Why Did We Build This Website?
A Manifesto on the Use of AI
September 2026
We decided to run an experiment with our internal documentation to free our HR team from routine questions such as how many leave days do I have for X? Now that we are more than 200 people, plenty of minor questions come up every day, and many of them can be answered simply by knowing which document to look at. This is exactly the kind of problem where RAG (Retrieval-Augmented Generation) works well. After trying it internally, we thought it would be fun to rebuild our public website around the same idea, launching our new intranet and website at the same time.
We have never liked conventional websites. This one is no exception.
We also built it this way because we are software engineerers and use AI extensively in our daily work. We believe we use AI better than most, and this website is one way of demonstrating that. At the same time, we think there is an enormous amount of hype around AI, along with plenty of unjustified expectations. A website like this gives us permission to say so without immediately being called neo-Luddites.
Today, we cannot imagine software development without AI, just as we cannot imagine it without Git, CI/CD, or an IDE. But we do not believe that anyone with no knowledge of software engineering or of the project itself, can use AI to replace good engineering. AI is like a mirror: the quality of the output depends largely on the quality of the input. It is not enough to tell it what you want. You also need to define how the work should be structured, planned, and verified.
We miss the world in which predictions about technological progress were tied to the professional reputation of those making them. People such as Nicholas Negroponte, who in Being Digital (1995) described the transition from atoms to bits, or William Rees-Mogg and James Dale Davidson, who in The Sovereign Individual: How to Survive and Thrive During the Collapse of the Welfare State (1997) anticipated cryptocurrencies and mobile banking many years before they became reality.
Today, we are surrounded by wild predictions about the metaverse, humanoid robots, quantum computing, blockchain-based traceability, trips to Mars that were supposed to happen back in 2018, and above all, predictions about AI. Instead of mocking people who make completely unfounded forecasts, we amplify them. Then, when the deadline arrives and the prediction has failed, we collectively kick the can down the road for another couple of years, or twist the definition of what was originally claimed (what does intelligence really mean? what exactly do we mean by Artificial General Intelligence?), until the prediction no longer sounds so ridiculous.
All these predictions ultimately serve the same purpose: attracting investment, with little concern for the damage they may cause in labor markets. We think it is important to say this plainly and not allow ourselves to be fooled about the incentives behind them.
According to countless forecasts, software engineers should already have been opening coffee shops or finding some other line of work that AI could not replace. We have supposedly reached the singularity several times already, and now, apparently this time for real, AGI will arrive again within a few weeks, for the third time, and will be capable of doing the work of a researcher, while it still struggles to reliably do the work of a secretary.
A great deal of AI is marketing designed to make us believe things that are not true. It starts with the name Artificial Intelligence itself, applied to systems that, so far, are not intelligent. Then we call tasks or processes agents when they lack the very thing the word suggests, agency or intrinsic motivation. If, after training the next, more powerful LLM, you switch it on and leave it there without giving it any instructions, it will wait indefinitely and do absolutely nothing. No impulse will spontaneously emerge from those trillions of circuits. That does not seem particularly agentic or intelligent.
We do not doubt that, given enough time, a systematic engineering process may eventually outperform biological evolution and produce a digital brain superior to ours. But the honest answer is that we do not know when that will happen, and such a brain will most likely look very different from today’s LLMs.
To predict when you will reach a destination, you need a plan. Without one, what you have is not really a prediction; it is a hope or a wish. And the wish today seems to be that, by giving LLMs more and more computing power, intelligence will somehow emerge by magic, without anyone knowing exactly how or when, although for several years now people have been feeling that it must be just around the corner.
If AGI is achieved, the organization that gets there first will hold a degree of power over the rest of humanity unlike anything we have seen before. Whether it will use that power for everyone’s benefit remains to be seen. We suspect it will only democratize access as far as it can without giving up its dominant position. It is therefore obvious that anyone would want to be part of that organization, at any cost.
So while OpenAI executives resign and leave the company, Meta suffers talent departures, Grok’s creators move on to do other things, or major DeepMind figures leave Google to start their own companies, the siren songs telling us that AGI is just around the corner do not matter very much. Reality clearly suggests something else. They do not know. If any of these organizations actually knew, and had even a moderately realistic plan for getting there, nobody would be leaving. People would be lining up to get in.
Until AGI arrives, we would like to make a case for common sense and honesty. For making decisions based on facts rather than wishes. For not promising solutions that do not actually work as advertised except in specific circumstances and under very particular conditions. For not forecasting breakthroughs when nobody really knows when they will happen. As software engineers and designers who use these tools extensively, we have a responsibility to stay grounded in reality and not amplify falsehoods or misleading promises from CEOs whose primary goal is securing the next funding round to keep feeding the machine.
LLMs are extremely useful and dramatically accelerate our software development cycle. But LLMs are probabilistic models that do not truly reason, so the code and designs they produce do not always make sense and must be reviewed by experienced professionals. We cannot simply generate work with AI that we do not even bother to check, expecting someone else to review it, or worse, release it to the world without any verification at all. There may be scenarios where this kind of rapid AI prototyping with very little human supervision makes sense, but that is certainly not the kind of project our clients hire us to do. For one thing, they are already doing those projects themselves.
With the help of AI, we have finally managed to complete and publish as Open Source a collection of Erlang libraries that had been sitting unfinished in our private repos for years. Without these tools, we would still be using those libraries internally, but they would not be complete enough to share with the world. We are not web experts either, and we could not have built this website without the help of an LLM. AI is extremely useful for exploring new paths, taking tedious work off our hands, replicating solutions, or finishing a task that is already well defined. But it should be the human who sets the course, defines the criteria, handles the critical parts, and takes responsibility for the result.
We are excited to be part of this new, more productive professional era. Sometimes we feel like we have acquired superpowers, giving us a greater ability to break down barriers and explore new territory. But we do not want to become secretaries to AI, sorting through and cleaning up what it produces. Nor do we want to flood our lives with meaningless code and content. At Nomasystems, we want to remain in charge of critical engineering decisions and creative processes, while handing AI the less important work that does not motivate us. We do not want AI to be a god. We want AI to be a servant.
This text was translated from Spanish using AI.