This week’s newsletter is sponsored by Unblocked.

[Webinar] 8 levels of context maturity in AI-native engineering

AI shows up in 60% of engineering work. But only about a fifth of it can be handed off without someone babysitting the output. That’s because agents are missing context.

This 8-stage context maturity model gives a real answer on why you haven’t seen meaningful productivity gains for all the tokens burned.

Join live June 24 (FREE) to learn:

• Why more MCPs provides agents access but not understanding

• What it takes to deploy agents you can trust without supervision

• How a context layer solves for quality, efficiency and cost

Register now

Thanks to Unblocked for sponsoring this newsletter, let’s get back to this week’s thought!

Intro

Last month, together with my friends from Augment Code, we hosted an event called: Engineering Leadership LIVE in San Francisco. It was a blast, and there were so many insightful discussions we had!

As part of the event, we also had 4 talks.

- Gregor Ojstersek, CTO & Author, Engineering Leadership newsletter
Talk: AI-Native Engineering Leadership

Full overview of my talk in this article:

AI-Native Engineering Leadership Gregor Ojstersek · May 28 Read full story

- Vinay Perneti, VP of Engineering, Augment Code
Talk: We Thought AI Transformation Was About Adopting Agents. We Were Wrong.

- Andrew Churchill, CTO, Weave
Talk: What Actually Works: AI Coding Patterns from the Top 1% of Teams

Full overview of Andrew’s talk in this article:

What the Top 1% of Engineering Teams Do Differently with AI Gregor Ojstersek · Jun 15 Read full story

- Anwar Haneef, GM & Head of Ecosystem, Canva
Talk: Your Product’s Next User Might Be an AI Agent. What Engineering Leaders Need to Know.

Today, I am sharing the overview and the recording of Vinay Perneti’s talk at the event.

Recording of the talk at the event

You can watch/listen to the talk below, or you can keep reading for the insights.

Let’s start!

The November 2025 moment

You might remember the release of Claude Opus 4.5 and the big shift it had on the industry. This is what Vinay mentioned in his talk:

“In November 2025, we were working with Anthropic to launch Claude Opus 4.5. At the time, it just felt like another model. The evaluations looked good, but nothing about it seemed extraordinary.

Then, in December, something changed. I had regular 1:1s with the team, and independently, several of our best software engineers told me the same thing: it no longer made sense for them to write code by hand.

That caught me off guard. These engineers had already been using AI agents for a long time. So I asked, “What changed?”

AI agents got a lot more reliable. The team mentioned that AI agents had become much better at following instructions reliably. So, they started trusting them much more.

Hallucinations had also dropped significantly, so they could confidently let agents work on tasks for much longer without constant supervision.

Now, let’s go through a very important message next.

The exponential trend of LLMs becoming better

The chart below shows the difference between models, based on the longest real-world software task it can reliably complete 50% of the time.

Even though the chart looks linear, there’s actually exponential progress, as you can see how big a difference it is between GPT-3.5, which can reliably do a 36-second task, versus Claude Opus 4.6, which can reliably do a 10h+ task.

Some people even call this super-exponential progress.

So what does that mean in practice?

If this trend continues, by the end of the year, we’ll have agents that are roughly 10× more capable than Opus 4.5.

Based on the data, that’s the trajectory we’re on. Now, let’s go more into the AI transformation journey.

Engineering productivity after adopting AI agents

In December 2025, when engineers started to regularly mention that it doesn’t make so much sense to manually write code anymore, the expectation for many of the engineering leaders around the industry (Augment included) was to get a 2-3x productivity increase.

But in reality, what they saw at Augment was the 20-30% increase. The reason? The main reason was the bottlenecks.

AI transformation doesn’t happen with just individuals transforming, you need a whole team to develop the same mindset. You need the whole system to change and go after the bottlenecks.

It was a wake-up call for Augment, and they experimented with a lot of different things in order to become more productive over time.

Let’s go through the 3 myths and 3 learnings from their AI transformation journey.

Myth 1: “Al-Native means everyone is using agents”

Most of the companies these days would say that they are “AI-native”, but in reality, there is a big difference between using ChatGPT to get answers and orchestrating different AI agents to do a certain work in parallel.

At Augment, they had some engineers chatting with an AI agent, and some people orchestrating multiple AI agents.

Here are also the 4 levels of being AI-native:

It’s a totally different way of working if you are on a level 1 than if you are on a level 3 or 4. You need totally different systems in place between these levels.

The message from Vinay is very important here:

Being AI-native is not a destination, it’s a journey, and the journey keeps changing as the capabilities are evolving.

Myth 2: “You can drive the AI transformation top down”

We saw companies trying to do a top-down AI transformation in 2025 everywhere. The most prominent example was the memo from Shopify’s CEO. I’ve also extensively written about this back in 2025:

Enforcing the Use of AI in Engineering Teams - Good or Bad Thing? Gregor Ojstersek · April 13, 2025 Read full story

“Everyone is adopting agents”. “This is our OKR”. “We are going to track adoption”. “I am going to make this a part of the performance evaluation”. These have been some of the common sentences from companies back in 2025.

And what was the outcome?

People did adopt the new way of working, but organizations did not make the full AI transformation.

The AI transformation journey is not just changing 1 behavior, it’s about changing the whole system. And at the same time, it’s human nature to resist change, it’s an even higher urge to resist if something is forced upon you.

If you’re forced upon something, the first time you’re going to run into a problem, you’re going to go to your manager, and say: “I told you this wouldn’t work”.

In order to have a successful AI transformation journey, you need a bottom-up initiative.

And the reason is that when a certain issue comes, or a new bottleneck emerges, people don’t wait for top-down direction. Instead, people look to resolve such things themselves, collaboratively.

