For a long time, at least for me, working with technology was almost synonymous with learning how to use new tools. New toys, new commands, new interfaces. We had to understand their capabilities, their limitations, and their procedures.

Lately, I’ve been asking myself whether this has changed with the arrival of Artificial Intelligence. On the surface, working with AI does look like working with a new tool. And in a way, it is. But it is also something else: learning how to think with an entity capable of responding, proposing, correcting, synthesizing, questioning, and executing.

That forces us to develop a new way of working: one that combines the art of asking better questions with the ability to design processes where those answers can be turned into action.

AI as an Interlocutor: The Value of Socratic Dialogue

Socrates did not teach by giving closed answers. He taught by asking questions. His method was about helping others think better, uncover contradictions, clarify concepts, and reach a deeper understanding of the problem.

Working well with AI has a lot to do with that. It is not simply about writing:

“Create a marketing strategy.”

It is about starting a richer conversation:

“Help me think through a marketing strategy for this product. First, ask questions about the audience, the positioning, the differentiators, the channels, and the risks. Then challenge me if you see inconsistencies.”

That change may seem small, but it is enormous.

AI stops being a machine for answers and becomes a machine for assisted thinking. It helps us ask better questions, detect blind spots, organize scattered information, and turn vague intuitions into clearer hypotheses.

Working with AI requires a new skill: knowing how to converse in order to think better. The person who writes the most prompts does not win. The person who structures the dialogue best does.

The quality of the answer depends on the quality of the thinking that comes before It

One of the most common mistakes when working with AI is expecting the model to compensate for our lack of clarity. We ask for sophisticated results based on poor, ambiguous, or incomplete instructions. And then we get frustrated when the answer feels generic.

AI amplifies the way we think. If our request is superficial, we will probably get a superficial response. If our mental framework is confused, AI can help us organize it. But first, we need to recognize that the work begins before the answer.

That is why Socratic dialogue is so valuable. Before asking AI to produce, we can ask it to question:

• “Before answering, identify what information is missing.”.
• “Show me the implicit assumptions in my request.”
• “Tell me what might be poorly framed.”
• “Ask me ten questions that would improve the quality of the answer.”
• “Give me three alternative ways to interpret this problem.”

These instructions turn AI into a critical mirror. Not a passive reflection of what we already think, but a set of responses that helps us see better.

Working with AI does not mean delegating thought, as many people fear. It means raising the quality of thought before delegating execution. Conversation is phase one. Production is phase two. Confusing them is a common and costly mistake.

From Dialogue to System: We Need to Design Loops

Socratic dialogue helps solve the problem of what to think. But it does not solve the problem of how to operate in a sustained, repeatable, and scalable way. For that, we need the other half of the work: designing loops.

A loop is a defined cycle of action, evaluation, and improvement. It is a repeatable structure that allows an agent (human or artificial) to move toward a goal without depending on a single perfect instruction.

Instead of asking AI to do something once, we design a process:

1. Analyze the objective and the context.
2. Propose a solution or first draft.
3. Evaluate the proposal against defined criteria.
4. Detect errors or weaknesses.
5. Correct.
6. Evaluate again.
7. Deliver a final version.

That loop can be applied to writing an article, designing a campaign, reviewing code, preparing a presentation, analyzing documents, responding to customers, or coordinating specialized agents.

The key difference is that we stop thinking of AI as an input-output tool and start thinking of it as part of a work system.

The prompt Is Not the Product. The System Is.

During the first stage of AI adoption, many conversations revolved around the prompt: how to write it, how to improve it, how to find the “perfect prompt.” But as we move toward more autonomous agents, the prompt stops being the central unit, it becomes a building block. The loop becomes what matters most.

A good prompt can generate a good answer. A good loop can generate a good operation.

An isolated prompt depends too much on chance, incomplete context, and the model’s interpretation. A well-designed loop, on the other hand, reduces ambiguity because it defines roles, steps, criteria, validations, and mechanisms for improvement.

For example, to create a piece of content, we could define a loop like this:

AI Content Creation Loop

• Analyst agent: reads the objective, the audience, and the context; proposes a structure and evaluates it against the objective before moving forward.
• Writer agent: takes the approved structure and produces a first draft following the defined tone and format.
• Critic agent: reviews the draft across four dimensions (clarity, precision, tone, and consistency) and returns concrete observations.
• Writer agent, second round: incorporates the observations and produces a refined version with publishing recommendations.

