To name something is not merely to attach a passive label; it is a profound, performative act of world-making that actively brings an identity into being. As media theorist Neil Postman observed, an entity "isn't anything until someone names it," and the chosen name reveals exactly how the namer wishes the world to perceive it—a definition that soon becomes reality for all who accept the name.
It is for exactly this reason that I found the choice “Anthropic” to be extremely interesting as a name for a high-profile AI studio. The word "anthropic" derives from a Greek root and fundamentally means relating to human beings or humanity. When adopted by an AI studio, this etymological choice is not merely a label, but rather a coherent, performative claim that the company is intrinsically "human-centred”. By pairing this overarching identity with a personal, human name like "Claude" for its flagship model, the studio deliberately cultivates a narrative that its immense technological power exists strictly "for" humans.
Anthropic has made numerous official statements and published framework documents that explicitly invoke these themes.
They explicitly categorise their research as an effort to build "more human-centred conversational AI experiences".
The company states that its long-term goals involve giving AI emotional intelligence and making systems capable of "human-centred conversations" that are useful in everyday interactions and real-world communication.
In a formal 2023 legal submission to the US Copyright Office, Anthropic’s Deputy General Counsel officially defined the organisation as an "AI safety and research company". The submission states their corporate governance is strictly aligned with the mission of "developing and maintaining advanced AI for the long-term benefit of humanity" and deploying systems that are "beneficial and useful to society".
In Anthropic's published "Constitution"—the document that dictates how Claude evaluates its own behaviour—the absolute highest-ranked core value (Priority 1) is to be "Broadly Safe". This explicitly prioritises safety over being "Genuinely Helpful" (Priority 4). They also stated that if AI safety proves unsolvable, its role would be to "push for halting AI progress to prevent catastrophic outcomes".
All this suggests that Anthropic does not just rely on its name -- and subsequent branding -- but actively adopts the rhetoric of technology at the service of humanity, aiming at societal benefit and prioritising human safety and overall harmlessness. I couldn’t agree more with that stance.
As a small-time Anthropic customer and, being overly committed and relatively ambitious as much as I am wide-scoped in my work, I have been a paying customer of theirs for some time now, climbing the echelons of the subscription tiers gradually to the point where, for the past few months I have been allocating up to 10% of my monthly revenue to Claude-related services, without a second thought -- at least up until recently.
To anyone who has spent a few hundred hours interacting with “frontier models” it soon becomes apparent that, the promise is only fulfilled in breadth, not in depth -- and that’s okay. It is not a matter of intent, I don’t think, as much as it is a restriction of the technology itself: Yes, the knowledge pool is vast, the power is immense, the progress is obvious, but every consecutive “groundbreaking” model, though a step in the right direction, only serves to reaffirm the power and capability of Large Language Models (LLMs) which can lead to impressive leaps in productivity, while accentuating the chasm between human thought and the semblance (or simulation) of it: I expect the LLM to strongly resist any truly innovative thinking (since it is trained on what already exists), to struggle with recursive architecture (since its vectors only ever move forward), to make obvious, stupid mistakes (since any task-specific memory available to it is purely transient) and -- by now -- to outright lie, only to come back with “honestly: this, honestly: that” when called out.
I’m pretty certain that, even if I could afford to hire human engineers, which would be ideal, I would come across some of those characteristics every now and then. I’m even willing to admit to demonstrating some of them myself on a bad day. So what do I do? What everybody else seems to be doing under the circumstances: Adjust. Try to maximise the benefits, minimise the downside. The industry as a whole, including the providers themselves, does this by adding substrates such as agents and harnesses.
A harness, especially, is a fine example of trying to impose human thinking onto AI. Since we do not want to allow “things” to go wrong, we try to make sure that the “dos” and “don’ts” that shape our needs and requirements are “imposed” (in a sense) onto LLMs so that the result does not altogether amount to wasted resources, a.k.a. time and money, but also that the outcome mostly obeys the rules and sufficiently meets the specifications of our original intent. The rest is up to us humans, I guess.
In and of itself, this need for harnessing, this ongoing search for some semblance of real control and direction only confirms what humans using this technology on a daily basis know to be true, at least at this point in time: LLMs cannot be trusted to do “the right thing” at all, and they can hardly be trusted to do “what they’re told.”
A result of this very realisation, by no means trivial, is that “autonomy” when it comes to AI is a pipe dream, a moot point, even more so because the concept of autonomy and decision-making is useless unless it is grounded in responsibility: An LLM cannot realistically be held responsible for making bad choices or taking wrong actions any more than a dog can be held accountable for peeing on the carpet: You should have taken it for a walk.
Code, and to a lesser degree content produced applied and packaged and sold as a product or service is the equivalent example: If anything goes wrong and the product results in any kind of harm or damage to its end users, Claude -- and by extension, Anthropic -- I assume will not be so eager to be held liable, despite the “Co-authored by Claude” banners that Anthropic makes sure are included in every LLM-assisted code commit. This is already a first sign (being eager to share any glory but not so much any responsibility) that Claude’s constitution, though masterfully crafted, does not necessarily align with that of Anthropic, the organisation behind it which, as it must, is first and foremost concerned with the benefit and wellbeing of itself and its shareholders.
