---
title: "Keep your judgement sharp: decide first, read the recommendation second"
url: https://xalt-northstar.com/blog/keep-your-judgement-sharp-decide-first/
description: "If you only approve recommendations, your judgement fades. With this exercise you decide yourself first once a week and read the recommendation second."
source: XALT Northstar
---

[Blog](https://xalt-northstar.com/blog/) AI leadership

# Keep your judgement sharp: decide first, read the recommendation second

If you only approve recommendations, your judgement fades. With this exercise you decide yourself first once a week and read the recommendation second.

Philipp Göllner 19 November 2026 7 min read

![The XALT team in front of the Enterprise AI Summit marquee in Charlotte, the first stop of the AI World Tour](https://xalt-northstar.com/assets/events/enterprise-ai-summit-charlotte/team-xalt-charlotte.jpg)

For seven weeks I ran XALT from the road, on our [AI World Tour](https://xalt-northstar.com/ai-world-tour/) from Charlotte via Vienna, Munich, Shanghai and Tokyo to Bangkok. Our whole team works in one shared agent session. The recommendations come in fast, and most of the time they are good. That is exactly what worries me about my **judgement**: a CEO who only approves recommendations loses the skill of judging.

My answer is a simple exercise, once a week: **decide first, read the recommendation second**, then compare. In this post I describe why this matters now, how the exercise works and which decisions suit it.

## Why judgement fades right now

A good decision paper used to take days. Someone had to collect numbers, write down options and weigh pros and cons. Today agents and teams deliver that in minutes. With us, several people write in the same session with the same agent and the same context; our article on [OpenClaw Multiplayer in the enterprise](https://xalt-northstar.com/blog/openclaw-multiplayer-enterprise/) describes how that works. For individuals there is the same thing as a [personal agent, the AI Chief of Staff](https://xalt-northstar.com/personal-agent/).

That is a big gain. It also has a side effect that is easy to miss. When the recommendation is on the table, well argued and cleanly written, approving becomes the default. Each single approval is reasonable. The sum of them is a CEO who **has not thought a decision through from scratch in a long time**.

I know the temptation well. On the road, between two meetings, phone in hand, "fine, go ahead" is the most comfortable answer in the world. The problem only shows later: on the day a recommendation is weak and I no longer notice.

## What we already know from automation

The pattern is old. The psychologist Lisanne Bainbridge described it in 1983 in her paper ["Ironies of Automation"](https://en.wikipedia.org/wiki/Ironies_of_Automation): when a machine takes over most of the work, the people next to it no longer practise their skills day to day. They are still needed, precisely in the rare, critical moments. Bainbridge's conclusion: such people need **more training, not less**.

Patrick Debois, who coined the term DevOps, applied this idea to AI-generated code [at the DevOps Enterprise Summit](https://videos.itrevolution.com/watch/1002959794/) in 2024. His point: if you no longer write code yourself, you gradually lose the basis for reviewing other people's code well. An [IT Revolution article on human oversight of AI](https://itrevolution.com/articles/human-in-the-loop-is-non-negotiable-leading-ai-adoption-in-safety-critical-systems/) paints a similar picture: someone who is only formally responsible and otherwise watches passively loses attention over time and steps in too late at the decisive moment.

For leadership teams I know of no solid study on this. That it works the same way there is my observation, not a measured fact. But the logic is the same: judging is a skill, and a skill you do not use gets weaker.

## The exercise: decide first, read the recommendation second

A round takes about ten minutes of thinking plus the time to read the recommendation. It replaces no recommendation and no conversation with the team. It only makes sure that I regularly think for myself before I read.

### Pick one real decision a week

Take a decision that is coming up anyway and for which you will get a recommendation, from the team or from the agent. No practice task, no made-up question. Only a real decision shows you how good your judgement is in everyday work.

### Write down your own decision before you open the recommendation

Before you read the recommendation, write down how you would decide, plus **two or three reasons**. A few sentences are enough. Also note which information you are missing and how sure you are. All that matters is that it is in writing. In your head it is too easy to adjust your own view afterwards.

### Then read the recommendation

Now open the recommendation and read it in full. Do not change your note while you read. If the recommendation convinces you, that is a good outcome; you have learned something. You make the actual decision afterwards as always, with everything you now know.

### Note where you differ

Write down where you would have decided differently and why. Three questions help:

- What information did the recommendation have that I was missing?
- What information did I have that the recommendation lacks, for example about people, customers or the market?
- Where did we weigh the same facts differently?

### Keep a short log

Record every round in a simple log. A table with six columns is enough:

Date Decision My choice and reasons Recommendation Difference Later: who was closer?………………

*Template for the log of the weekly exercise*

You fill in the last column only once it shows how the decision turned out. For some decisions that takes weeks, for others a quarter.

### After eight to twelve weeks, look at the pattern

After about eight to twelve weeks it pays to look at the whole log. Where was the recommendation better than you? Where were you better? Are there topics where you are regularly off, or topics where the recommendation regularly misses something? **You learn more from this pattern than from any single round.** It also shows you where you can safely follow a recommendation and where you should look more closely.

