Hey Reader,

I just listened to Tesla's 2026 Q1 Earnings call this morning, where Elon and others talked about Tesla's self-driving capabilities, where things are at, when are we going to have unsupervised FSD (Full self-driving), etc.

It made me think about the parallels on how we delegate operations and decisions to AI in our businesses, and I'd like to tell you a bit about where things are with AI, and what the future holds.

I know a lot of entrepreneurs reading this are a bit shocked by how fast things are improving with AI, because you were used to a certain rate of technological improvement in the past decades, but with AI, the rate at which things change is 10x faster.

So it's natural for business owners to feel a little left behind or scared, because the playing field shifted so much, your old playbook doesn't work that well anymore.

Self-driving cars

Self-driving cars are workers for hire. If a car like Waymo offers to take your from A to B, the offer competes with the promise of the actual human doing the driving.

They are responsible for what happens. And because of how many thousands or millions of miles these cars have proven to drive safely (or safer than humans), you trust them.

Currently, no production grade vehicle offers full autonomy.

That would mean you can be hands-off, eyes-off while the car does its thing. You are not driving anymore, you are a passenger.

What we have now, even with the best self-driving Teslas is that the driver needs to be ready at any moment to take over. They can take their hands off the steering wheel, but they can't take their eyes off the road (the internal camera monitors this).

This is Level 2 autonomy, and the liability is on the driver. See the table below, Level 5 is the full autonomy where you can sit in the passenger seat of your own car instead of the driver seat, and the car will take you.

Right now, the driver cannot fully delegate the task to the AI, because the AI is not capable enough. That means the driver needs to have more capabilities than the AI driving the car, in order to go from A to B successfully.

The task being delegated can already be done by the person wanting to delegate the task. The human is a supervisor.

Self-driving Claude

The same is still true for working with AI.

You can only delegate what you can supervise.

Expert programmers can actually ship working software, even if they use Claude Code to write the whole codebase.

But if you are a business owner who doesn't know anything about coding, you can't "hire an AI agent" to build software for you, because you don't have the ability to take over in case it goes wrong. You don't have the ability to project manage the AI, because you've never done that project.

I see this a lot when working with AI. When I delegate a task I've already done and I know how to do well, I can work amazingly fast with the AI.

But when it comes to doing something I've never done before, I get slowed down by a lot and I waste days.

So even though it feels like AI could replace a marketing agency, a copywriter or even a CEO, it can't (yet).

Because when you make the decision to hire someone, you shouldn't be micro-managing them, but an AI has to be micro-managed. Heck, sometimes it needs to be nano-managed!

You hire them because you also want to delegate the responsibility of getting the task done well, and delegate the overhead of managing the project.

Every project, every new AI chat, every single thing that is now possible because "Claude Design just killed designers" or whatever crap is the newest headline, also comes with AI management overhead that's invisible at first.

AI agents at this point are still stuck at Level 2 autonomy. We are still eyes-on, hands off.

Claude makes the presentation, we review.

Claude writes the contract, we review.

Claude makes a decision, we review.

What has changed recently with Opus 4.6 and 4.7 is that we need to chime in less and less, and the AI can work for longer and longer.

According to METR.org (Model Evaluation & Threat Research), the length of coding tasks AI could do was doubling every 7 months based on data from 2019-2026. That's the orange line in this chart:

However, if we look at data from models launched between 2024-2026, that doubling speed is down to 4 months.

So, when you popped the champagne this NYE, Claude Opus 4.5 (released in 2025 Nov), could do a coding task with a 50% success rate that was around 5 hours long.

When I'm writing this, the latest data is only available for Opus 4.6 (released in 2026 Feb), and the 50% success rate is for tasks that are 11 hours and 59 minutes long.

What this means is that you can send Opus 4.6 to do a task that would take a whole WORKDAY, and 50% of the time, it will successfully carry it out WITHOUT any intervention.

This is roughly on par with what I see from my work with these models. Opus 4.6 did make me a full presentation in one go that would've taken me roughly a day.

If the trend continues, we could see AI models that do month long tasks by 2027.

But the trend might slow down or it might accelerate even more, we don't know.

What this means for businesses

Well, first of all, don't be afraid that you'll "miss it". You won't. You'll see it everywhere, it's going to be major headlines, far more often than when ChatGPT came out in 2022.

The whole world is watching now.

Second of all, the chances of AI replacing you in 5 years is little, but it's a lot higher than it was a few years ago.

Your potential clients will have access to the same tools as you do, but they don't have the experience and the taste to direct these tools and make tough decisions.

So, you will need to rethink your pricing structure and your offers, because your costs to ship the same work will be lower. For example, it will take you 2-3 prompts to ship a full website instead of a week of blood, sweat and tears.

But for your potential client, it might be a day of prompting back and forth and still not knowing if it's right.

So maybe instead of shipping two websites a month for $10k each, you'll be able to ship twenty websites a month for $1k each, and still make the same revenue of $20k.

This is just an example, by no means I'm recommending you to drop your prices and start competing on price.

But this will be the reality, things are going to get cheaper & faster (they already are).

So if you'd still like to work with fewer clients at high ticket prices, you need to make sure your offer reflects that and add things that can't be done by AI, or can't be done by beginners in the field using AI.

You'll have to develop new offers, new services, and keep reinventing yourself like you did when we went from billboard advertising to Facebook ads, or when we went from TV infomercials to YouTube videos.

The only way I can guarantee you that you won't fall behind is if you keep yourself up to date with what's going on in AI and what the new tools do. You don't need to try them all, but read some of the news, watch the YT demos and follow your curiosity if you see something you like.

Things are going to get less and less technical, and I think over time a lot of the graphical interfaces will disappear and we will be just prompting AI tools that are connected to everything.

But for now, you still need supervision to work with AI, you need your eyes on the road and your seatbelts on.

Dave Talas

P.S: If you'd like to have a consultation about your business and how to use AI in it or how to prepare for what's coming, reply to this email and I'll tell you how we can work together.