But the loop can go further. We could ask for several versions of the same article, serve them to different audiences, measure what worked best and in which channel, and use that learning to feed back into the analyst, the writer, and the critic.

That process is much more powerful than simply asking: “Write a post about this topic.”

The intelligence is not only in the model. It is in the design of the system that uses the model.

Agents need good environments, not just good instructions

When we talk about AI agents, we often imagine autonomous entities capable of solving problems from end to end. But an agent does not work well simply because it has access to a powerful model. It works well when it operates inside a well-designed environment.

That is why designing the agent’s environment is just as important as defining its task. That environment needs:

• Clarity of objective: what it needs to produce, not only what it needs to do.
• Clarity of role: who the agent is in this context: analyst, writer, reviewer, coordinator.
• Clarity of context: what information it has available and what information it needs to look for.
• Clarity of limits: what it can decide on its own and what requires human supervision.
• Clarity of success criteria: how the agent knows it has finished well.
• Clarity of contingency: what to do when something fails or when the information is ambiguous.

A poorly oriented agent can move very quickly in the wrong direction. And that is one of the biggest risks of applying AI to work: accelerating processes without having defined clearly enough where they are supposed to go.

Designing loops is not a minor technical task. It is a strategic one.

The new skill: Thinking in conversations and systems

Productivity with AI combines two planes.

The first plane is conversational. It has to do with knowing how to ask, ask again, challenge, clarify, and explore. It is the plane of Socratic dialogue.

The second plane is systemic. It has to do with designing processes, loops, evaluations, roles, and control mechanisms. It is the plane of work architecture.

Those who master only the first plane may have good conversations with AI, but may fail to turn them into repeatable results. Those who master only the second plane may automate processes, but risk building rigid systems without enough judgment or capacity for reflection.

A major advantage appears when we combine both.

We converse to think better. We design loops to execute better. And we use agents to scale that combination.

From tool users to designers of intelligence

This shift has a profound consequence: the human role is moving.

We are no longer just software operators. Nor are we merely passive supervisors of intelligent machines. We are becoming designers of applied intelligence.

Our task is to define problems, formulate hypotheses, structure conversations, create loops, establish criteria, and continuously improve the system.

AI can produce a lot. But it needs direction, purpose and vision. It can write, summarize, classify, program, compare, translate, analyze, and propose. But it still depends on the quality of the framework we give it.

Without a good framework, it produces sophisticated noise. With a good framework, it can become a powerful extension of our ability to work.

There is an interesting paradox in all of this: the more capable AI becomes at answering, the more important our ability to ask becomes.

The answer becomes cheaper. Generation accelerates. Production multiplies. That is true. That is why everyone is talking about the bottleneck moving from “coding” to other areas such as quality control, security, marketing, legal, and so on.

However, there is also work to be done before execution. There is a major opportunity for us to bring more maturity and a more distinctive style to what we ask from AI, as long as we improve the definition of the problem, the quality of our criteria, and the architecture of the process.

In a world where anyone can generate content, code, analysis, or ideas in seconds, the difference will not be in producing more. It will be in producing with more meaning, more personality, and more differentiation.

And that requires knowing how to dialogue and how to design.

AI forces us to recover something very old -the art of asking good questions- and combine it with something very contemporary: building systems where humans and agents work together in cycles of continuous improvement.

Socrates did not believe in writing, because he argued that it weakened the mind. At an individual level, he was probably partly right: something changes in us when we externalize part of our thinking into technology. At the same time, he was deeply wrong at the level of the species. We would not be where we are without writing.

And in the same way that Socrates criticized writing for “weakening the mind,” some studies and leading voices are now saying that AI weakens the quality of our thinking.

But it does not have to be that way.

Socrates gave us the recipe to avoid it: working with AI is not simply learning how to ask a machine for things. It is learning how to think and work with it.

Ask. Think. Design. Iterate.

Seen this way, nothing has changed in thousands of years.

You may also be interested in reading:

GeneXus and Neuro-Symbolic Architecture

The Problem with Prompt-Based Development

GeneXus in the Era of Agentic Development