Which HR tasks you should not automate with AI

It is getting easier to automate almost anything, and that is why the useful question has changed. It is no longer "can we automate this?" but "should we?". In this guide we share how we think about that decision at Método y Estrategia, what the latest data and examples say about artificial intelligence in HR processes, and how you can apply it in your company.

What is the "human reserve"?

It is the idea of reserving for people those tasks that, because of their impact, complexity or sensitivity, should remain their responsibility even if they could technically be automated. Jose Cardona, co-founder of Método y Estrategia, put it forward on LinkedIn starting from a reflection by Bill Gates, with an image that sums it up well: a robot giving someone the news of an incurable illness. There is no technical reason it could not do it; still, it should not.

Jose adds something very practical: an outside perspective helps avoid uncontrolled tool purchases, the kind driven only by the urgency to automate without being clear about which problem needs solving. Read Jose's post on LinkedIn (in Spanish).

What is happening today with AI in HR processes

These three recent examples show why it pays to decide calmly.

1. In hiring, AI filled the processes with near-identical applications

An article published by the World Economic Forum on August 13, 2026, written by professor Marvin Starominski-Uehara (the opinions are his, not the Forum's), describes a problem many HR teams already know. According to the data it cites:

• LinkedIn applications rose 45.5% in late 2024, while job postings fell 10.6%.
• In a Greenhouse survey of 4,100 recruiters and hiring managers (U.S., U.K., Ireland and Germany), 34% of recruiters say they spend up to half of their week screening junk applications.

His proposal is called trace hiring: instead of filtering résumés, evaluate how candidates think, adapt and build when facing a real problem, for example with a video response to a company challenge, an observed group task and an individual reflection at the end. Some companies are already adding friction on purpose, with longer applications, more human review and skills tests.

To read it calmly: it is the author's proposal, not a practice with measured results, and it requires people's time (live sessions and interviews). Still, it illustrates the underlying idea well: if we automate without judgment the very signal we wanted to measure, that signal disappears. Read it at the World Economic Forum.

2. Interviewing a candidate is now a task an AI can do

Amazon Web Services introduced a tool called Amazon Connect Talent in April 2026 and made it generally available on September 16, 2026, for now in two U.S. regions (Virginia and Oregon). According to AWS, its AI agents run structured voice interviews with adaptive questions, apply competency-based assessments and score candidates consistently; recruiters receive scores, transcripts and detailed evaluations.

We are not saying it does not work. We are saying it is exactly the kind of task where it is worth asking before delegating: which part of getting to know a person can be structured without losing what we are looking for? Who reviews, and with what criteria? The AWS page does not detail these points, so each company should define them. Official AWS announcement.

3. AI can make everything look fine today while tomorrow's talent fades

In an article published by the World Economic Forum on August 17, 2026, Nacho De Marco (co-founder of BairesDev, in an opinion column) uses data from his company's developer barometer, with more than 1,500 developers from 77 countries, and a roundtable of 26 U.S. tech leaders. 54% of senior developers agree that AI makes the junior role less relevant.

His warning is simple: when a senior solves with AI the task a junior used to learn from, the work turns out fine, but the junior loses the chance to learn. In the words of the roundtable, "AI productivity is masking the loss of capability" (our translation). Short-term metrics look healthy while the path that trains future experts thins out.

To read it calmly: it is about software developers and is an opinion column based on surveys from a single company. The lesson, however, applies to any trade that is learned by doing repetitive tasks. Read it at the World Economic Forum.

What changes the outcome: the people who lead

A Gallup study of 23,717 U.S. workers (February 2026) asked how their organization's culture changed over the past year:

• Where AI has already been implemented, 51% say culture stayed the same, 25% that it got worse and 24% that it got better.
• Manager support tips the balance: 31% of those who feel their manager actively supports the use of AI on their team see a cultural improvement, versus 21% of those who do not.
• In a separate survey of 102 chief human resources officers (CHROs) at Fortune 500 companies, 50% say they are not very or not at all confident in their managers' ability to guide their teams in using AI.

To read it calmly: this is U.S. data, and for the CHROs it covers large companies; it is not a measurement of Colombia. But it points to something we see in the field: the tool matters less than how it is introduced and who supports the people. Read it at Gallup.

Tatiana Agudelo, from the Método y Estrategia team, shares an experience along the same lines. We were hired to strengthen a team's leadership, but we realized the real challenge was not the leaders but the management model: teaching delegation does little when nobody is allowed to decide. So we stopped the process and proposed working with senior management first. Her conclusion: not every problem is solved with more workshops, nor with more tools. Read Tatiana's post on LinkedIn (in Spanish).

Four questions to decide what to automate in your company

This is our way of applying the human reserve idea. They are our own criteria, not a rule; use them as a starting point for a conversation with your team.

1. Which specific problem do we want to solve? If we cannot say it in one sentence, it is not yet time to buy a tool.
2. How sensitive is the task for the person on the other side? The more it affects their job, their health or their opportunities, the more important it is that a person is present.
3. Who answers if it goes wrong, by name? Automating a decision does not remove responsibility for it.
4. What did people learn by doing this task, and where will they learn it now? It is the junior and senior question: if automation takes away their practice, it must be replaced with another way of learning.

If the answer to the second or third is delicate, that task probably belongs to the human reserve, or at least it is best automated only in part and with a person reviewing.