Lesson 5 · 20 min
McKinsey, Deloitte, Microsoft and Gartner publish about the agentic company almost every month, and those numbers reach your desk as a decision: cut, hire, promise. In this lesson you read what each one says and learn to separate research, self-reported experience and forecast. At the end, the owner of Horizonte shows up with a report, and you are the one who answers.
Start the lesson
Goal: read a consulting or vendor report and separate what is research, what is self-reported experience and what is forecast.
Before you use a number from a report in a decision, it goes through these four questions, in this order. Click each one.
All the facts in this lesson come from the October 2026 market overview of the program, with date and source. Anything that only had a secondary source was left out.
Each consultancy and vendor sells its own frame. See what each one says and what kind of claim it is.
McKinsey describes the "agentic organization" (2025–2026): small teams built around an outcome as the unit of the organization, with 2 to 5 people supervising 50 to 100 specialized agents, and accountability designed in from the start (accountability by design). It also says the span of control may narrow at first and widen later, with supervision tools.
Type: self-reported experience from the consultancy, not a measured study. It is what the firm observes and proposes to its clients.
Who publishes it: a consultancy that sells reorganization. How they measured: reporting. Forecast: no. What I take away: the unit of management becomes the team built around an outcome.
In Tech Trends 2026, Deloitte treats agents as a "silicon-based" workforce: integration with data and operations, graduated autonomy with human supervision triggers, and training for supervisors. In Human Capital Trends 2025, it discusses the future of the middle manager.
Type: consultancy view, with trends and recommendations. It works as vocabulary and as a checklist, not as a number.
Who publishes it: a consultancy. How they measured: trends and cases. Forecast: in part. What I take away: whoever supervises an agent also needs to be trained.
Work Trend Index 2026 (05/2026): 20 thousand people in 10 countries. 16% are "Frontier Professionals", who use agents in multi-step workflows; 58% do work they could not do a year ago (80% among the Frontier); when the manager uses AI in front of the team, perceived value goes up 17 points.
Type: survey with a sample, run by the company that sells the tool. An analysis by The Microsoft Cloud Blog (05/2026) reminds readers that these numbers should not carry a business case on their own.
Who publishes it: an AI vendor. How they measured: opinion survey with a large sample. Forecast: no. What I take away: the manager setting the example matters; the 58% is not from my company.
Forecast made in October 2024: by the end of 2026, 1 in 5 organizations will use AI to flatten their structure and cut more than half of middle management roles. In 2026, in the Hype Cycle for agentic AI, the same Gartner recorded that only 17% of companies have agents in use.
Type: forecast, with a deadline that is coming due now. Critics point out that taking a layer off the org chart does not eliminate the work it did: coordinating, reviewing, training the people who join. That is where the agent manager comes in.
Who publishes it: a research consultancy. How it measured: analyst forecast. Forecast: yes, for the end of 2026. What I take from it: list the management work before cutting the layer.
One sheet for each report or news item that reaches your desk looking like a decision. It takes ten minutes and saves a whole meeting of opinion against opinion.
CRITICAL READING SHEET · <report title> · <who publishes it> · <date> Arrived through: <who brought it and which decision it is meant to justify> 1. WHO PUBLISHES IT AND WHAT THEY SELL <consultancy / vendor / institute / press> · sells: <...> What they would have no interest in showing: <...> 2. HOW IT MEASURED [ ] survey with a sample: <how many, where, when, which question> [ ] self-reported experience (what they saw with clients) [ ] case study [ ] opinion / column [ ] doesn't say 3. IS IT A FORECAST? [ ] no [ ] yes: made on <date>, for <deadline>. Can we check it yet? <yes/no> 4. THE NUMBER THAT WILL BE USED "<number and exact sentence>" · type: <survey / account / forecast / opinion> Definition used by the report: <what counts as "uses", "integrates", "adopts"> 5. WHAT APPLIES TO ME What I take from it (1 sentence): <...> What I need to measure here before deciding: <measure, for how long, who>
You are a skeptical analyst. I will paste an excerpt from a market report. Do not summarize and do not give an opinion on the subject: classify the claims. Excerpt: """<paste here>""" For each claim with a number or trend: 1. Say whether it is a survey with a sample, self-reported experience, case study, forecast or opinion. If the excerpt doesn't make it clear, say "not stated". 2. If it is a forecast, point out when it was made and for when. 3. Point out the definition used (what counts as "adopt", "use", "integrate"). 4. List what I would need to measure in my company to get the equivalent number. Do not invent data that is not in the excerpt.
Work Trend Index 2026 · Microsoft · 05/2026 · brought by Carlos to expand Hora to collections (Horizonte, fictional study company)
Simulated conversation for practice. The owner of Horizonte arrives on Monday with numbers from a report and a decision almost made. You are Patrícia.
Choose your answer. The comment appears right after, and the conversation continues based on your choice.
The conversation with Carlos for practice; a real report for keeps.
Run the simulation above. If you don't reach the top score, read the comments, start over and choose again. For each number Carlos brought, write down the type of claim.
Done when: you reached 6 out of 6 points and have noted the type of all three numbers: self-reported experience (McKinsey), forecast from Oct/2024 (Gartner) and vendor survey (Microsoft).
Pick a report, news story or post that reached your company citing an AI number, or one of the four from this lesson. Fill in the critical reading sheet and run the prompt on the main excerpt. Compare what you classified with what the prompt flagged.
Done when: the sheet has all five parts filled in, the number in quotes with the type beside it, the definition the report uses, and the last line with a measure, a deadline and a name.
5 questions. The explanation appears after your attempt.
Question 1 of 5
"2 to 5 people supervising 50 to 100 agents" (McKinsey) is what kind of claim?
Question 2 of 5
When was Gartner's forecast about cutting more than half of middle management made, and for when?
Question 3 of 5
Why does a Work Trend Index number deserve extra care before going into a plan?
Question 4 of 5
What is the argument of the critics of "flattening" middle management?
Question 5 of 5
What is the last question on the critical reading sheet?
Choose an option. The explanation appears after your attempt.