What leadership sets and what the teams take on
Salesforce describes the agentic enterprise as a business where humans, AI agents and platforms work together for customer success.[S2] Our approach to this is the hybrid organisation. Two sides have to come together. Leadership decides which work stays with people, which people and agents do together, and which agents take over entirely. The teams give individual agents a described role, an accountable person and feedback.[S1]
Bottom-up only ends in one-offs: every agent brings its own knowledge, every correction improves only that agent, and maintenance grows with each one. Top-down only stays a slide, because nobody learns from real cases what the target picture means for people and processes. At the digital unit of an automotive supplier we talk to, the target picture is forming at the top right now, and its digital unit wants to try it on itself first. Both are needed at once: the first agents emerge in teams, and what they need becomes the foundation for everyone.[S5]
Three leadership decisions and the teams’ role
The table separates three leadership decisions from the teams’ role design. The bar is agreed before the first pilot, the target picture and foundation before scaling. The central unit that owns this becomes a kind of HR department for agents: it knows which agents exist, who is accountable for them, and when one is reused or shut down.[S1]
| Decision | Who makes it | What it sets |
|---|---|---|
| Target picture | Leadership | Which work stays with people, which agents take over, and which areas become agentic, hybrid or human. |
| Shared foundation | Leadership decides, the central unit (for example a staff unit) builds and maintains | A standard for role descriptions, a register of agents with roles, rights and accountable people, shared knowledge, approvals, onboarding and the capacity to check results; every further team starts on it. |
| The bar | Leadership with the pilot teams | What must be true after the pilots for the programme to scale: time per task, quality rated by your own experts without knowing whether a person or the agent produced it, cost per correctly completed case and the value of freed capacity. |
| Roles | Business teams | Which agent takes on which tasks, who onboards it, reviews it and takes over when it fails. |
Where scaling gets stuck today
In a Lünendonk survey of 180 IT and business leaders at mid-sized companies and corporations in Germany, Austria and Switzerland, 66 percent say fewer than a quarter of their previous AI pilot projects have reached productive use. The study names missing integration and data connectivity, security and data protection, a shortage of staff and poorly coordinated initiatives as the obstacles.[S3] In June 2025 Gartner predicted that over 40 percent of agentic AI projects would be canceled by the end of 2027.[S4]
Another obstacle is human in the loop: people reviewing and approving results. As long as every output must be checked in full, human effort grows with each agent. Our approach therefore starts with training the agent for its role: corrections checked by subject experts are recorded in shared knowledge with a defined scope and help the next agent too.[S1]
Reliable results build trust. The accountable person then approves more autonomy for clearly bounded tasks. These need fewer individual approvals; exceptions and high-risk cases stay with people. Sampling and fresh tests after changes check that quality holds. The aim is to reduce review effort per correctly completed case.[S1]
The capacity gained belongs in the target picture: reviewing additional tenders or advising customers the team previously had no time for. Leadership and team therefore measure both the full effort, including human review, and the value of the additional work.[S1]
Start with one team and a clear role
One team, a recurring task and an accountable person are enough to start. In a corporation, the target picture and first teams develop in parallel. MING starts with an orientation conversation and an assessment workshop that produces the role description, business case and agreed bar. After testing and again after the pilot, the accountable person decides whether to scale, revise or stop. Reviewed rules become reusable by other teams within a defined scope. Our approach is described in our lead article on building a hybrid organisation.
Our hybrid organisation service explains how we introduce an AI agent into your team.
Evidence and limits
- The three decisions and the starting plan are MING’s method from client projects and its own agents; no measured client result.
- Lünendonk figures are self-reports and Gartner makes a prediction; neither proves the effect of this method.
- Observations from corporations are anonymised; duration, cost and results depend on the individual company.
