I recently went to a seminar about the use of AI in the workforce. Part of the description of the event was some homework that could be summarized down to having Claude installed on your laptop, so you could run Claude Cowork. My thoughts were that we'd learn how to summarize your email, organize your desktop, or get an agent look into your calendar to plan for the coming days. Safe to say that my expectations going into it weren't high. I really only went, because my girlfriend had shared the event with excitement.
Quite fast I realized that this was not that type of event, however. Instead of the basic introduction to a tool with system access, I was met with genuinely inspiring speakers, an academic point of view on AI's effect on the workforce, and a startup founder who was experiencing what could only be described as AI euphoria.
I left the seminar - one I'd had no interest in - feeling heavily inspired and excited about the possibilities of the future. We arrived home late that evening, but it stayed with me for some time. The next day, out of excitement, I started writing a formal description of a position that I believe my workplace - and any workplace for that matter - will need, if they want to stay competitive in the future.
However, stepping out of the infectious AI euphoria and back into real life hit me hard. What had started in excitement, quickly grew to a realization of how these grand predictions of the future rarely manifest as quickly as one would have hoped. Being met with the same barriers that the job description I was working on should erode, served to calm the excitement I was feeling in the moment. It's strange how, depending on the social arena you find yourself in, the conversation of AI in the workforce can both feel like the most natural thing to discuss, or an isolating thought that you alone are reflecting on.
My excitement was slowly but surely stopped, and I went back to making sturdier and more efficient processes locally, while my grand plan of establishing something bigger was put on ice for a while. This was until I had a conversation with my nearest manager, and shared that I had been working on this description, and he asked me to finish it. It gave me a bit of that excitement again - not because it's a job I necessarily want, but it's a person that I am convinced every company will eventually need. The result was a brief, six-page write-up that I believe reflects some of the truths of the AI era, as well as some misunderstandings people have of how this new technology is actually useful.
My main thesis is that the frontier models we have nowadays aren't just useful because of their perceived intelligence or capability. On the other hand, I would argue that they are incredible in their ability to make knowledge more widely available. I envision a future, where each team is independently building the tools they need, without any technical competencies. Where it previously had to be governed by a central team with capabilities, these capabilities are no longer needed for the actual production of solutions, but simply for validating the governance of the tools created. Knowing better how to build tools solving the pain points the user is experiencing is a capability that only few of us have - and none of us has the time for. Especially with frontier models in the hands of average employees, genuinely useful tools are only a few minutes away, capable of deleting hours of work a week.
But the model itself was never really the hard part on a local level. Whether a company gets anything out of all this comes down to two things sitting outside the model - whether you can reach it at all, and whether the organization lets people use it.
The best part of leaving the creation of tools up to the users is the joy and pride I see amongst the people I know who have started building. Instead of having to wait months or years for a central department - they build it themselves. It's there when they need it, it evolves as their needs do and it's capable enough to be of genuine help. It might not be perfect, but it's theirs.
Fable 5 was released earlier this week, and it's the first time I've felt that "wow"-experience since the release of Opus 4.5 in late 2025. That might sound recent, but with how fast this space moves, it feels like ages if you keep up with AI news. Especially when it comes to coding or other technical tasks - not to mention the huge upgrade to front-end work - this model just feels more competent. Despite using twice the tokens of Opus 4.8, it - in my experience - does the work faster, better, and more efficiently. I used to burn through my usage balance in half an hour to an hour, but now I'm solving multiple complex problems inside a single 5-hour credit window, faster and with more condensed, precise answers. It felt great.
In that regard, Fable 5 being the first of the Mythos-grade models released feels like a genuine leap ahead. But to me, it also underlines a genuine flaw in how we look at this new technology. Days after its release, the US government issued an export-control directive suspending access to Fable 5 and Mythos 5 for any foreign national - inside or outside the US - and to comply, Anthropic had to disable both for every customer, paying or not. Anthropic disputes the decision and says it's working to restore access, but for now no amount of money brings it back. A model that felt like a leap forward one week was simply gone the next, by a decision none of its users had any part in.
This fragility in access really highlights one of my key approaches to the AI conversation. It's not about the technology itself or the intelligence it can sometimes display, but it's much more about the tools it allows us to build. Building your business on using a technology you don't own, that can be taken away at any time, is completely unsustainable. Building your business on a tool that you used AI to build, but that you own and can run independently of the AI service being available is completely different. That's real value being added - and if done right, it comes with the benefit of being deterministic and escaping the vague definitions of truth that are inherent to LLMs.
Building tools without technical understanding doesn't come for free, though. I ran into this myself recently, building a tool to handle CSV files with multiple lines of text - not understanding the technical foundation can be a major flaw in a workflow. The output format shifted between iterations, with each version is technically correct, but without understanding why or how, you run the risk of creating problems downstream. The governance of making a decentralized suite of custom built tools is something I don't quite grasp yet.
Another part of frontier models I find interesting is how they seem to be diverging - Anthropic focusing on enterprise, Google on a general, broad intelligence rather than being the domain expert in a specific field. All that technical focus has one funny side effect: Fable 5 isn't much good at creative writing. A model named Fable that can't write one.
I wonder if gating frontier models to people within your own region will become the norm. And if so, what effect that will have on open-source models going forward. I have noticed more and more people online talking about switching to different models than the big three - Anthropic, Google, OpenAI - and I'm curious to see if that's a trend that is here to stay. Given the capabilities of last-gen's models - do we really all need to be using frontier models like we are today?
We're getting to a point where there can be no doubt about the technical competencies of frontier models. But given how fragile the outputs and access to them can be, I don't think the true power lies in the model itself. It lies in the capabilities and efficiency that arise when a high competency level within any field is achievable for anyone, despite not having a specialist title.
But from my anecdotal experience, people don't understand this intuitively. They see Claude and ChatGPT as just that - a chat. For that exact reason, I believe we will soon start seeing companies recruiting for people with capabilities that are not just technical, but focused on communication - people capable of building an organization where employees aren't afraid of progress, don't see it as a challenge, and actively participate in building it. Part of that role, focused on a fast-moving, highly technical field, will be educational, with dedicated time for learning. Despite the technical nature of the field, the focus won't be on the person's technical abilities - as long as you have a base-level understanding, that will be enough. The focus will be on their ability to learn and share.
The real value of the current frontier AI models lies in the liberation of technical capabilities, what it allows people to learn, to build, the things they own, even if the technology was to become unavailable. But in order for people to build what they need, they first and foremost need to know that the possibility is there. The existing friction in any given company needs to be removed - give access where it makes sense - ensure that governance is up to date and not too restrictive.
We are at a time now, where it's hard to say that the capabilities or knowledge of models are of low quality. I'm certain that many tasks would be carried out better, if more people were piloting an AI rather than carrying out the work themselves. However, for this to be a reality, the organization needs to be mature and modern enough to allow these interactions to happen with as little friction as possible. It's not a systems issue, it's a people issue.