel like magic. And of course it creates big opportunities for N10. And I think one interesting way of thinking about what’s changed is that building with a tool used to mean learning that tool’s grammar, learning that tool’s way of thinking, which sometimes was hard. I mean, with Niten, we we were a pretty good example of that. Um, and that’s changed with with

35:32 natural language input. Okay. And there’s really two aspects to building, both of which have changed. First of all, you can build new things. And second of all, you can build them in new ways. And I’m going to talk about both. But first of all, let’s talk a little bit about what you build. We’ve talked about these two architectures at a high level. So you have the agent, you have the workflow. And by workflow, I I really mean

36:03 deterministic logic. It doesn’t have to be a workflow. It could just be code. There are risks to both of these sides and they’re different. So if agents can figure things out for you, they may figure out the wrong thing. And a workflow that does exactly what you said every time also repeats your mistakes. On the other hand, agents, you know, they’re they’re great at handling varied different flexible input formats. They’re great for natural language

36:33 interfaces. Workflows on the other hand are faster, they’re cheaper, they’re more reliable. And one nice way of of thinking about this difference is where does the thinking happen? You know, does it happen once at design time? If so, it’s a workflow, or does it happen every time it runs at runtime, in which case it’s an agent. So, which one is better? Yeah, they’re different, right? They’re

37:04 different. They have different strengths. They have different applications. They’re essentially two puzzle pieces with different shapes. So if your task you need if you need it to run constantly, if it barely varies, that sounds like a workflow. If it’s rare, if it’s messy, if it’s different every time, then an agent is probably more suitable. On the other hand, an agent doing the same thing 10,000 times a day is just a very expensive workflow. And for all the great stuff about AI, it

37:34 does have this pesky little quality of non-determinism. So it’s just a little bit unpredictable. And of course, if you want your system to be reliable, you want as little unpredictability as possible. And this um gets even worse when when you have multiple steps in a chain. So if you have an agentic step that’s 95% reliable and you do it 10 times, that system is only 60% reliable. So when intelligence is abundant, being

38:04 sure is what’s scarce. So you should take anything any part of your system that can be deterministic and you should make it deterministic because that improves the overall reliability because of course trust is slow to build and it’s fast to lose. And this goes even when the agent is the top level architecture of your system. What you want to do there is you want to take parts of the problem, take parts of the process and put them inside deterministic tools and you want to

38:34 encode as much business logic in those tools as possible because that makes the models choices simpler. So if you think about an IT department trying to automate employee onboarding for example, you could give your agent a tool to create a account in Slack. You could give it a separate tool to create an account in Google Workspace or in Salesforce. What you might want to do instead though is just create a single tool for onboarding a user to a given department that does those account creations itself

39:06 because that kind of task can be deterministic and you’re taking it away from the agent given that it doesn’t need to be nondeterministic. Okay. So, so that’s the what. Let’s talk a little bit about the how. Obviously AI has revolutionized this too. We believe that no one will build manually in the future. And so if you haven’t checked out the assistant, if you haven’t checked out MCP, we highly recommend that you do that. [snorts] But agents are also the front door to automation. They’re like the starting

39:37 point. So many people when they first come to automation, when they first come to an agent, they’re going to ask for some quick one-off task. I might say, if I’m feeling ambitious, you know, I’m going to Spain next month. Can you book me a flight, please? And that’s the easy mode for agents because I’m sitting in front of the computer and I can answer any questions. I can steer the agent if it goes wrong. And then maybe it’s successful, it works, and I decide I

40:08 want to do this every month. I fly to Spain every month for business. I mean, I don’t. I wish I did, but let’s imagine I did. Um, and and the difference here is not just that it’s a repeating task. it’s that I’m probably no longer sitting in front of the agent while it’s doing this stuff. So, it needs to be more independent. And so, the the the quality bar has gone up. And then finally, I may say this is a really important process. This needs to have another level of reliability. And that’s where this hardening, this taking of the logic

40:41 out of the agent and putting it into deterministic workflows can happen. And of course, if you’re an automation expert like many of you in this room are, you may jump straight to step three. Uh you don’t have to go through this entire process. But equally, there are many things, many types of automation that might never get past step one or stop at step two. So the point is it’s all a continuum and agents very often are the front door to automation. And we talked about these two ways of building. Um they’re both important and

41:13 we’re going to continue to invest in both. because Nan has always been an all-in-one solution. You know that uh that life cycle I showed before. We cover all of it and that’s by design and it’s [snorts] always been easy to get started because of that and that’s exactly why we have the assistant. That’s our front door for uh new users coming in. But on the other hand, we don’t believe in trying to control users or dictate where they do work. And that’s why for people who prefer to live