This principle is important everywhere, but even more important in the AI transformation journey, because the bottleneck keeps changing over time.

Myth 3: “This is entirely technical problem”

The biggest mistake an engineering leader can make is thinking that AI transformation is entirely a technical problem.

“Let’s bring the new best tool”. While a tool is important, it’s not enough. A tool will just help you with the journey, but it won’t make it successful. You need to get the buy-in from the people.

As part of the recent report done by Augment Code, they asked 219 engineers and engineering leaders the question: “What are you fearing the most right now?”.

89% mentioned that they fear that their skills will no longer be relevant.

Basically, the majority of the people (9/10) are feeling this, and they probably haven’t voiced it enough. And if someone is in that mental state and it’s not acknowledged, their willingness to embrace something new is going to be very low.

As an engineering leader, you really need to think about this side as well. You can’t just bring a new tool to the team and expect that there’ll be a successful transformation.

Now that we have gone through the 3 myths, we’ll go through the 3 learnings from Augment’s AI transformation journey.

Learning 1: Slow down to speed up

This is a great analogy that Vinay mentioned:

“Even a Formula 1 car has to slow down when it makes a turn”.

And that is true for all the organizations. If you don’t slow down, you might be going full speed in the wrong direction. This can be very problematic (even more so these days, as the speed of building has increased).

At Augment, they slowed down and took 2 days offsite for the whole engineering team, and did the following:

• Day 1: Hackathon

The theme of the hackathon was: 10x your agent, 10x your team, and 10x yourself. With one additional constraint which was: Whatever you’re building, build it entirely with AI agents.

That’s how the team was able to experiment, share their ideas, and implement them as well.

• Day 2: Human side

Vinay mentioned that day 2 was a LOT more important, and that day 1 was actually preparation for day 2.

They opened day 2 with Mentimeter: Live question being asked and put on the screen: “What feelings are you going through now after yesterday’s hackathon?”.

And then asked everyone to answer it (anonymously), and the answers showed up on the screen. The 2 biggest themes of the answers were:

• Fear, uncertainty

• Curiosity, excitement

And an interesting thing happened after this:

Everybody felt like they were not alone, feeling that way. There’s something very powerful about seeing on the screen what you’re feeling, and seeing that everyone feels the same way.

Secondly, Vinay asked the team another 2 questions:

1. “What do you think we need for AI agents to be more productive?”

2. “What are some of the new behaviors that we need to adopt as a team?”

They created 30-minute breakout groups to think about these 2 questions, and everyone came up with very interesting conclusions like:

• Verification is the biggest bottleneck

• Producing the high-quality spec is the highest leverage work a human can do

• Code review is also a big bottleneck

• Agents amplify both good and the bad

Learning 2: The throughput of a system is governed by its slowest link

There’s a book called The Goal, by Eliyahu Goldratt, which talks about the importance of always finding the slowest link and working to improve it.

If we look at software engineering throughout history, coding has been a bottleneck, until now. And as agents get better and better at coding, the bottleneck shifts to something else. You need to actively be thinking about that.

This is what they found out at Augment regarding the bottleneck shifting over 2025 and today:

Learning 3: You need a system

As we know, smaller AI-native teams are the preferred way of working these days. And these teams usually work with a large number of AI agents.

And in order to have that way of working possible, your system needs to evolve as well. You can’t expect to just create smaller teams and put 1-3 engineers together working on different projects and expect that things will remarkably change for the better.

In order that the change becomes successful, your system needs to evolve as well.

3 main takeaways

1. AI-native is a system redesign problem

Al-Native transformation journey is so much more than adopting AI agents. You can’t just give the team a new tool and expect that things will magically work out for the better.

At this time, it’s really important that you show up as a leader: be supportive, provide space for the team to experiment, and give them the psychological safety, so they can make mistakes and learn from them.

2. Tackle bottlenecks in your system, one at a time

If coding is not a bottleneck anymore, think about where the bottleneck is going. Figure that out and focus on resolving it.

3. AI transformation is a part human and part technical

Everybody in software at this time is going through an identity crisis, so you need a way to acknowledge that and be supportive on this journey. That is very important to keep in mind.

2 questions for reflection

2 very important questions for you to reflect on. Make sure to block some time!

1. 6 months from now, when agents are 10x better, what does your org look like to take full advantage of that shift?

2. Where does human involvement have the highest leverage?

Last words

Let’s end this article with the following:

The teams that succeed in the AI transformation journey are not the teams that have the largest number of AI agents. The teams that succeed are the ones that adapt their systems so that humans and agents can work side-by-side effectively together.

Special thanks to Vinay for sharing his insights in his talk at the Engineering Leadership LIVE event in San Francisco.

Liked this article? Make sure to 💙 click the like button.

Feedback or addition? Make sure to 💬 comment.

Know someone that would find this helpful? Make sure to 🔁 share this post.

Whenever you are ready, here is how I can help you further

• Join the Cohort course Senior Engineer to Lead: Grow and thrive in the role here.

• Interested in sponsoring this newsletter? Check the sponsorship options here.

• Take a look at the cool swag in the Engineering Leadership Store here.

• Want to work with me? You can see all the options here.

Get in touch

You can find me on LinkedIn, X, YouTube, Bluesky, Instagram or Threads.

If you wish to make a request on particular topic you would like to read, you can send me an email to info@gregorojstersek.com.

This newsletter is funded by paid subscriptions from readers like yourself.

If you aren’t already, consider becoming a paid subscriber to receive the full experience!

Check the benefits of the paid plan

You are more than welcome to find whatever interests you here and try it out in your particular case. Let me know how it went! Topics are normally about all things engineering related, leadership, management, developing scalable products, building teams etc.