Newsletter #30 · Human Skills in the Age of AI
Build or Buy Is a Human Skill Now — What 19 Days and One Cancelled Licence Taught Me
In short: If you take one thing from this: AI has reopened an old question: what do we build, what do we rent? The answer needs people who know both business and technology. Ahead: why the pendulum swings back at the edges, not at the core; what I could hand over to agents, and what I could not; why almost nobody asks the question; who is allowed to try, and one task for whoever signs.
On 18 October I asked a simple question: do I still need the all-in-one platform that ran my website, blog, podcast and newsletter for €180 a month? Nineteen days later my new website was live. AI agents had built it with me and moved about 190 articles and 96 podcast episodes; 222 old addresses now lead to new ones. The licence is cancelled.
This is a small case, and I want to mark it as one: one person, one website, no IT department, no works council. But the question behind it is not small. For years the answer to "build or buy?" was to rent finished software. AI has reopened the question, and I found that answering it had less to do with technology than with patience, judgement and the courage to try.
If you run a company, the question is already in your building. In McKinsey's 2026 survey of 1,719 people, nearly a third said their organisation had decided against buying at least one software product or feature because it could be built in-house with AI coding tools. Gartner sees up to 234 billion dollars of application spending at risk by 2030, yet expects rented software to change rather than vanish. Who in your company may ask which software you still need to rent, and who may try an answer, is decided at the top, or by nobody.
Before you read on, may I ask you a favour? The result is at digital-leader-program.com. Is it clear who the site is for? Do you find your way around the blog, where every edition of this newsletter now sits in one place? I would be glad to hear it, in a message or in the comments.
The pendulum swings back, at the edges
Companies once had most of their software written for them. Then came software as a service: ready-made packages, rented by the month. By Gartner's figures and forecasts, that market grew to almost ten times its size between 2015 and 2025.
Now the pendulum moves again. With AI, software made to fit is possible again, without joining five services together. How far that goes depends on a company's size, its rules, and where anyone is allowed to build at all. The cases of full replacement that The Information documented in July are small: five firms of 20 to 70 people that cancelled their Salesforce or HubSpot contracts and built their own. A US health insurer, Curative, says it cancelled a Salesforce contract worth 600,000 dollars a year after building a replacement in about two months. It looks like a game for small firms.
The evidence is less tidy. Klarna, the best-known example, did shut down Salesforce, but its chief executive later wrote that it had not replaced its rented software with a language model. And AI is used far more in large companies than in small ones: 55 against 17 percent in the EU's 2025 figures.
So the line does not run between small and large. It runs between edge and core. At the edge, in internal tools, daily routines, a website, people build again. At the core, in payroll, finance, the HR system, the bought software stays for now, and agents begin to work on top of it. Sanofi, with about 75,000 employees, runs its own AI assistant. Some 60,000 use it, and it draws on data from systems the company rents. "Software is becoming far more individual to the company," its chief digital officer told Fortune. The vendors have noticed: Workday has said it will charge for outside agents working on its data.
What I could hand over, and what I could not
The agents did the moving and the code. What stayed with me was everything that needs judgement or consent: the tone, the prices, which client voices appear. The people in my workshop photos have not yet agreed to be shown, so for now the site runs those photos with AI-generated faces in place of theirs. I switched the domains myself. And when the clean-up of the old server turned up three suspicious files, I deleted them. The design changed too: my font and colours had felt too hard to me, and the move was a good moment for small corrections.
The technology was the smaller part. I needed some knowledge of servers, and that can be learned quickly with AI if you like trying things out. The harder questions were human. How much time do I lose in the worst case, on something that turns out useless? How much can I break? And it took patience: waiting, and not knowing whether it would run as smoothly as I pictured it. It cost attention too: about 280 requests.
The sum, because whoever signs will ask. The platform cost €180 a month. The new site costs about €15 a month to run, and my share of AI costs is higher, perhaps €60. I expect maintenance to be easier than before and to save me time; that is an expectation, not yet a record.
