Anthropic’s latest launch on Claude Co-work, which they say has been fully developed by its AI Claude Code, has opened up the debate about whether we are at the stage of full automation of work thanks to AI. I would say no, because as we approach the surface of this “magic”, we encounter a technical fragility that we cannot ignore.
The reality is that although the Claude Cowork code was generated 100% by artificial intelligence, human intervention is essential to planning, designing and iterating with Claude Code.
Entonce Are we looking for an autonomous AI? Let’s start with the reward Productivity isn’t just about getting more done now, it’s about delivering more real and better value. If 90% of the administrative work is automated but introduces errors that require 20% extra time to review and correct, the net benefit is diluted.
Data from various studies confirm that AI in collaboration with humans brings us the best results. So I reflected on it October 2025 OpenAI study which evaluates AI performance in terms of large real-world economic value. As a result, iterative processes in which AI works with human supervision offer productivity multipliers and cost reductions that are sufficiently superior to independent human labor.
This study reveals that the most advanced models approach experts with 14 years of experience at the level of their superior models. And now for this amazing performance it is not autonomous. To ensure that the AI is at a “higher” level
requires tare context aa “andamiage” process designed by humans. Without this structure, the model breaks down.
The point is not to “buy” AI, but to design a collaborative ecosystem where humans bring context and AI the power of execution. Without forgetting that you need to know how to think about what we do and how much it really costs us.
Are we looking for complete autonomy in software development? In our experience without a solid head Government ITautonomy is a possibility rather than a dream.
Data also supports this need for human management and monitoring. El HAI-Eval studio (arXiv:2512.04111) Our synergy in the cooperation program is one of our destroyers. Before complex logic problems, the most advanced AI models can only reach one stage of completion 0.67% when they work alone. But when you join a person in a collaborative thinking environment, efficiency soars 31%which with its score surpasses even the experts. If we blindly trust AI without checking how it is applied, we cannot see it through to the end. In fact, government is everything.
AI with human supervision is significantly better, but for this combination to work, you need to drive it through three pillars. I have to take one to the top objective medicine. Don’t let the “feeling” of moving faster go away. It is vital to use specified product metrics (such as international standards centered on functional scale) to verify whether AI is effective or simply “more scripted”.
On the other hand, it is essential specialized human supervision. Artificial intelligence is fantastic, but human intervention is crucial for planning, design and control.
finally The use of a rigorous estimation model is essential. Estimate management is critical to protecting IT assumptions. We need to know that we are paying, regardless of whether the code is written by a human or an algorithm, 80 euros per month.
The future of business in the AI era is not a career that has complete autonomy, but has controlled precision. The most valuable AI, at least for now, is not the one that “does everything” until it is integrated into processes that excel at what the organization needs, and in those whose quality and cost are under low control.
This is exactly the goal that we meet every day with our customers and where we support ourselves to be able to drive an AI application. Because doing things faster and easier. Doing it better, in a sustainable way and with given fairy tales is still a deeply human activity. And we care about being the people we lead to achieve results.
***Alexandra Blanco He is the Content Director of LedaMC.
Leave a Reply