I will not lie to you. When Anthropic launched Claude Code, I saw the logical next step coming.
The terminal is fine for those who can type bash without shaking, but for a business owner or a freelancer, it is a barrier.
Then comes Cowork. The idea is simple: your coding sessions, your files, and your thoughts are no longer stuck in a local folder on your Mac or PC.
Everything is in the cloud. You start something on your desktop, you continue in your browser while waiting for your train, and you check a detail on your iPhone from your couch. It is fluid. It is efficient. It is even quite impressive technically.
But this is where the trap is set. The current narrative, the one you read everywhere on LinkedIn or in tech newsletters, is that Cowork is your new miracle employee.
They sell you a kind of digital brain that manages your stack, your files, and your execution without you having to lift a finger.
That is false. It is even dangerous if you believe in that kind of promise. Cowork is a productivity tool, not a substitute for your intelligence or your entrepreneurial courage. I use it every day, and I see exactly where the machine stops and where your work begins.
Why I am testing Cowork / CODEX right now
I am interested in Cowork because friction is the number one enemy of execution.
For a non-technical profile or a decision-maker managing an SME, installing a development environment, managing API keys, configuring Git, and ensuring everything runs on the local machine is a monumental waste of time.
Cowork removes this step. The fact that it is linked to your Max plan and deployed gradually shows that Anthropic wants to test the stability of the cloud infrastructure before opening it to everyone.
It is a cautious approach that I appreciate.
The illusion of total continuity
The real gain is synchronization. Being able to switch from a desktop interface to a mobile interface without losing the thread of the conversation or the state of the file I am working on changes the game in terms of speed.
I save about thirty minutes a day just by not restarting my contexts.
But be careful, this fluidity creates an illusion. We get the impression that the AI understands everything because it has access to everything.
This is an error in judgment. Access to files is not the same thing as understanding the business.
The real cost of abstraction
By moving everything to the cloud, you agree to lose a bit of control over the infrastructure.
That is the deal. You trade total mastery of your local environment for a disconcerting agility.
For me, the calculation is quick. I prefer a tool that works everywhere and allows me to delegate technical management, even if it means I depend on a proprietary infrastructure.
The ROI here is measured in available brain time. If I spend less time configuring my tool, I spend more time thinking about my strategy.
Tone and voice cannot be delegated
This is the first point where Cowork will lead you astray if you are not vigilant. The AI is capable of producing clean code and consistent documentation.
It can even imitate a style if you give it enough examples.
But it does not possess your intention. It does not know when to be provocative, when to be humble, or when to be sharp to close a sale. It offers a statistical average of what seems correct.
However, in business, the correct is often invisible. What leaves a mark is singularity.
Final arbitration belongs to the human
I see many people using Cowork to write client emails or sales pages directly in their project files.
The result is often smooth, polished, and totally interchangeable.
If you let Cowork decide the final tone, you transform your business into a commodity.
The AI proposes, you arbitrate. That is the only rule that matters. You must be able to take a generated text and inject your own blood, your own doubts, and your own humor.
If you do not do this polishing work, you send a signal of laziness to your clients.
The risk of standardization
The hidden danger of Cowork is that since your entire environment is synchronized and optimized, you will be tempted to automate everything.
You will end up producing content that looks like all the other content produced by people using Claude. This is the paradox of efficiency.
The more the tool is capable of standardizing the output, the more the value of human deviation increases. Your competitive advantage does not lie in your ability to use Cowork, but in your ability to know when to ignore its suggestions to add a personal and unpredictable touch.
Legal and business responsibility remains yours
This is where I become very serious. Cowork can modify files, push code, and even suggest structural changes to your application or website.
This is great until the moment it breaks everything or, worse, until the moment it introduces a security flaw or a major legal error.
Nobody tells you that the AI can be wrong about a clause in the general terms and conditions of sale or about the management of sensitive data. Yet, this is a daily reality.
The mirage of technical reliability
I do not know exactly how many times per week Cowork or another similar tool makes a subtle logic error, as it depends on the complexity of your project.
But I can assure you that it happens.
If you consider Cowork to be a certified accountant or a lawyer, you are heading for disaster.
The AI does not sign contracts. It does not go to court. It does not pay fines to regulatory bodies. Every line of code or every legal text generated and deployed via Cowork/CODEX is your exclusive responsibility.
The fact that it is in the cloud does not dilute your risk; it simply accelerates it because you can deploy errors faster.
The cost of an unverified error
Let us take a concrete example.
You ask Cowork to modify your payment tunnel to optimize conversion. It does it, it is clean, it seems to work. But it forgot to handle a specific tax case for a specific country.
You lose money or you find yourself in conflict with the tax authorities.
How much time did you save by not checking? Maybe ten minutes. How much will it cost you in stress and legal fees?
Potentially thousands of dollars.
The immediate productivity gain must never mask the necessity of a systematic human audit.
The unique context of your business is invisible
This is the most frustrating point. Cowork has a shared memory. It sees your files, it follows your sessions. This is a huge leap compared to a classic chat where you have to explain everything again.
But be careful: technical memory is not strategic memory.
Cowork knows what you wrote in your README.md file, but it does not know why you decided to pivot your offer last month after a tense call with your biggest client.
It does not know the unspoken truths of your market.
The difference between data and knowledge
There is a major confusion between data (what is written in your files) and knowledge (the experience accumulated in your head). Cowork processes data. It is excellent at that.
But your business relies on nuances that you have never taken the time to document. Nobody documents their intuitions.
Nobody writes a text file to explain why a certain partner is toxic despite correct figures.
If you rely on Cowork to make decisions based on context, you are ignoring 80% of the reality of your company.
The exhaustion of artificial context
Even with giant context windows, the AI eventually saturates or prioritizes certain pieces of information over others.
I often notice that Cowork or CODEX clings to an idea mentioned at the beginning of a session and ignores a correction made later.
This is what is called attention loss in the research field. For you, this means that you must constantly reframe the tool.
You cannot just say “do as usual.” You must be the guardian of the context, the one who reminds the AI of the global vision when it locks itself into a stupid local optimization.
Does total accessibility kill reflection?
We have reached the end of the reasoning. Cowork and CODEX are formidable tools for those who want to stop fighting with their machine to finally focus on their product.
I recommend the tool without hesitation for the raw time gain. But I ask myself an honest question: by making everything so fluid, do we not risk atrophying our ability to conceptualize things in depth?
When everything is accessible in three clicks on an iPhone, we tend to skip the slow reflection stage. We want the result immediately. But the best business decisions are rarely those made in fast mode.
Do you finally want to take action?
I have a special offer to help you see more clearly and obtain concrete results quickly.
The DHS Discovery Pack is made for this: we look together, in one session, at what can be delegated in your business, and you leave with a clear roadmap.
No bullshit, no commitment, no 7-step method. Just concrete action: this is my why: to be USEFUL, and everyone who knows me can testify to that.
Thank you for reading me… until the end…
CDAL
