# AI Native DevCon: Shifting From Output To Outcomes

**Podcast:** The AI Native Dev - from Copilot today to AI Native Software Development tomorrow
**Published:** 2026-06-23

## Transcript

Hello and welcome to AI Native DevCon by TESOL.
This is the fourth edition of this conference, the second in person and the first in London.
We're at the brewery in central London near the Barbican Centre.
Let's go in and take a look at some of the expo hall and we'll give you a flavour of what AI Native DevCon is like.
Follow me.
The AI Native Dev is a podcast for developers and engineering leads at the cutting edge of AI and agentic coding.
Join your hosts Guy Pajani and me, Simon Maple, every week as we chat with the most exciting voices in AI and tackle the biggest questions facing developers today.
This is the AI Native Dev.
Okay, so here we are at the expo center of AI Native DevCon.
We're going to walk in.
It's a break.
It's busy.
Let's try and talk to some attendees.
So follow me and you can see the expo hall is packed.
full of people, full of attendees and full of sponsors.
There's areas to get some free shirts and the sponsors going right round.
So let's find someone to talk to you.
Follow me.
Tell us a little bit about yourself.
What's your name and who do you work for?
I have a couple of companies, but the reason why I'm here, I have a fintech startup I'm building alone.
Oh, nice.
Combining AI and blockchain, public and private markets.
So I'm here to learn what are the tools available to make my life easier to expedite the process and to make the production process output.
more reliable.
So looking for tools and practices and processes.
Correct.
What are some of the sessions that you've really enjoyed so far?
Max, the guy from...
Max Thurman?
No, no, Max Maximilian from Argentina.
Oh, yeah.
Web MCP.
Yeah, Max Thurman.
Yeah, yeah.
Max, I don't remember his last name.
He was very interesting.
Then Dave Farley.
Yeah.
Dave Farley with N-Wave.
So I'm planning to try all those.
tools in my stack and see what it can do.
Yeah, very interesting.
And one of the things I learned with Dave Farley is everyone has a different approach to doing agentic development and agentic coding.
And so some of the things that Dave was saying works obviously really well for him.
And there's some other people kind of like saying, oh, actually, I agree with this bit, but not that.
So I think it's really important for us to find our own way, but learn from others who are doing this at SCARE as well.
Correct.
And one more session that caught my attention yesterday was Matthias Lubeck from Pi.
The company is different, I forgot the name, but he's using Pi.
So I want to try that one as well.
Amazing.
Well, enjoy the rest of the conference.
Thanks for speaking.
Awesome.
Thank you.
So Chris, this is Chris Beatty, one of our speakers.
Chris, tell us a little bit about yourself and what you presented.
Yeah, so I've been an engineer for 20 years or so.
And what I've done over the last couple of years is I've been building a product team within a consultancy.
So we've hired new people, we've tried to build products.
So my talk was about our lessons learned of adopting AI in that team and what we're able to achieve, which we wouldn't have been able to achieve, I think, if we tried to do this five years ago.
Yeah, yeah.
And what are some of the key learnings that you would say if you were to give advice to someone who is on this journey as well?
What are the key things that you wish you could have told yourself?
I think avoid vanity metrics at all costs.
like code commits or how many tokens people are using.
Have a look at your flow of delivery, right?
From idea to someone using your piece of software.
Don't stop at like commit or, you know, outputs.
Think about the actual outcomes you want to achieve and find the friction, right?
Because it's not normally in code production.
For our experience, it's not even idea production, like product management.
We've been pretty good at that now.
The bottleneck for us is actually getting real people to use our software and quickly feeding back on those.
And how well are you finding, I guess, execs or senior folks within the engineering teams to accept those types of metrics over and above token usage?
Well, actually, when it comes to it, the higher up you go, if you can talk in the business outcomes, they love that, right?
Outcomes over output.
Yeah, exactly.
Outcomes over output.
So I'd say it's normally like middle management that are going for vanity metrics because they're easy to measure.
