AI is a junior colleague

Here is an interesting perspective to apply to your employees’ acceptance of AI in their work – consider AI is a junior work colleague that every employee has at their disposal. With the roll-out of new AI capabilities and agents, there is often push-back from employees who fear that their job is at risk, or they don’t trust AI. Taking a psychological approach that AI is not a tool, but instead is a junior colleague, may assist the adoption and understanding of the new era.

AI is not just a tool

We are in an AI world. It is not just another tool at our disposal – AI is not like a new piece of software on our laptop or a new device. Instead, AI is everywhere, and we need to get our heads around the new paradigm to ensure we are not left in the wake. Admittedly, what most people think when they use AI, is that they are going to a website for ChatGPT or opening a new app for Copilot or Claude. However, the range and reach of AI is so much further, we need to consider AI as a colleague instead of a tool.

Resistance to AI

Employees may be hesitant to the world-changing capabilities of AI being let in to the realm of their job. People fear that AI will replace them, or that it will remove so much of their normal job that they will no longer be useful to the company. There is also resistance from the knowledge that AI hallucinates and makes errors or even makes up information. Staff are logically concerned that not only will their jobs be changed, but it could be something that is worse than the work they do, just because it is cheaper.

What is AI good at, and what is AI bad at?

AI is great at repetitive and standard tasks. It can understand small variations in instructions, and adjust to produce consistent output. It can be incredibly useful at removing “boring” tasks from people – but the best people to set it up to do that, are the people who normally do those tasks. However, AI can be really bad at creativity and problem solving, because it is limited on the information it has access to – it can’t make innovations based on people’s personalities or understand the physical world’s impact on a project or initiative. That is where people excel.

AI as a junior colleague

So, to take a different viewpoint, if we tell employees that they need to treat AI as another person – a junior colleague, but someone who needs to be monitored and guided, their work checked, a colleague who is a person who is happy to do the menial and boring/repetitive tasks. The AI systems will need to be guided by the employee, trained and mentored, and no new work released until a human has reviewed and assessed the output. The trust in the AI needs to be won by the AI, through monitoring, guidance and oversight – like taking a green new graduate under your wing.

A delegation ladder for supervising AI

If AI is a junior colleague, the practical question is how much freedom it should get, and when. A new starter earns responsibility one task at a time, and AI should too. Here is a simple ladder:

Level	What the AI does	What the human does	Move a level up when
1. Suggests only	Offers ideas, summaries and options	Decides and does the work	The suggestions are consistently useful
2. Drafts for review	Produces a draft	Reviews every output before it is used	Few corrections are needed over a sustained period
3. Acts with sign-off	Takes the action once approved	Approves each action	Approvals are routine, with no serious errors
4. Acts, with sampled checks	Handles routine cases alone	Audits a sample and handles exceptions	The error rate in the sample stays within the agreed limit
5. Acts within limits	Acts alone inside defined limits	Monitors results and sets the limits	Limits are reviewed regularly and never exceeded

Three rules make the ladder work:

1. Trust is per task, not per tool. The same AI can sit at Level 4 for tidying meeting notes and Level 1 for contract advice.
2. Move down as well as up. After a serious error, a change of model or new kinds of data, drop a level until the evidence returns.
3. Set the limits before you start. Decide what the AI may never decide alone, such as complaints, refunds above a set amount, personal information and safety issues.

One caution about the metaphor itself. Calling AI a colleague can make people feel less responsible for its work. A recent Fortune article reports Boston University research finding that people caught 18% fewer errors when work was described as coming from an “AI employee” rather than a chatbot, and Computerworld reports a Harvard Business Review study in which some managers took less personal responsibility for results once AI was framed as an employee.

So I add one rule: unlike a real junior colleague, AI cannot be accountable. The person who delegates owns the result.

None of this means AI should replace real graduates, who need that routine work to learn their craft.