Programming has changed: what I focus on now

Every day in my communities, people who are learning AI engineering or starting out as freelancers ask me some version of the same question. With all of these AI agent harnesses and agentic engineering developments, will the things I'm learning right now still matter a year from now? Two years from now?

Things are definitely going to change. But I don't foresee a future where we don't need software engineers, AI engineers, and data scientists at all. That's not what I'm seeing. The work itself is shifting, and you can position yourself ahead of that shift. This post is the same advice I share inside my communities on how I'm preparing, where I'm upskilling, and how I see things playing out.

The short version is to learn to help companies become AI-first. To explain what that means, I need to walk you through three concepts.

The three stages: AI-enabled, AI-first, AI-native

These three terms describe how effectively a company uses AI, and almost every business sits somewhere on this spectrum.

Most companies today are AI-enabled. Picture the anatomy of a business, with shareholders, the board, and the CEO at the top, and underneath them the operational layer where the work happens. In most digital companies, most knowledge workers now use some type of AI tool, whether that's just ChatGPT or far more advanced tooling. But from a workflow perspective, not much has changed. The business runs the way it ran before. Employees have simply become a little more effective because they have AI tools.

AI-first is the next phase. An AI-first business is optimized around AI as the way to fuel every process and every system. That means either re-engineering processes that were designed before AI reached its current level, or, for newer companies, engineering workflows from the ground up. The anatomy flips. The AI agents come first, and employees manage their day-to-day work through that agent layer. The agents take different forms. It can be someone using Claude Code. It can be tooling with a much deeper level of automation and integration. It can be an entirely custom AI platform, or even an AI operating system built for the company.

Notice the language change too. At the AI-enabled stage, I say AI tools. At AI-first, they really become AI agents.

AI-native goes one step further. An AI-native company is built and engineered from the ground up with AI agents in mind. Every process starts from the same questions. How can an agent do this? What data do we need? What systems, what inputs, what outputs? With that design, even less headcount can manage the company. Eventually we'll most likely see businesses that run entirely on agents. In today's reality, that's still really hard to do. There are probably already some businesses out there doing it, but at any reasonable size, you still need humans.

Why the move to AI-first is inevitable

Shareholders will force this transition. From their perspective, AI-first is very interesting. You get the same output with fewer people, and in some cases more. That means more revenue and more business growth.

So shareholders at companies small and large look around, they see AI-first companies, they see AI-native companies, and they simply have to keep up. Otherwise they get competed away. New companies have the advantage of starting fresh with the latest technologies. They move much quicker than a huge enterprise stuck in old processes and lots of headcount, and they can take that enterprise's market share.

In my opinion, this makes the shift inevitable, especially for businesses that operate in the digital world. Existing companies move to AI-first. Newer companies start AI-native.

The skill to bet on: moving companies along the spectrum

If you take that trend as a given, the way to use it to your advantage is to play along with it. Learn how to help companies become AI-first, and later AI-native.

The good news is that from a development perspective, the way you build AI agents for an AI-first company is pretty much the same as for an AI-native one. The difference is the process and how the business is set up, which is mostly the responsibility of the founders and shareholders. The engineering carries over.

That skill, knowing how to build AI agents that can automate entire roles and entire workflows, is going to be one of the most in-demand qualities of a developer. And it doesn't map to a single vertical. Being really good at AI engineering alone, or software engineering alone, or data engineering alone, won't get you there. The work demands a combination.

End-to-end automation pushes you full stack

In the age of AI, if you can only do one vertical, there's a high chance you eventually get automated away. You become the bottleneck in the system. The developers who stay valuable can solve problems end to end. They take a particular problem or workflow inside a business, identify the starting point, the data sources, and the process around it, and then build the automation or software that handles it completely.

End-to-end automations touch multiple parts of the stack by nature. You need a database. You need some type of deployment and a backend. Maybe a frontend. Maybe a dashboard, so the people managing the system keep some control over it and get human-in-the-loop interaction with the agents.

That production layer is its own discipline, and it's the foundation everything in this post builds on. For the technical path, start with how to build production AI systems.

Think in workflows, not headcount

There's a second skill beyond the engineering, and this part is new. When you come into a business, whether as a full-time employee or as an external service provider like a freelancer or consultant (which is what I've been doing for the past years), you bring two things. You can build these systems, and you can map out the processes and identify the workflows worth automating.