But if you say, hey, to your execs, to your higher ups, we can get these customers doing this by this date.
And that's the only measure of success.
And I don't really care how many tokens you use, how many commits you make, how many other features that you build.
I think we're all suffering from building features that are not being used because it's so easy to build features now.
In fact, if you can improve your outcomes as high as you can with as fewer tokens as you can use, even better.
And as fewer features in your product, really.
Absolutely.
Because we've all used...
I'm on camera, I'm not going to name a product.
Any sessions that you've been to that you've thought have been really amazing or have you mostly been focusing on the hallway track?
Both, really.
I mean, we just went to the session about measuring the impact of agents and AI development on...
organizations and like how people that can become low performers or high performers.
But overall, I think it's been like a really good conference with a nice balance of like, we need a bit of hype, we need to get excited about our work.
But there's been a lot of like practical takeaways as well.
So people like dealing with reality.
Yeah, amazing.
Well, thanks for joining us.
Thanks for speaking to us.
And thanks for the support and conference talking at Native Dev.
Tell us a little bit about a little bit about yourself about what you do.
Okay, I work at TrainGuard on the AI automation team.
I've been dabbling with AI stuff since co-pilots first started telling me what to do.
I tried to make that into a predictive text generator for functions and ever since I've been sort of captivated with making AI a productive tool for my development workflows.
So I'm here to pick up on what the current trends and you know fields are in the industry among everyone that's here.
Amazing and of course we first met, Alan Pope introduced us and we first met at Starbucks like 18 months ago, other bad coffee establishments are available.
But from when we first chatted to now, what would you say are some of the biggest changes in your working environment and style?
Well, I think when we spoke back then, we were talking about the capability of the models.
And then shortly after that, the whole sort of MCP and a Gentic thing kicked off.
So I think it's really the tooling around the models that has changed the most.
Obviously the model capability has improved significantly, but I think the force multiplier has been the tooling around the models.
So skills, MCP servers, they've really sort of delivered sort of frecency of information to the models that really make them much more capable.
So I think that's the big thing for me.
And other than my conference opening, what would you say is your highlight for the sessions?
The sessions?
I don't know.
That's a tough one.
I think there's been a couple of the exhibitors that are interesting.
So the virtual SRE I really like.
It looks inside your existing observability stack and acts like a historical SRE, looking at where the issues were and automatically proposing fixes.
So I kind of like that.
Which one?
Which one?
Is it Autonomy AI?
Autonomy AI.
Awesome.
We'll talk to them in a bit.
Martin, thank you for spending time with us.
Cheers.
What's your name and what do you do?
My name's Ryan, I'm a QA team lead.
QA team lead, very interesting.
What are the things that you've come to the Our Native DevCon to learn or practice more about?
Yeah, I've come to potentially see gaps, like from what I know and what I don't know, and I've identified a few gaps.
Awesome, would you mind sharing a couple of those with us?
Sort of harnessing or harnesses as a whole, that's a new concept to me.
and evals is something I'm very interested in.
Evals for your context.
Exactly.
What are the kind of things that you feel like you can take away and try straight away?
The evals for sure.
We've got skills spread all over the place and kind of want to consolidate and look into the evals.
What would you say are some of the biggest problems around skills?
Is it distribution?
Is it duping of skills?
A bit of duplication.
But I'd say at work, it's just skills coming in too quickly and more on the mid to junior levels.
knowing about the skills and how to use it so it's a bit of it's consolidation in my opinion.
I guess it sounds like a little bit of education about education.
That's correct.
Anything else you're looking forward to?
Anything else you got your name by in the workshops or sessions?
Next workshop is the test.
Yeah, I can't remember the name of it now.
Yeah, it's a test strategy, I think.
Oh, awesome.
Yeah, yeah.
Is that one with Baruch and Macy maybe, perhaps, from Tessel?
Yes, I think so.
Awesome.
That's going to be a great workshop.
Excellent.
Thank you very much for speaking with us.
Enjoy the rest of the conference.
