AI-first: our philosophy
We are building AI into companies that were created without AI.
That is perfectly understandable, to begin with.
A chatbot here, automatic invoice recognition there, Copilot for emails and somewhere a meeting summary. All sensible. It just isn’t AI-first yet.
Xerox PARC / Personal computing
Alan Kay
„Alan Kay (3097597186)“ by Marcin Wichary from San Francisco, U.S.A., 2008
Source: Wikimedia Commons · Licence: CC BY 2.0 · Image credits
Editing: Cropped (image detail), scaled and compressed; otherwise unchanged. Edited by Today is Life.
In the early 1970s, Alan Kay was thinking about a computer that practically couldn’t exist yet: small, personal, graphical and as natural to use as a book. His “Dynabook” anticipated much of what laptops and tablets became decades later. Kay wasn’t trying to make the mainframe of the day a bit more comfortable. He asked what computers would have to look like if you started from scratch, from what they make possible. That is exactly what AI-first means: not building AI into the old organisation. Rethinking the organisation with the new possibilities.
Why this isn’t AI-first yet
We are building AI into companies that were created without AI. That is perfectly understandable, to begin with. There is a customer service team. So it gets a chatbot. There is accounting. So we let AI recognise invoices. Employees write a lot of emails, so they get Copilot or ChatGPT. And somewhere an automatic meeting summary is almost certainly being created right now.
All sensible. It just isn’t AI-first yet. It is roughly as if, in the mid-nineties, you had discovered the internet and then put the printed product catalogue on the web server as a PDF. It was online. It just wasn’t Amazon yet.
The real upheaval only begins when a new technology is no longer used to make the old a bit better. It begins when someone starts to ask what, with the new possibilities, we still need of the old at all. That is exactly where we stand with AI. The interesting question is not: where can we build in AI?
Take a customer service team. You can have an AI summarise the customer’s email. Then an employee reads the summary, opens the CRM, looks for the customer, checks their order, maybe checks an invoice, uses AI to draft a reply and then documents the case. Great. The employee was faster.
But you can also ask a different question. Why is a person doing all this in the first place? An AI can receive the request. It can recognise what it is about. If we connect it accordingly and give it the necessary permissions, it can find the customer and their cases, gather information from different systems, work out a solution, reply, document and trigger the next work step. And if it isn’t sure, it asks a person.
That is an entirely different customer service. Not because we automated the old one. But because we built it again. That is AI-first.
The interesting AI questions
And then some pretty fundamental questions suddenly come up. Do we still need this report, or do we only need the decision that someone used to make after reading it? Does an employee have to gather information from five applications, or can a system bring it together exactly when it is needed? Why does this approval process exist? Because someone really has to decide? Or because our software had no better idea twenty years ago?
And why does a customer wait for a problem to arise, for someone to complain, for a ticket to be opened and then for six people to discuss it very professionally? Perhaps the system can spot the problem beforehand. For us, these are the interesting AI questions. Not: which model should we buy? But: what would we build differently today?
For that, though, you need to know what is possible today. And that is where it gets a little strenuous. The capabilities of these systems are currently changing faster than the processes in most companies. AI can now do much more than generate text or summarise documents. Modern systems can process images, speech and documents, use software, access company data and work with tools. Agentic systems can break tasks down into individual steps, carry out actions, check results and then keep going.
That doesn’t mean you should hand them the company tomorrow morning. AI makes mistakes. It can misjudge connections, invent information and produce nonsense with impressive confidence. The more a system is allowed to do on its own, the more important permissions, limits, logging, control and the question of when a person has to take over become.
AI-first doesn’t mean AI-only. Quite the opposite. If machines can do more, we have to decide more precisely what we leave to them.
That is why we don’t start with a four-phase model. First we look at what actually happens in a company. Not just the org chart and the process description. Those are usually tidy. Reality is more interesting. Who gets which information? Who transfers it where? Who is waiting for whom? Why is an Excel file being passed around? Why is someone retyping data that already sits in another system three metres away? Which decision does a person actually make, and where do they really just click “OK” because the process called for a person?
Then we look at which of this classic software does better, where AI makes sense and where a person still belongs. Sometimes the result is an AI agent. Sometimes new software. Sometimes an entirely different process. And sometimes it turns out the existing Excel spreadsheet was not the problem at all. We can live with that.
AI-first doesn’t mean: people out. I find that debate too simplistic anyway. On one side, we hear that AI will destroy millions of jobs. On the other, we are told by reflex that AI will of course only support and replace no one. We don’t know. But what we can already see: if machines can take over more and more activities, people’s work inevitably changes. Some tasks disappear. Others emerge. Some professions will change more than we would like today.
That is why we also think little of the reassuring line that AI should simply “make every employee more productive”. Maybe. But maybe for some work we will in future actually need fewer people. Then companies should talk about it. And before the software is introduced.
For us, AI-first therefore doesn’t mean optimising people out of the company. It means thinking quite precisely about what we want to need people for in future. That is a difference.
And why now?
Not because everything will be over tomorrow. Nor because every company urgently needs 37 AI agents. But because we are still in a rare phase: the technology is already powerful enough to change real work fundamentally, and at the same time surprisingly little is settled.
We can try things out. Discard. Learn. Rebuild processes. And we can still decide what role AI should play in our companies. At some point much of this will be taken for granted. Then we will probably talk about “AI-first” as little as anyone today seriously claims their company is “internet-first”. Then it is simply the way companies work. Until then, there is quite a lot to do.
Common questions
Do I have to rebuild everything straight away for AI-first?
No. That would often be the wrong start. AI-first doesn’t mean big bang. It means looking honestly at which processes still make sense today at all. Sometimes this starts with a small pilot project, sometimes with a bigger systems question. More important than the size of the first step is the direction.
Is AI-first just automation with a new name?
No. Automation makes existing processes faster or cheaper. AI-first first asks the more uncomfortable question of whether that process should still exist at all. That is the difference: whoever thinks AI-first builds anew where it makes sense, instead of just becoming more efficient.
Does every company need AI agents right away?
No. The term sounds bigger at the moment than it needs to be in many cases. Sometimes an agent makes sense, sometimes classic software is enough, sometimes a clean process with better data helps more. AI-first specifically doesn’t mean forcing agents in everywhere.
How do I find out where AI really makes a difference?
Not with a tool list, but through the reality in the company. Who is waiting for whom? Which information is entered twice? Where does a person click “OK” only because the software knows no better way? That is usually where it gets interesting. Not with the question of which model has the most hype at the moment.
Does AI-first ultimately mean fewer people after all?
Possibly, in some areas, yes. And that is exactly why it should be talked about early, instead of pretending it is only about a bit of productivity. Some tasks will disappear, others will become more important. For us, AI-first doesn’t mean optimising people away. It means deciding more clearly what we really need them for in future.
Ready for AI-first?
We look with you at what is actually happening in your company today, where AI makes sense and where it doesn’t. No AI theatre, no agent obligation, no selling of methods. Instead, an honest assessment of what you should rebuild and what is better left as it is.