## Which decisions suit the exercise and which do not

Not every decision fits this exercise. Good candidates are decisions

- where you have experience and context yourself, such as filling a role, the priority between two customer projects or where a budget should go,
- whose outcome you will see within a few weeks,
- that matter enough for you to take them seriously, but are not so urgent that there is no time for ten minutes of thinking.

Less suitable are decisions

- that have to be made immediately, such as during an outage or a security incident,
- for which you lack the expertise altogether, such as a detailed legal question; there a specialist's recommendation is exactly right,
- whose outcome you will never learn, because then the last column of the log stays empty,
- that an agent handles better and faster anyway, such as purely operational coordination.

## How to set up the exercise with an agent

With an agent on the team, the biggest risk is seeing the recommendation too early by accident. It often sits in the summary or in the first line of a message already. Three settings help:

- **Facts first, recommendation later.** For the practice decision, ask the agent or the team for a fact sheet without a recommendation: situation, numbers, options. That way you decide on the same basis and still train your own judgement.
- **Keep the recommendation separate.** It goes into its own document that you only open once your note is done.
- **Compare together afterwards.** Once your note is written, the agent can put both versions side by side and list the differences. You write the reasons for your own choice yourself, beforehand.

A fixed date in the calendar, for example every Monday morning, makes sense to me. Without one, the exercise slips away in exactly the weeks when the most is going on.

## Typical pitfalls

A few mistakes are tempting, and they make the exercise worthless without anyone noticing:

- **Picking only easy decisions.** Then you learn nothing about the questions that matter.
- **Reading the recommendation in the subject line.** Once you know what is recommended, you no longer decide independently.
- **Writing vague reasons.** "Gut feeling" is allowed, but then write down what the feeling is based on.
- **Skipping busy weeks.** That is when you approve the most.
- **Turning it into a contest.** The point is not to be right, and the team should not keep score.

## What the exercise is for

The exercise is **not distrust of the team or the tool**. I rely on both every day, and that should stay that way. It is meant to keep one specific skill: noticing a weak recommendation when it comes. That only works if I still know what a good decision in that area looks like.

It helps the team too. A CEO who can still judge asks better questions and recognises good work more precisely. That is why I would make the exercise open: tell the team that once a week I decide first and read second, and talk through the differences with the people who wrote the recommendation. It fits what we have learned about AI adoption: [learning speed counts for more than prior knowledge](https://xalt-northstar.com/blog/change-management-ai-adoption/), and that applies to leadership too.

## Judging stays a leadership job

I do not know yet what my log will look like after half a year. Maybe the recommendation was nearly always closer than I was. Then at least I know it for sure and can approve with a calmer mind. If not, I will have learned something important while it was still cheap.

## Sources

- [Lisanne Bainbridge: Ironies of Automation (1983), summary](https://en.wikipedia.org/wiki/Ironies_of_Automation)
- [Patrick Debois, DevOps Enterprise Summit Las Vegas 2024](https://videos.itrevolution.com/watch/1002959794/)
- [IT Revolution: Human-in-the-Loop Is Non-Negotiable](https://itrevolution.com/articles/human-in-the-loop-is-non-negotiable-leading-ai-adoption-in-safety-critical-systems/)
- [Also published on xalt.de](https://www.xalt.de/en/blog/keep-your-judgement-sharp-decide-first/)

**Related:** [Personal Agent](https://xalt-northstar.com/personal-agent/) – The AI agent that grows with you.

For CEOs and leadership teams

## Keep your leadership judgement sharp

A conversation for CEOs and leadership teams: which of your decisions suit a decide-first exercise, and how to set it up with an agent.

[Talk to Philipp](https://xalt-northstar.com/talk-to-philipp/) [Run the AI maturity audit](https://xalt-northstar.com/maturity-audit/#questionnaire)

Read next

- [Personal agents When the AI guesses: what I got wrong during onboarding 7 November 2026 · 5 min read](https://xalt-northstar.com/blog/when-the-ai-guesses/)
- [AI leadership Change Management for AI Adoption: Ask the Dumb Question 7 November 2026 · 6 min read](https://xalt-northstar.com/blog/change-management-ai-adoption/)

---

Solutions

- [Agentic Engineering](https://xalt-northstar.com/agentic-engineering/)
- [Regulated AI](https://xalt-northstar.com/regulated-ai/)
- [AI Services](https://xalt-northstar.com/ai-services/)
- [AI Brain](https://xalt-northstar.com/brain/)
- [Visual Review](https://xalt-northstar.com/visual-review/)
- [Personal Agent](https://xalt-northstar.com/personal-agent/)

More

- [Events](https://xalt-northstar.com/events/)
- [Blog](https://xalt-northstar.com/blog/)
- [Careers](https://xalt-northstar.com/careers/)
- [Contact](https://xalt-northstar.com/contact/)

Company

- [xalt.de](https://www.xalt.de/)
- [Imprint](https://xalt-northstar.com/imprint/)
- [Privacy](https://xalt-northstar.com/privacy/)

**XALT** Northstar

In progress SOC 2 Type 1 · ISO 27001 · ISO 9001 · GDPR. Not certified yet. [Trust Center](https://trust.inc/xalt)

© 2026 XALT LLC