But we have Manny Sarkar here, long-term supporter of the AI Native Dev, and it's great to see you here at the conference as well.
So Manny, tell us a little bit about your background.
Tell us a little bit about why you're here.
What are you trying to learn at the conference?
Yeah, definitely.
So I'm a software engineer turned AI slash ML engineer.
I've got a history of 25 years of software engineering, and in the last 8 to 10 years, I switched to, or I started taking on data science and machine learning a lot more seriously.
I leverage my software engineering experience to learn and implement all of that.
And so today what I do is, if I solve a machine learning or an AI problem, I wear my software engineering, my best practices hat, and I solve it from that perspective.
Interesting.
What would you say are some of the best practices that you feel an agentic coder of today could very easily drop or miss when they're trying to build out their agentic...
workflows and processes.
Yeah, so I was just talking to one of the attendees here.
I just really interrupted you.
Yeah, yeah.
No, that's okay.
I think it segues very well into what we're saying.
So what's happening is these days is we are reading less and we are passing on the buck over to the other side, which is non-deterministic.
Yeah.
We need to build and discipline ourselves with deterministic processes and steps and try to reduce the non-determinism.
So...
using a whole swarm of agents to build something, but then not knowing what it has built.
So you don't know how to use it.
And then that's where it's one of the things.
The other bit is like not having a plan and just like, you know, wipe coding itself is like just writing prompts after prompts isn't the greatest practice.
You plan your work.
So like I was telling the attendee that I remember Einstein's quote, he says that if you give me a problem and you give me an hour, I'll spend 55 minutes trying to think about the problem, trying to brainstorm the problem.
And I'll spend five minutes.
trying to solve the problem.
Now, Einstein, great mind, has done some great work or lots of great work.
We can learn something from that and then we can apply it in our space.
Yeah, yeah.
Amazing.
What is some of the, what, for example, are some of the sessions that you really took things, good practice?
So guys' session kicked off really well.
I love, and I'm going to randomly name them, but Patrick's session was amazing.
I'd love to work with someone like Patrick who knows how he needs to run, how teams need to be run while we are in the AI space.
Dave Farley really closed it off well yesterday.
And then there are some other sessions where I took little bits of things like do treat your current practices just like you were before AI came in.
Don't drop those practices.
What we've been given is AI to augment our skills and not to pass it on to AI to do all of the stuff and not know what's happening.
Because it's a great steer, it's a great skill, a great way to make us 10x but we need to be constantly disciplined have a have a plan have a structure and use it in that manner and not slip away by a lot of people who are doing by coding billions of lines of code and not knowing actually what it has created yeah it's locked basically yeah yeah amazing well mani thank you very much for your time So kid you're the founder of near fall.
Yes, tell us a little bit about what you're learning here from attendees some of the conversations that you've had at the conference.
Oh my god I've learned a lot about Harness engineering right well I've been here It's it's crazy from when the conference was announced to now like it's become a thing like and you know, it wasn't even a thing like three months ago This is the horrible thing about CFP for a conference like this because you open a CFP and then something that happened, you know, we opened it three four months ago And the content and the things that are going to be hot, you don't know what's going to be hot by the time the conference comes around.
So Harness Engineering, Tokenomics, everyone's going like, how are we going to afford to pay for this stuff in a few months time?
And just a really vibrant community here, you know?
And Neofom, obviously, a consultancy that come and help people in a very agentic way, which is super, super cool for a consultancy to really be pushing that.
Tell us a little bit about what some of the problems, some of the things that your clients and customers are asking from you these days.
Yeah, we're doing a combination of helping companies to start doing things in an agentic way.
You know, you have in most companies like one or two teams that are doing things in a more advanced way and it's like, well, how can I get to 100 teams?
So we're solving that problem for them.
So, you know, we're coming in, we're doing kind of hands-on training, workshops, building with the customers.
So we technically, you know, pick a couple of Lighthouse projects, work with their teams and get them into a higher confidence level of what they're doing.
So doing a bunch of stuff like that.
And then we're doing some applied AI stuff as well.
