What can a hotel doorman teach us about using AI well?
Most people assume that the doorman of a hotel has one job - to open doors.
Automatic doors are a lot cheaper than doormen, so looking purely at what’s visible makes a pretty good case for getting rid of the doorman altogether. If you take this approach though, you’ll soon notice that there are all sorts of tasks that a doorman does that don’t fit into the job description.
The doorman does indeed open the door, but they also welcome guests, identify confused people, discourage antisocial behaviour, and solve ad hoc issues before they turn into something major. The hotel runs more smoothly with them around.
There’s a temptation to automate this visible and ostensibly simple task but miss out on the hidden complexities of the role and the value of having knowledgeable people interacting with the hotel “system” day-to-day. The advertising executive Rory Sutherland called this the doorman fallacy in his book Alchemy.
The main reason this tends to go wrong is that you can see the visible output of the doorman in that they open doors, and with data insights you might even be inclined to capture data on how many times they open the door and how long it takes them, but optimising around visible outputs misses half the story.
The “doormen” in your business
So how can a software business apply this insight to automating the right things?
As we’ve written about in our other blogs, AI coding agents such as Claude are almost too good. They can generate impressive-looking code, and there’s an increasing move to boil down the role of software engineer to reviewing AI output (and so maybe you don’t need so many software engineers…), but this ignores the hidden value that a human touch adds to your software solutions.
A good software engineer writes good code, but they also do all sorts of other things. They notice duplicated business processes, they draw on past experience to question the requirements, they take a strategic view across the whole business to ensure consistency in good practice, and they onboard new people to the business into this same way of thinking. If you distil the process of creating software down to just the visible metric of the code, you miss out on all these extras.
More than that, your developers are active synthesisers of organisational knowledge. They remember what was tried before and why it failed, and they’ll see the kinds of issues that keep coming up with customers before you capture them formally. This human insight gives you an edge over the AI that needs the right input to be able to make the right judgement.
Humans acquire this context fluidly by being embedded in your business. They identify opportunities for improvement before they become problems for your customers in the same way that the doorman sees a flustered guest in the lobby and helps them before they leave a bad review.
Where AI really works
This isn’t an argument against using AI coding agents though.
AI is great for helping get to grips with unfamiliar libraries, writing well-trodden functions quickly, and managing large codebases, and you absolutely should make use of it for these purposes. AI performs brilliantly when there are already lots of examples of good solutions. Solving a standard problem is quicker than ever.
These are all constrained problems and given the right constraints AI can be great. Be careful though that not all of your problems are constrained like this.
How automation goes wrong
Whilst AI can add a lot of value, what I want to caution against is assuming that because AI generates good code it’s the right code for your situation. Humans only develop the experience to identify the right code through practice and getting the battle scars of seeing where their solution didn’t quite work.
A business that uses AI correctly will avoid the temptation to look only at the visible metrics on software output and will continue to incentivise thinking deeply about the problem before diving into a solution. If you get too reliant on the AI to solve the problem, you’ll get fine solutions most of the time, but you can really use it as an accelerator when your people have got really good at making judgements on what works and what doesn’t.
The hotel doorman adds value not because they follow procedures well, but because through hundreds of interactions, they’ve developed an instinct for when you shouldn’t follow the procedure. Of course, some companies really are paying developers to produce largely routine applications. In those cases, AI alone probably does change the economics. The mistake is assuming every software problem looks like that. And much like a hotel doorman, those instincts once lost can be almost impossible to replace.
AI-enabled change with a human touch
AI has the potential to change how experts spend their time. Using AI agents for problems with established solutions frees up time to focus on where people add the most value – getting the broader architecture right and really understanding your customers deeply.
A place where software specialists can really add value is understanding your context in-depth and selecting the right solution for you. They’ve seen the sorts of problems you have and what did and didn’t work for other people, and they’re able to use this experience to know when the design isn’t right.
When you really understand the requirement, you can use AI as a powerful accelerator to get to the right outcome quicker than you could before. Getting that right means recognising the hard-to-measure hidden value, and getting help from experts, rather than assuming the AI will be good enough on its own.
AI works best when you pair it with people that really understand your business. Don’t mistake the visible output as the only valuable output. AI can allow your experienced teams to use their knowledge to deliver bigger and better things for your business, multiplying their value, rather than simple cost savings. And as with our rueful hotelier, you may find that these savings can come with a great cost of their own.
