In January, I fired off 5 theses on Agentic AI.
Today is half-time. Time for an honest review.
Based on the numbers rather than gut feeling,
the numbers that have landed on the table since then.
1️⃣ Agents become operating systems ✅
Came true, faster than expected.
McKinsey describes the shift to agent-native enterprise software as the defining architecture pattern of 2026. The key question has moved away from the model and towards the orchestration layer.
2️⃣ The agent labour market emerges 🟡
Here I was too fast.
What is actually emerging are new roles for humans, not for agents. Forbes maps around 20 new job profiles around Agentic AI.
IBM shows: one in four companies already has a Chief AI Officer today.
3️⃣ Orchestration becomes a core skill ✅
The thesis was too cautious.
Orchestration is more than a skill. It is the bottleneck.
McKinsey measures: 62% are experimenting with agents.
Only a quarter scale them into operations. In most business functions, the scaling rate is around 10%.
Pilot yes, production no. That is exactly where everything will be decided in 2026.
4️⃣ Lifecycle management becomes reality 🟡
For the pioneers yes, for mid-sized businesses not yet.
Gartner predicts: more than 40% of Agentic AI projects will be stopped again by the end of 2027. Escalating costs, unclear value, lack of control.
Bitkom backs this up from a German perspective: a third of AI users report that the rollout turned out much more expensive than planned.
Lifecycle sounds unsexy. Until it is missing.
5️⃣ Ownership becomes a question of power ✅
Came true harder than expected.
Bitkom 2026: 41% of companies use AI. Only 21% have a formal strategy. It is exactly in this gap that ownership is being fought over today.
19% have already cut jobs as a result of using AI.
The number that appears in no keynote.
Responsibility has long stopped being a tool question.
It is a leadership question.
And now a new thesis I was missing back then:
6️⃣ The data foundation beats any model
In January, I wrote about architecture, orchestration
and ownership.
About what really decides things in the end, not a word:
the data the agent gets to see.
A recent study from May puts it in one number:
41% already use Agentic AI in production.
Only 15% have a data foundation built for it.
85% are building autonomous systems on a foundation that was never meant for autonomy. The result is more than “a little worse”.
It is “wrong, faster”.
My half-time verdict:
Three theses fully came true. Two partial hits.
And a sixth that, in hindsight, turns out to be the most important.
The studies agree on one point:
In 2026, success or failure will hinge on something other than technology.
It hinges on the invisible prerequisites: data, responsibility, leadership.
What do you think?
Which topics do you currently see as the most exciting in the AI market?