What Becoming AI-First Means In Practice
Last week I sat down with a CEO who told me he wants his team to move faster to make a shift towards being AI-first. I asked him what he meant by it, and he wasn’t completely clear.
I don’t blame him. We all feel the disruptive potential of the technology, but we’re also tired of vague videos about how the future is going to be different thanks to AI. I agree that it’ll be different, but what does is mean today, for that particular business?
Here’s how I think about it. There are three metrics that can be used to discuss the progress towards operating as an AI-first company:
Strategic: is AI a tailwind?
Human: has everyone made a shift from IC to agent manager?
Technical: are we writing close to 100% of the code using AI?
Let’s dive into all three.
Is AI a tailwind or a headwind?
AI is going to be either a tailwind or a headwind for your business. Very few businesses will be completely unaffected by the impact of AI.
A tailwind means that as AI models get more powerful, the value delivered by your business automatically increases. For example, if you’re in the business of offering AI customer support agents, your agents likely get better as underlying models from frontier labs improve without you lifting a finger. AI becomes a tailwind, a wave you’re riding. You can’t wait to try Claude Opus 5.
A headwind means that as AI gets more powerful, your business delivers less and less value. For example, if you’re running a human call-centre, more powerful AI makes AI-powered solutions better and cheaper, making your solution less valuable compared to the rest of the market. For you, AI would be a powerful headwind. You dread the release of Claude Opus 5.
Becoming AI-first means, above all, formulating a company strategy that clearly explains how improvement in AI models and tools directly leads to more value being delivered to your customers.
It needs to happen without any intermediate steps1 and be easily explainable to everyone on your team without any management-speak like “synergy” or “augmentation”.
Is Every IC a Manager of Agents?
If you’re a CEO or a manager, you probably take your management skills for granted. You have years of experience setting goals, checking the results, providing context, resolving conflicts, navigating a multitude of working relationships and, above all, explaining clearly what you’re looking for.
For most individual contributors (IC), that is, most people on your team, this is a new skill to learn. And now, the shift to AI-first is requiring them to think and act like a manager with as many reports as they can handle.
This shift requires learning new skills, but above that it requires a shift in identity: away from someone who is valuable because of what they can do towards someone who’s valuable because they can get their team to do great things.
Some people will lean into it, but many, maybe the majority, will grieve and regret the loss of a sense of safety and familiarity that their old identity offered.
That grief will probably manifest as quiet resistance. Not out of stubborness, but your team will keep doing things the way they used to out of desire to feel okay. If top management responds with force instead of understanding, it’ll likely result in good people leaving the business for no good reason.
What needs to happen instead is taking the entire team on the journey. The business needs to start with the strategic AI tailwind question and then translate it into language that will resonate with everyone for their own reasons.
The outcome, if this work is done well, is a shift towards everyone acting as a manager of AI agents, leveraging AI to maximise their unique human contribution.
Is Nearly 100% of Code AI-produced?
Software engineering is, at this point, largely a solved problem. It doesn’t mean that every company can just buy a Claude Code or Codex license and fire their engineering team. Far from it. However, it is beyond doubt by this point that going forward the craft of software development will consist of defining what needs to be done and how, and then orchestrating AI systems to deliver it.
There will be exceptions, of course. But just like today very few software developers are writing assembly code2 by hand, going forward very few software developers will write any code at all by hand.
So one of the metrics that we should use to discuss whether a company is shifting towards an AI-first future is whether its engineers are focused on building systems that then write code instead of writing the code directly.
There three metrics aren’t the only one that matter. There are questions about security (where is your data going?), data (is it accessible to AI systems that need it?), ethics (what does responsible use of AI mean in your business?), choice of specific platforms (is a migration to Google Cloud because of Gemini worth it?) — all will need to be addressed during a shift to the AI-first future.
Yet the three most important questions are about the AI tailwind strategy, the shift from ICs to managers and reinventing the entire software development process.
Bad example: “Our team will use AI to optimise internal processes, so that we can deliver better products”. That’s not it. That’s slightly improving a non-AI-first business model, not reinventing it as AI-first.
Low-level machine code that all programming languages are translated into before the code is executed. Decades ago, software developers were reasonably expected to read assembly, if not understand it. Today, very few people need to: the rest use high-level languages like Go that give them leverage. AI takes that another step forward.