41:43 in cord, they can use the MCP instead. So we have lots of plans for the building experience. Um they include improving the assistant, improving the MCP. They include making that front door easier to walk through. So we recently released an automation mining skill to give you ideas for what you can automate, for example. Um they include bringing your own context because agents perform so much better when they know more about you. And they also include self-healing automations so that you

42:13 don’t have to be pouring over the uh the executions all the time. Now building in this process was traditionally the bottleneck. It was traditionally the hardest part and the most time consuming part and that’s gone. That’s gone with AI. Unfortunately, in any system that’s closed like this, um when you remove a bottleneck, you get a new one somewhere else. And the bottleneck now is most definitely with auditing. We see this with engineers because they are drowning

42:46 in PR reviews. They’re drowning in code review that’s been of all this code that’s been generated by agents. And I think auditing is the last part of this system that we’re ever going to hand over to AI since it’s about trust. So how do we think about auditing? When the building experience is really fast, auditing can become actually slower. Um, so you can build something 40 seconds and then spend 40 minutes checking it works. Anyone can relate to this experience.

43:17 Yeah, I see a lot of hands. [snorts] Um, auditing doesn’t always matter though, right? There are certain conditions I think you need to to care about auditing. For example, your task needs to be relatively complex because if there’s no scope for things to go wrong, then auditing is less relevant. Secondly, it’s much more relevant for business critical processes because that’s when going wrong has consequences. And then finally, it’s more important in collaborative environments where there’s

43:49 more than one person involved because then there’s more scope for misunderstanding and not everybody has the full context. And many people build these kinds of things with NAN. So it’s very important for us that they have a solution for auditing. Two types of auditing. Um there’s inspection which means looking at the logic and how it flows and then there’s testing which means running the thing right and checking the results.

44:19 um inspection would be that PR review for engineers that I mentioned before, but it would also be looking at the workflow graph of NAS and testing is actually [snorts] much more intuitive than um inspection. It’s the kind of thing that people do naturally when they first start using AI, but it’s also hard to do rigorously. So talking a little bit about inspection. Um I mentioned before that we we want want to make sure we move

44:50 with the times and when AI got really good at coding we wanted to make sure that our product wasn’t constraining our thinking that it wasn’t holding us back. So we tried to reason from first principles what would we build today if we didn’t have NAN um and we wondered you know how important is the canvas? So we devised this question. We said, “Here’s the good news. You’re living in this utopian future where AI is perfect. AI can do exactly what you ask it to. But the bad

45:21 news is you don’t have the canvas anymore.” And we surveyed about a thousand users about how they would feel about that. And what came back surprised us because 55% of the users we surveyed said, “Not only do we need the canvas, if you didn’t have it, NA10 would lose much of its reason for existence for us.” So, I just want to reiterate the canvas is not going anywhere. It would be crazy to remove it. Um, [snorts] but it just goes to show like how important that

45:52 inspection is. [applause] And we then asked, okay, why the canvas? Why is the canvas important? And these were the top three results. Most importantly, understanding how the data moves between steps. Um, which is certainly a form of auditing. Um, and the other two, I mean, I think they’re all about either understanding or auditing. And so these are the things that we’re going to be leaning into. So that’s inspection. Let’s talk a little bit about testing.

46:22 So these are now figures about the overall agent landscape. This is not NAN specific. This is just some market research that I wanted to share. I think it’s fair to say that agents have not fulfilled their promise yet. Everyone’s using them, but only half of organizations have a single agent in production. And I’m guessing you probably know the answer to this. There’s one reason that is much much more common than any other and that is

46:55 performance quality. These agents often just don’t work right. They they don’t get productionized because they’re not good enough to be useful. And this is where auditing comes in. You know, it’s hard to inspect agents. So, we need to rely more on the the running them, more on the testing. And we call this eval. We call this evaluations. and they work. So, one report found that teams with rigorous evals were 70%

47:26 likely to have excellent reliability. And for those that didn’t, it dropped to just 32%. So, eval this is not a secret. But there’s just one problem. No one does them. So if you take this left hand column, only 15% of companies or teams fall into it and it’s pretty damning, right? I mean, you know, why do people not do this when everyone talks about them all the time? I think there’s two parts

47:56 really. First of all, they’re complex and secondly, they’re tedious. They’re really boring. Um, they’re complex because they’re hard to understand conceptually. It’s hard to choose the right metrics. It’s hard to track down the right data very often and they’re tedious because you have to comb through results looking for problems and you have to figure out the business logic for obscure edge cases. Now fortunately the problems that AI creates sometimes AI can also solve.

