Freshly back at my desk after a week of wandering around the countryside and swimming in the sea, the thought occurred, “Do I really need to be here? Why haven’t I been replaced by a robot yet?”.

After all, we have been promised that artificial general intelligence (AGI) is around the corner for some years now, and at large scale fewer humans being needed to do the same amount of work seems inevitable. But could it do my job? I decided, for just one morning, to note down everything I did in my current role managing the IT dept of a small charity in the education sector and to ask the question “Could a robot have done this?”. In each case I would consider the answer in some (some would say tedious) detail, to try and get a sense of how round-the-corner the new utopia is.

9.00 – Catching up on emails, responding quickly to easy queries, deleting salesy ones, leaving the those that need a more considered response.

Of course I’m only seeing a fraction of what is emailed to me because the majority of the spam sent to me is pre-filtered by my email provider. Some of the easy queries could perhaps have been responded to by a bot, but if so the sender could have simply consulted a bot themselves rather than emailing me. Ergo we can suppose this task, as long as it exists, will need the human touch.

Humans 1, robots 0

9.38 – A colleague needs help with a workflow built into a proprietary application

Were this a problem with Power Automate, Co-Pilot would have easily volunteered a solution. However, in this case the software is unknown to internet-trained generative AI, which would only be able to help at a high level, and not the detailed step-by-step required by my colleague.

You could argue that this software could one day have its own AI assistant. However this is unlikely if the software is very bespoke, unless AI becomes much, much cheaper. Lets award a point to both humans and robots for this one.

Humans 2, robots 1

9.48 – An external API (Hubspot) has been updated requiring a change to a power query that is part of a symantic model in Power BI.

In order for a robot to do this it would have to

1. Read and interpret the email from Hubspot.
2. Know that this particular API is in use by a symantic model.
3. Find and read the necessary API documentation.
4. Find and open the file containing the symantic model.
5. Identify and implement a solution to the power query
6. Update the Power BI service and test.

Each of these steps seems to me well within the capabilities of AI. (Confession: AI helped me write the power query in the first place). What’s missing here is all encompassing bot that can read the email and has access to the symantic models in the Power BI service. This would have to be an AI that has access to all the cloud services necessary (email, Hubspot, Power BI) and be given the permissions necessary to make changes. In principle though, there was/is no subtask here that couldn’t be done by a bot.

So… Humans 2, Robots 2.

Conclusion

It appears there are still plenty of occasions when only a human will do, even in an ostensibly technical role. In just one morning I identified the following features where AI currently cannot replace human intervention:

1. Overall oversight. AI can read emails, research online documents and write code. But it still needs a director prompting it to do each of these subtasks and tying them together.
2. Initiative. AI with overall oversight would still need to have initiative to do anything with that oversight.
3. Permissions. AI with overall oversight and initiative will still need permission to make changes which relates to…
4. Responsibility.
5. Dealing with novelty. In common situations AI excels. For example Co-Pilot will create a complex excel formula quicker and better than any human given the right prompts. But with novel, bespoke software, it simply hasn’t had enough material to learn from.