If Your AI Builders Left Tomorrow, Would Your AI Capability Go With Them?
· 4 min read
· 4 min read

Consider this example. An agent handles a routine operations task, and its builder moves to another department. She does a handover before she goes: a colleague is shown what the agent does, where its outputs land and who to call if it stops. For several weeks nothing goes wrong.
Then a model update changes how the agent handles one type of case. The colleague finds that the agent still runs under the builder’s account and her connections to the case system. Nobody on the team knows where its instructions are kept, what she used to check after a change, or how to test a fix. The handover covered the task. It did not cover the maintenance.
The people who build AI systems may also be the only ones who understand their instructions, design decisions and workarounds for tool limitations. The rest of the team uses the outputs but may not know how to maintain or adjust the systems when something changes.
If those people left, what could the team keep running without them?
The builder’s maintenance decisions can go unnoticed while an agent is working. Someone taking over needs more than the prompts. They need the instruction files and skills the agent runs on, the systems it connects to and under whose credentials, the permissions it has been given, the test cases used to check it after a change, and what to do when the underlying model is updated.
That knowledge covers both what the operation needs to achieve and how the system delivers it.
A system can look stable because its builder catches problems before they reach the wider team. Instruction changes, error checks and workarounds keep the outputs reliable, so the dependency may never appear in the risk register.
The team’s ability to maintain the system remains untested while that support is available. The skills gap may only become visible when the specialist leaves or changes roles.
Recording the instructions and workflow does not necessarily capture why they were designed that way. The builder may rely on judgement developed through daily use without recognising that someone else will need it.
Without that context, whoever takes over may have to reconstruct the reasoning behind the build before they can maintain or adapt it.
The ICO’s guidance on explaining AI decisions asks organisations to document design choices and the reasons behind them.
An agent can be used across a function while only a few specialists know how to maintain it. Rising usage shows that AI is becoming part of daily work, but it does not establish whether the wider team can keep the system running or adapt it when requirements change.
When reporting adoption, identify who else can maintain the system if the specialist is unavailable. If performance still depends on that person’s support, include the dependency alongside the usage figures. Leaders need to understand both how extensively AI is used and whether the team can sustain it.
Many agents run on their builder’s own account, with connections to business systems authorised under that person’s credentials. When the builder leaves and their access is removed, the agent may stop without warning. If their access is left in place, the agent keeps acting in live systems with no owner. Both outcomes come from the same gap: the agent was never set up as something the team owns.
If no one takes over maintenance, instructions may become less effective as conditions change, and workarounds may no longer be applied. Outputs can deteriorate gradually, with the early signs mistaken for unrelated problems.
The team can see that something is wrong but may not know what to check or change. Finding the cause takes longer without the specialist’s knowledge.
Knowledge transfer may have been left to everyday collaboration rather than planned into the work. Colleagues taking over may have a description of the task but neither documented reasoning nor experience of maintaining the agent alongside its builder.
They are then learning how the system works while trying to restore its performance.
Our article on how to build AI skills across your team sets out the method: map who holds the capability, pair colleagues on live work, document design decisions and write AI responsibilities into roles. Two further steps deal with the departure risk directly.
Give every agent a named second owner. Move its connections from personal credentials to a team or service account, and keep its instruction files and skills in a shared repository.
Ask the second owner to run and maintain the agent for a week while the builder is still available but not involved. Whatever they cannot do without asking is the handover that still needs to happen.
Identify which AI systems depend on one person and what prevents others from taking over. The gap may be in skills, design knowledge, access or assigned responsibilities, so training alone may not resolve it.
Envisago’s AIVOM™ Reading shows where your operation stands across Value, Design, Capability and Performance, including whether capability sits with too few people, and what to prioritise next.
Start your free reading at aivom.envisago.com.
The Power of AI. The Potential of People™.
The AIVOM™ Platform. Redesign, manage and measure how your operation works with AI. Start for free at envisago.com.