48:28 Um and that’s where we are headed with evaluations. Um, we already have evaluations in the product of course, but what we are working on is just surfacing the business decisions to the user and taking care of the rest automatically with AI. So, expect more on that soon. Okay, we’re halfway through. Um, this part’s going to go faster, I promise. Um, deployment. So, you’ve got to run your automation somewhere. Inexperienced

48:59 automators tend to forget that. um they they tend to think about the beginning of the of the funnel um and they underestimate how important it is to have things run reliably. And of course the the ability to run things is something that’s been built into NA10 from the start. In fact, it’s fair to say the engine of NA10 is the heart of NA10. It was the first part of NA10 and much of it is still Yan’s original code written by his own fair hand. And that’s because it’s the kind of

49:29 critical thing that you don’t touch unless you have a really good reason to. Well, we are touching it. And in fact, we’re completely rewriting it from the ground up. And you might say, okay, we talked about going back to the drawing board and not being constrained by existing NAT. Why not just run code? AI is great at writing code. Running code is a very well understood problem. It’s kind of easy. Um, why not just do that? And we tried that and what we found was that

49:59 when we started to list out the requirements of what we’d like to see when the code runs, we started to get something that looked a lot like a workflow. And that’s because we believe it’s very important, for example, to be able to see and understand every step. So when you run a workflow in edit, you know, it lights up green across the screen. You can look into every step and you can look at the data. We think that’s really a really important part of what makes NASN special. And of course, if you do that, then that

50:31 allows you to resume where you left off. You don’t have to start everything from the beginning every time. And then that allows you to retry failures automatically. Hence the new workflow engine. And when we were thinking about the design of this engine, we had a framework for reliability. And this is our framework. It has six things in it. And I’m going to talk you through them using the example of a restaurant just to make it a little bit easier to understand. The first one is you’ve got to write it down first. So a waiter

51:03 needs to take the order, write it down on paper, put that ticket on the rail inside the the kitchen because if something happens to that waiter, the order can’t be stuck in that waiter’s head. It can’t be stuck in that waiter’s memory. And that the same is true of systems. Secondly, there needs to be only one version of the truth. So if the sign on the front of the restaurant says open, the restaurant needs to be open. Thirdly, every job must reach an end. So it’s not okay for an order to get stuck

51:35 in the kitchen going cold and congealing. And then when things go wrong, you need to be able to contain the damage. So if you have a fuse trip on a on a on a fridge, all the other fridges should remain unaffected. It also needs to be safe to try again and you need to know when it’s safe to try again. So pressing the lift button twice is fine. Being asked to pay twice, not so fine. And finally, you need to leave a trail.

52:05 Um so even the orders that maybe are cancelled because the uh the restaurants run out of ingredients need to be tracked. Okay. What does that mean for the the engine? First of all, it’s going to save your progress after every step. This is something that’s optional right now in NA10, but this is going to be core to the new engine for the reasons that I gave before. Second of all, there’s going to be a reconciliation sweep. So, that’s going to find jobs that have become stuck for some reason, and it’s going to repair them. We’re going to

52:37 have safe retries. So, we’re only going to retry steps automatically if they’re marked safe, so that we don’t repeat any side effects. uh we’re going to run code in a sandbox so you don’t have to trust that code as much. We can have worker pools. That means that you can assign specific capacity to one or two workflows so that the other workflows you’re running don’t interfere with them. We’re going to have resource isolation and limits and waiting. When your workflow has to wait, it’s going to be a

53:08 stored timed job rather than living in memory. So, lots to come on the new engine. Um, stay tuned for that in the coming months. With monitoring, we already have a load of stuff in this area. You know, we have insights, we have log streaming, we have audit logs. We recently added open telemetry support a few months ago, but sophisticated users are running more than one instance of NA10. Hands up. Who’s who’s got more than one instance in their organization? Yeah, lots of hands. And you might be

53:40 doing that because you have a development um instance, you have a production instance. You might be doing that because different departments require separated instances. But what we’ve been hearing very clearly from you is that you need an observability layer that sits across all of them. And so with that, I present to you manifest. Manifest is a separate app that runs in the same ecosystem as your NA10 instances and it helps users with all the problems that automation creates

54:10 like what exists and is running, who owns it, uh which systems are being touched by which automations and we’re currently triing it. So if you are a multi- instance user and you’re interested in joining this trial and providing feedback, please do get in touch because we’d love to have you. And at first this is only going to be for NAN automations. Uh but we want to expand this to cover all kinds of automation in the or because we recognize that we do live in a world

54:41 with with multiple automation opportunities. Okay. And then finally onto the the manage side. So this is the human the organizational side of running a tool like N10. Many of you know just how significant that can be. Um, and because automation is exploding, the stakes become higher. So before, um, this is probably back in the 80s or something, but you know, shadow IT might be a spreadsheet macro. Uh, today it looks very, very different.