But the money is the smaller gain. I can now create pages in large numbers, improve them and keep them current. The site is filled with far better data, far more information comes back from it, and far more runs by itself. When an edition of this newsletter is written, it is published on the website, translated and sent out by email, without me carrying it from tool to tool. The connection to my customer system is far better too. That is new marketing, and with it new chances for revenue, none of which was possible so cleanly before.
If you sign budgets and nobody in your company is allowed to ask which software you still need to rent, one thing is enough. Forward this edition to whoever runs people development for you, with one line: "I want us to work on this over months, not in an event. Find me something like it." What to look for: a programme that starts with the leadership team, runs over months rather than days, and treats human skills and AI as one subject. If you run people development and just received that line, or still have budget for this year: that is the work I do, since 2010 with organisations from start-ups to adidas and Fraunhofer (references). One conversation, and you'll leave with a clear next step whether or not you work with me: book a call.
Why almost nobody asks
I believe many people in companies are afraid to build something themselves, or even to ask whether an existing system still makes sense. There are servers, networks, and data that many others depend on. Too much hangs on it.
Part of that fear is bias, and it cuts both ways. We stay with what we have, and we hold on to what we have already paid for: the licence, the rollout, the training. Researchers described both decades ago, the status quo bias (William Samuelson and Richard Zeckhauser, 1988) and the sunk-cost effect (Hal Arkes and Catherine Blumer, 1985). And the builder is not immune. Michael Norton of Harvard Business School and colleagues found that people value what they assembled themselves more highly than the same thing assembled by someone else, and called it the IKEA effect. So the person who built a tool should not be the only one to judge it.
Part of the fear is right. In a Stanford study, people who wrote code with an AI assistant produced less secure code than those without, and were more confident that it was secure. Curative's chief executive calls maintenance "definitely one of the most challenging pieces", and a textbook rule of thumb agrees: maintenance takes 40 to 80 percent of what software costs over its life.
Britain's National Cyber Security Centre published a scale in June: the higher the stakes, the more care and oversight. Letting AI write the code without reviewing it closely "can often be perfectly fine" for test versions and internal tools with no sensitive data; sign-in logic, sensitive personal data and passwords need more care. The EU's AI Act has made this a duty since February 2025; in its current wording, companies that provide or use AI systems must "take measures to support the development of AI literacy of their staff". Helping people understand these tools is the employer's job by law.
Who is allowed to try?
So the question needs an owner and a boundary. Stefan Thomke of Harvard Business School describes Booking.com, where anyone may test anything without a manager's permission, on a platform built for it. Only about one in ten experiments there brings a positive result. Trying is cheap when the area is safe.
The German Economic Institute asked nearly 5,000 employees: of those who use AI at work, 29 percent use tools their employer did not provide, and among managers 35 percent. The authors read it as a signal of demand. Even the managers are faster than their own companies.
Who should hold the boundary? My guess: people who understand both the business and the technology. In Germany about 233,000 people study business administration, 158,000 computer science, and 70,000 the subject that joins the two, business informatics. I know of no clean evidence that these bridge-builders decide whether such projects succeed; what exists comes from surveys. But I think this link is more useful now than it has ever been.
That is why I call build or buy a human skill now. The tools can do the building. Deciding what may be tried, by whom, and where the core begins is a judgement, and somebody has to own it.
Your turn
If you sign budgets. Consider naming two things: who may question a licence, and who may try an alternative at the edge, with which data. Then send this edition to whoever runs people development, with the line from the middle of this page.
If you lead a team. List the tools your team rents. For each, one line: core or edge, used or unused. For one edge tool, try asking what an agent on top of it could do, before anyone talks about replacing it.
If you develop people. Put understanding of technology into leadership development, not only into IT training. Look for the people who speak both languages and give them a mandate. If you need the mandate, forward this upward; the opening is written for the person who signs.
Your 1% this week, whoever you are. Take one piece of software you pay for. Write down what you really use of it.
Until next week: which licence in your company would you question first, if you were allowed to? Tell me in the comments. I read every reply.
P.S. This newsletter now also comes by email: every Wednesday morning in your inbox, and you can simply hit reply. The sign-up form is at the bottom of every page on digital-leader-program.com.
If you want to bring these themes into your organisation, here's that link again: book a call. You can also explore the Leadership Essentials Workshop Series for teams.