So far so good, mind you. Nothing unusual, nothing extraordinary. There is no hate, no anger, no complaints, no unrealistic expectations, nor can you expect any: Things are what they are.
In fact, I meet this entire “AI revolution” business with an enthusiasm reminiscent of the days of the desktop publishing revolution when, instead of Anthropic, we had another “A” company, Adobe, open up a world of possibilities to the creative community. Coupled with the radiance of the Apple paradigm and made highly visible by the monumental, by now legendary “Photoshop” -- which ended up becoming a verb, even -- Adobe brought immense (for those times) power into the hands of the many at an accessible cost.
In the 1990s, I was one of the millions who contributed to making their tools central - I even taught Photoshop to PwC employees at one time, used Premiere extensively, talked about their software with the kind of enthusiasm that edges into evangelism.
And then, something happened. The relationship changed – on their side: Subscriptions, lock‑in, dark patterns around cancellation, shifting terms, new AI‑era policies about data that felt more extractive than respectful. I never hated Adobe. I never waged a campaign against them. But I did, quietly, lose respect. I did the hard work of moving to alternatives. At some point, Adobe became irrelevant. A bad aftertaste that fades with time. Though Adobe was never a villain in my own story, it is clear to me that they were guilty of hubris – another Greek word – and that story has an established, archetypal ending.
The word "hubris" originates from ancient Greek culture, where it was used to describe a specific way that power goes wrong.
In plain English, hubris is not "lying" or "hypocrisy," but rather a structural pattern of behaviour: it begins with rising success, which leads to an overreach that feels entirely natural to the successful party, ultimately resulting in a "blindness about limits". It manifests as a powerful actor assuming that the boundaries and environments of others can be progressively reorganised to serve its own interests.
According to the ancient Greeks, who knew a thing or two, when hubris occurs, it inevitably triggers a "correction that flows from reality itself”, a result is called nemesis -- not an arbitrary punishment or a divine thunderbolt striking the offender; rather, a self-inflicted downfall caused by ignoring fundamental facts and realities. Which brings us back to Anthropic, the human-centred AI studio - mega-corporation, at this point in time valued in the hundreds of billions, only this time the timeline is not spread over a number of years, as was the case with Adobe, but is much more compressed in time, as all things tend to be lately.
The pattern is, by now, a familiar one: From $20 a month to $100, then onto $200 - fine, my need, my choice. Increasing reliance on Anthropic’s services based on a relationship which they defined and to which I conformed and on which I relied to plan, produce and build much of my own work: Again, so far, so good.
But lately, the ground has begun to move beneath my anthropic feet. A model that performed decently is upgraded, and the new, “better” model seems to be worse, not better. I find that I get less work done for the same money and that this fluctuates -- sometimes dramatically so. Usage of my subscription is then confined to Anthropic’s own harness, which makes me think that they might be reaching for a “system lock-in” strategy, or at least a hybrid of it: This is never a good thing for the end user. Again, it is what it is, and I work around it. I jump through all the hoops, obey all the rules, patiently -- even stoically -- keep quiet to all of it, despite the loud stream of complaints, even thrashing, of thousands of others who will not succumb to their fate as nonchalantly as I: People whose paid accounts are suspended or banned with generic references to terms and no clear explanation; companies whose Claude access is cut off for hours by automated systems, freezing work for dozens of employees; support channels that route through generic forms while teams wait.
And then, I am notified via email the other day that “non-interactive” use of Anthropic’s harness will, into the future, be excluded from the monthly subscription. I stop and think:
“Non-interactive” use means I use the same technology, I use the same (Anthropic) software to do so, pay the same amount of money but I cannot do so in the way that I choose. I cannot choose the way in which I want to perform my tasks, it has to be done in the way that Anthropic chooses, I need to be talking to that same Claude personality -- and if I want to impose my own rules, I need to do it their way. For a technology that’s supposed to work “for” me, that’s where I draw the line.
All this has nothing to do with my own personal problems, which are silly, really, given what’s taking place in the world right now. And yet, in a way, it does. It is a pattern of behaviour, an archetypal “history repeats itself” that is becoming visible and very much felt on many levels, across domains, all across the world in which we live -- whether it be the political system, the financial system or big tech.
My specific problem was solved by making a concession: Since I am hand-holding, disciplining even, the LLM, making it “bend to my will” (to the degree that I can of course), I simply “tightened the screws”, slowed the work down somewhat, started switching to Chinese models that offer 80% of the power at 20% of the price -- which is probably something I should have done long ago. I’m quite pleased with the results, by the way. If things continue down this path, a few months down the line Anthropic will become less and less relevant to this humble, solitary knowledge worker.
I know: I am less than a rounding error for Anthropic. I was irrelevant to them to begin with, so my migrating to other providers is a non-event for them. Yet, that is the essence of hubris. And we all know how that ends. I just wish it didn’t have to be this way. Because it was good – while it lasted.