Most shareholders still think in roles and headcount. There's a problem, the company needs to grow, so the question becomes who to hire. That's an old way to think about it. In a world where AI agents can do a lot of the heavy lifting, you should think in workflows. Take the problem or process, decompose it to the actual workflow, and figure out how to hand that specific workflow to an agent.

The catch is that in real businesses, a single employee is often responsible for multiple workflows, and those workflows are rarely well defined. Someone knows that at the beginning of the month they need to create the report, pull data from one system, and check a second place if the data isn't there. Lots of edge cases, and humans handle them naturally. Usually this is the result of management never having everything dialed in. As a company grows, it always happens. John takes on one task, then another, then he hands part of it to someone new, and things get messy. That's the reality of how most businesses operate as they grow.

Data Freelancer

Technical skill was never the bottleneck

Most developers can already deliver work clients pay $150/hr for. What they're missing is an offer, a pipeline, and the first client. That part is learnable.

Learn the System

The audit: finding type two waste

The skill I think really matters beyond the engineering is being able to come into a business and run some type of audit. It can literally be asking questions, interviewing employees, and figuring out where the waste is going.

There are two types of waste. Type one is necessary for the process; you can't get around it. Type two is just waste you can get rid of. Duplicate steps, data manually moved from one place to another, people creating and shuffling stuff around by hand. Those are very good triggers to start digging in and say, I think we can optimize this using AI, and we can build a process around it.

You can do that effectively because you, as the engineer, know the capabilities of AI automation. You know what can be automated and what can't. For most people outside tech, AI is still one big black box. They know ChatGPT, maybe Claude, and that's where it ends. You can look at a process and say, are you really doing it like that? I can build a simple script in literally 30 minutes to automate that for you. That's the value you bring to the table.

That position has to be earned, though. If you're just working as a developer, it's often very hard to get enough information from a company to look at processes at that level. Developers are usually kept separate from the systems and processes of the business overall, and in larger organizations this can be really tricky.

Every company is becoming a software company

It used to be that developers worked for the big enterprises, in large development teams. What you'll now find is that almost every company will become somewhat of a software company. There will be a software component, because you simply need AI to manage things and survive.

That sets you up to step in as a developer at a small company too, literally the local business down the corner. Maybe as a regular job, maybe as a freelancer or outside consultant. Most business owners have no idea what they need to do with AI. They're still catching up, still figuring it out. Coming in and showing them what AI agents can do for their workflows is really going to be the play. And as we move closer to AI-native, that role only gets bigger.

That's the future of your tech job as I see it. The developers who come out on top are the ones who can help businesses move from AI-enabled to AI-first to AI-native. The hard skills are knowing how to build end-to-end AI automations, which generally covers the full stack. The rest is being able to identify the right processes and get to an ordering and prioritization of how to implement them. This is the recurring theme I'm focusing on myself, and it applies wherever you are in your career.

Next step

If you want to take the freelance route and sell solutions like this to businesses, but you don't know where to find that first client, read how to become a freelance AI engineer. I run a community of hundreds of freelance data and AI professionals, and we're all there to make more money, work on fun projects, and create freedom.

The full whiteboard walkthrough is in the original video, Programming Has Changed - Here's What I Focus On Now.

FAQ

Will software engineers still be needed as AI gets better?

Yes. I don't foresee a future where we don't need software engineers, AI engineers, and data scientists at all. The work changes. Less writing code by hand, more designing automations and judging where AI fits into a business. Demand shifts toward developers who can solve problems end to end.

What's the difference between AI-enabled, AI-first, and AI-native?

AI-enabled means employees use AI tools, but the workflows haven't changed. AI-first means the business is optimized around AI, with an agent layer as the primary way work gets managed. AI-native means the company is engineered from the ground up with agents in mind, with every process designed around what an agent needs to run it.

Do I need to be full stack to build AI automations?

Mostly yes. End-to-end automations touch multiple parts of the stack, including a database, a backend, deployment, and often a dashboard for human-in-the-loop control. If you can only cover one vertical, you risk becoming the bottleneck that gets automated away.

How do I find which workflows in a business to automate?

Run an audit. Interview employees, ask questions, and look for type two waste, like duplicate steps, data moved by hand between systems, reports rebuilt manually every month. Those are the triggers worth digging into, because they're the processes a simple automation can take over.

Can I do this kind of work as a freelancer?

Yes, and it's what I've been doing for the past years. Almost every company is getting a software component, including small local businesses, and most owners have no idea what to do with AI. Coming in as an outside consultant who can both map the workflows and build the agents is exactly the gap.