So we have an interesting use case for an AML project with a financial institution.
So AML departments are creaking at the seams to do KYC stuff and all that.
So I've got an agent to do that stuff recently.
So I think we cleared like a thousand AML cases in 20 minutes.
Wow.
Whereas that would have been about three to four thousand hours of human time to do it.
Amazing opportunities for scaling and being more efficient and effective.
Yeah.
The other stuff we're seeing is around modernization.
So we have a fun one with a client where a couple of years ago we were...
We budgeted a modernization project to be like, it was over 10 million euro project, big old code base, been around a long time.
And we kind of speculated up free AI, and recently we've kind of looked at doing it, and it's coming in well under 2 million to do the same project in about 35% of the time.
So you've seen this massive acceleration.
in reduction in the amount of time it takes to do something.
So we're seeing it from a consulting perspective.
It's a huge opportunity.
Amazing.
Have you found the crowd here at Age of Devcom?
Oh great.
Lovely community, very chill there it was.
And yeah, people are very nice here, you know.
Amazing.
So really enjoyed it.
Wonderful.
Well, thank you very much.
And thank you for all your support for AI Native TechCon.
I think it's a really interesting diversity in where people are at with their adoption of AI.
Yeah.
From the kind of cautious, but curious end of the spectrum through to the kind of YOLOing.
And I think my take on that is there's a lot of wisdom in both ends.
You know, people are rightly cautious with wisdom that's been hard-won over the years.
They're looking for something that helps them to ensure that AI stays in the right track.
But I'd be able to unleash the speed of AI without just letting it do whatever and just accepting the risk.
What would you say are some of the biggest problems that people have in their organizations today where a mature security solution, which is very AI first, can help people with?
A lot of it comes down to trust.
and how you can trust that things are going to stay on track without having to be hands-on and watching everything.
And so I think where Snyk has a lot to offer there is providing the trust, is giving a solution that people can rely on Snyk to ensure things don't go wrong.
And don't feel like they have to be hands-on, you know, gripping the wheel really tightly.
At scale, I presume, as well, across the organization.
It's got to scale.
It's got to scale.
It's got to be codified.
It's got to be something that can fit into the workflow, again, without slowing things down, and actually integrated in the way that flows.
Awesome.
Great to see Sneak here.
Great to see him supporting the AonAid dev community.
It's been really fun.
Amazing.
Thank you very much.
So Adir, tell us a little bit about Autonomy.
You're the founder and CEO, I believe?
CEO?
Yes.
Yes, there we go.
I didn't just promote you, did I?
No.
The founder and CEO of Autonomy AI.
Tell us, what is Autonomy AI?
So Autonomy AI is basically an operating system on existing code bases.
We enable, we understand existing code bases, understand the components, the design system, all the design library, and then we enable non-technical or semi-technical individuals.
Think about your PMs, your designers, your professional services.
to be part of the delivery process.
Instead of going back and forth with engineering, they can deliver everything end-to-end on your production code.
And what would you say are some of the biggest differences between a non-technical person and a technical person when it comes to them doing work in the agentic workflow, raising PRs, all that kind of thing?
What are the differences in needs?
Well, let's start from the basic, the interface.
All the interfaces you have today are usually CLI interfaces, right?
And today what we enable is a web interface that is super easy.
That's the first thing.
The second thing, in order to actually push a PR that is valid, viable, in a consistent, reliable way, the PMs don't know what are the constraints of the platform itself.
What autonomy can help them with and what Face Studio, our product, is helping them with is just click a button.
The pull request will understand the entire system.
PMs need to know how to ask visual things.
KPIs on the product not how to build a code behind the scenes.
Super interesting.
Of course we had Tomas, your CTO here, in doing one of the sessions.
When you're talking with some of the folks here at the booth or in sessions or on the hallway track, what are people's biggest problems today that they're facing with their agentic coding movement?
Everyone is spending so much money, so much effort on this, but not actually seeing any needle movement.
The needle movement is very, very small, around the 5% to 10%.
