Using AI can prevent friction and save and money, when used wisely. AI won't replace experts and craftspeople though. Here is why (not how, though, that would be another post) and how to explain to our customers what AI can do and what it can't.
Can AI make a Website in 2 Days?
"AI can do a website in two days" isn't web design and development. It's called prototyping. Web designers, project managers, and web developers fear those customers who don't know what they want, but who'll tell you what they didn't want only after their work has been done.
Are you that type of Client?
If you're that type of customer, you can leverage AI to sort your thoughts and do some sketches and click dummy mock ups. AI isn't great at producing quality code or taking responsibility, but it's good at producing anything that you can look at and talk about. AI won't get tired and doesn't charge you hourly rates (at least not yet at the time of writing) for discussing over and over verbosely and trying to answer the most basic questions.
Hallucination, Stubbornness and Alternative "Facts"
As ChatGPT, Claude, and other AI services tell you in their fine print, AI can make mistakes. AI is quite likely to make mistakes. AI is known to hallucinate, tell you what you want to hear or else insist on a made up random claim not sustained by any facts.
Too long, too often
When AI says "often", it often means "always" or "never" or "maybe, I don't know". When you see a word like "often" or "many", take it as a guess or ask to elaborate and research, if it matters. Often, it doesn't. AI often makes too many words and you don't have to read it all.
Cliffhangers
AI tends to ends its wordy answers with cliffhangers offering additional things it can do for you. That's a trap. Even when it's free or paid with a flat rate plan, you still pay with your time.
Understanding the Limits
You don't need to understand AI any more than you need to understand the inner workings of your car or your smartphone. But you should know how to use it and what it can do and what it can't.
AI is neither Creative nor a Craftsperson
AI is no creative, but a remixer. AI is no craftsperson, it's only well-read. AI has likely read 90% of everything relevant that has been published in recent years, including everything that's outdated now.
Prototyping as Basic Preparation
Using AI is like discussing an idea with your brother or reading a book from your local library to understand some basic concepts. Not bad for a start. You'll get an idea, proceed on your creative path and have smarter questions to ask when you meet with a real expert.
That can save everyone a lot of time, friction and money.
Discussing AI with Coworkers and Customers
AI can be helpful in certain situations. Accepting that as a fact can help us clarify what it can do and what not and why. AI can't replace creative professionals (unless for basic requirements) but telling customers and coworkers how to use (and not overuse) it, can give us back more time to focus on the real work that requires human expertise.
AI Cost Projections
Asking Claude AI about the total cost of using AI, it starts with a a forecast of the obvious pricing:
- For individual users: prices are largely stable or declining
- For businesses: total spend is rising fast ...
- A looming risk: the end of subsidies
According to Zylo's 2026 SaaS Management Index, organizations spent an average of $1.2 million on AI-native apps — a 108% year-over-year increase. A big reason is the shift away from flat fees: unlike traditional software where the cost to serve an additional user is near zero, AI involves significant variable costs including GPU compute, token processing, and data transfer, which vendors increasingly pass through to customers.
And that's not the total cost of ownership yet.
Maintenance issues, already known as technical debt in traditional software development (exact coding), are compounded in AI systems. Models drift, outputs degrade, prompts need constant tuning, and integrations break with every model update, all of which require ongoing human time and expertise that rarely appears in vendor pricing sheets.
The true cost of AI
The true cost of AI is the sum of compute costs, integration work, maintenance overhead, staff time, and the strategic risk of building on foundations that may shift when subsidies dry up.
Last, but not least, there's the rising risk of low-quality "vibe code" only superficially reviewed by human engineers or other AI agents, making a large codebase inconsistent, violate established principles of software engineering, and introduce security risks that might may go undetected until it is far too late to address them cheaply.
The Human in the Loop
Another poorly worded business buzzword, "the human in the loop" is the necesseary supervisor: an experienced software engineer skilled in coding, debugging, thinking outside the box, and communicating. Such a person acts like a translator between machine logical and what human stakeholders (business owners, founders, CTOs, designers) and their clients actually need.