55:11 It’s probably, you know, a ton of agents with access to really sensitive systems. Um and actually governance we believe is becoming even more relevant for than before. Not just for the large companies that have always cared about governance, but also for the small companies because of this fact I mentioned before that you’re probably going to have more agents than you do have employees and they may be just as powerful. So we have a little framework for how we think about governance. It has four things. access, guard rails,

55:42 observability and quality and security. Access is about who’s acting and what can they reach. So that’s things like identity, credentials, permissions. Guard rails is about within that what is limited and what does a person decide. So spend and usage limits, policies, anonymization, approval gates, exception routing, this kind of thing. Ownership and quality is about who owns

56:13 it and can we keep it, can we trust it to keep working. So this is about managing the life cycle that I’ve been talking through. It’s about testing. It’s about review. It’s about environments. And these three principles together, they govern the system when it’s working as designed. The fourth one, security, is about what happens when someone tries to make it work against you. So this is things like prompt injection, tool poisoning, data exfiltration. Now, we have a lot in these areas

56:44 already. And don’t worry, I’m not going to talk through everything on this slide. I just wanted to to give you guys an idea. Um I want to just um pause on the security topic for a second uh and talk about security advisories because AI has changed the landscape in many areas but no more so than insecurity and as NA10 you know we see so many more tools that scan public code bases. Also, because NA10 gets more attention these days, we’ve seen a rise in in reports

57:17 and every report gets triaged and it gets fixed and we send you an advisory via email with enough information to uh for you to act on it. And we prefer that you hear it from us, you know, clearly and early rather than finding out some other way. And the good news is we’re seeing the volume of these things kind of level off over time. What we’re not assuming is that that trend continues. So, we’re continuing to invest in this. Um, but at the end of the day, NAN is

57:47 public code and what that means is we have many people looking at it, many eyeballs on it. Fixes are verifiable, so you don’t have to take our word for it. And to be honest, you know, that’s a trade-off that we would take again. Okay, back to um this framework. I just wanted to run you really quickly through some of the things we have planned in these areas. So under access, most agents today, they borrow a person’s login. So you can’t actually tell was it the human or the

58:18 agent doing these things. And the agent gets all of the permissions that the person has. We want to have our own identity for agents, which is narrow. It’s attributable. It’s revocable. Skim provisioning is about when a person leaves, do they do they lose access to the system or do their automations continue running? And user groups is about managing users or making managing users easier. On the guardrail side, we have policies and we have limits. Now, policies allow

58:49 you to stop things before they run and limits cap the damage after something goes wrong. For example, when an a when agents can loop and retry, they can sometimes get stuck. Costs can just run away unnoticed. And I’m sure that’s a familiar experience to many of you. So, you need hard caps to to set the bounds. on the ownership and quality side. Um you know software discipline is kind is about bringing that to automation and bringing it to agents too. Um and also

59:19 bringing the you know the rigor of evals also to the deterministic side of the house with with workflow testing. And finally with security I talked a bit before about code sandboxing. That’s not just about security. It’s also when you when you don’t have to trust the code you’re running you can be much freer with it. So you can load random libraries into your automation for example which is something we’ve wanted to add to N10 forever. Okay, so I’ve kind of talked about this already, what’s next. Um, but I just

59:51 wanted to give a really quick recap, you know, um, on the building side. We’re going to continue to improve the assistant. We’re going to continue to improve MCP and agents. There’s more context coming. There’s um, self-healing coming. On the auditing side, we want to make evals much more intuitive and easy to understand and basically less painful. On the deployment side, we have this new engine coming. Um, and while we don’t have a single point of failure today, we also want to make this even

1:00:21 more robust uh with a new engine. We have manifest for cross in cross instance observability and then we have a load of stuff on the governance side that I’ve just talked to. Okay, so there is a lot to come. the pace of change is increasing and NA10’s output is increasing with it and we’re very excited uh for what comes next. And on that subject, I do just have one other thing to show you.

1:00:52 [music] >> [snorts] [music] [music] [music]

1:01:54 >> Okay. [applause] So, this is a new product from NASA Labs. Um, if you are interested in checking it out and joining the weight list, please do uh take a photo and join it. We’d love to have you build it with us. And with that, you know, uh thank you so much for your attention uh and being a great audience. I hope you have a great rest of your day and you enjoy all the great stuff that we have planned.

1:02:24 [music] Good job, Becky.