And that's the main challenge.
And when we talk with people, we don't actually just talk about our product.
We talk about the actual workflow.
Where are the real bottlenecks?
And we start exposing this.
They understand that the bottlenecks are not, you know, the coding velocity, writing code.
It's actually the workflow that you have within the organization.
And in one word...
What is the thing that people should focus on that they're not doing, spending enough time on?
Change management.
Utilize those tools, not just to write things faster, but actually how you can orchestrate your organization faster.
Change management.
And we'll call that with a hyphen.
What are the biggest problems that people in organizations are having today in and around their agentic coding journey?
Who wants to take that one?
So we've been talking to a bunch of people who are getting very excited about agents and then installing...
them all over the place and making lots and lots of skills and then they're a bit nervous about all the skills so they're either worried about security or they're worried about how they keep track of all of the different ones and work out how to share them properly and not lose track.
So I've been hearing that a lot from different people.
So sharing them through distribution, being able to actually validate them and understand.
If they're good or if they're secure, those types of things.
Have you been to many sessions?
I'm excited to go to my first session today.
Oh, which session are you going to?
I'm seeing Don's time talk on automation in GitHub.
So I'm really looking forward to that because I had a chat with him yesterday and he kind of gave me the gist.
And I think he has a point about GitHub and GitHub Actions actually being a big part of automation in the future.
So I'm really curious to see what else he has to say about that.
How ready is Git and GitHub, would you say, for an agentic coding workflow of today?
Let me put this questionable question to you.
Contentious question.
I think...
They have the opportunity, but being able to seize that is different.
What I mean by that is when you've got the core primitives of, like an action is just basically running a workflow in the cloud that the user is able to define.
And so that as a core primitive feels very valuable when you've got an agent that you can now run as part of that workflow in terms of the sequencing, in terms of forking, in terms of branching.
But to be able to capitalize on that.
They need to make sure that it's reliable and it is effective and that they have built the interfaces for people to do that, particularly given that code is now so cheap that people can go build their own versions themselves.
And so having the ease of onboarding, so much so that it is easier for me to get an agent to use GitHub Actions rather than for me to go have an agent write its own orchestrator on my own system, I think is kind of the important tension point there.
Amazing.
And we'll wrap up with one final question about a very interesting early access.
that Tesla announced here at Gaipo's session in and around an agent harness.
Tell us a little bit about what that is, what people can expect.
Yeah, so for Tesla agents, I think the main framing is...
people are going to need to shift up the abstraction sack to be doing harness engineering rather than just telling an agent what to do.
And to do that effectively, you need to set up your environment so that the agent can run effectively in that.
You've got hooks, you've got context, you've got verifiers, you've got all these kind of core concepts that you can use to make sure the agent is producing the correct code.
The Tesla agent is there to help you set up that environment and set up your ecosystem and set up your code base for a genetic success in it.
So it shifts your human leverage point to the point where you have the most value in terms of defining how the agent should be operating rather than telling the agent what it should be doing because that's something that can be much tighter, much more easily encapsulated in a ticket.
Amazing.
And if you're interested in that, tesl.co forward slash agent and you can learn more about how to get onto the early access list.
So that's a wrap for AI Native DevCon 2026 in London.
I'm here in the VIP room where we essentially let speakers and some of our VIP attendees hang out, grab good coffee in nice comfort.
I'm pretty tired now.
My mind is absolutely full of amazing information from tons of great speakers, tons of great conversations in the hallway track.
I think it's time to maybe have a little rest and then start putting all that great knowledge into practice from tomorrow.
Thanks all for watching.
Thanks for attending.
And hopefully we'll see you in New York or London next year.
Bye for now.
The AI Native Dev is brought to you by TESL, the package manager for skills and context.
Your hosts are Guy Pajani and me, Simon Maple.
Our producer is Tom Dowler.
The AI Native Dev is not just a podcast, it's a community.
And we host monthly meetups at the TESL offices in central London.
Visit tesl.io forward slash community to learn more and I hope to see you there.