
Your Website Has a New Job
AI agents are changing how people search and decide. Your website stays relevant—but its job is changing.
Read more: Your Website Has a New Job
Over the past year, AI coding agents like Claude Code have changed how fast we build. One example from our own projects: a budget of €50,000 used to pay for Phase 1 of a client project, typically the MVP of a web application or a content platform tailored to the client. Today, the same budget often covers later phases and features that used to be optional follow-ups. For anyone who sells skilled work by the hour, this raises an uncomfortable question about what we are actually billing for.
This post is written for freelancers, agencies, and consultancies in web, BI, data, ERP, or DevOps, and for the clients who work with them. We don’t have every answer yet. At b13 we have always tried to be honest with our clients, and that includes being open about how we think about this while the industry is changing.
The obvious answer sounds fair enough: we are faster, so we bill fewer hours and the client saves money. Once you run the numbers, it stops working. If a 10-hour task now takes two, an agency billing by the hour earns a fifth of what it did for the same result. The better we get at our craft, the less we earn. Salaries, training, infrastructure, and open-source contributions don’t drop by 80% because the typing got faster. Clients never really wanted to be billed by the hour. They wanted a working result and someone who takes responsibility for it. Hours were simply the easiest thing to count. What clients need hasn’t changed with AI, but counting hours has become a far less reliable way to measure it. Speed on its own isn’t what we sell, either. Anyone can vibe-code a quick web application or their own TYPO3 extension with a few prompts. We work differently. We listen first, think about the problem, and then write code. Understanding each other takes time, and in our business that is the part AI can’t take over (yet). Where AI helps a lot is when something isn’t working. It often points us in the right direction much faster than hours of trial and error would, which saves the client money and saves the rest of us a few gray hairs.
For well-defined work, we think billing by the hour is increasingly the wrong model, for agencies and clients alike. Agencies like selling hours because it feels safe, and clients like buying them because it feels controllable. In the end, neither side gets what it actually wants.
With hourly billing, the client also pays for our learning curve, our detours, and our slow days. With a fixed price for a defined result, the client pays for what we agreed to deliver, and the risk of getting there stays with us, where it belongs. That way our experience benefits the client instead of reducing our revenue.
Two terms often get mixed up here. An hourly rate is simply a price. Time and materials (T&M) is a contract model in which the client pays for the actual effort plus costs and carries the risk if things take longer. With AI, “materials” now also include tokens, model subscriptions, and compute, which also have an environmental cost (see Oli’s blog post). T&M can still be a fair model if both sides can see what the time and the materials consist of today. What doesn’t work is treating an AI-assisted hour as if it were the same unit as an hour five years ago.
Other parts of the industry are moving in the same direction. In a September 2026 WFA study of 69 of the world’s largest advertisers, labor-based agency pay had fallen from 54% in 2011 to 19%. The largest group, 33%, now pays fixed fees for outputs.
We don’t use one model for everything. In practice, we combine:
We have heard big promises about new ways of working before. Years ago, the industry adopted Agile, first Scrum and later Kanban. For some agencies and clients it worked well, especially product teams with a dedicated product owner and a long-running backlog. For many others it had little to do with how clients actually need us. Most of our clients don’t need a sprint review every two weeks. They need us to listen, think through problems with them, and give feedback when it matters. Quite often, a sprint turned out to be a block of hours with a new name.
Scrum by the book, with fixed sprints and all its ceremonies, never really suited our company culture. Kanban came closer, but we never treated either as a rulebook. Maybe we were already flexible enough: we talk to clients directly, adjust the scope when reality changes, and ship when something is ready. For us, pricing the result is not a new framework but a price tag on how we have worked for a long time.
For years, that has meant two things in practice: a retainer for support and ongoing smaller projects, and a fixed price for anything above a certain budget. When a fixed-price job took longer than planned, we covered it, not the client. So hourly thinking never worked for us internally either. We always wanted to become faster by using our tools and our heads better, and hourly billing would have punished exactly that.
Pricing the result doesn’t mean we keep quiet about how we work. During this transition, we want to be open about where AI supports us in our daily work:
AI also brings a new kind of tooling cost. The hourly rate has always covered the licenses for all the tools we use. Most of them were seat-based, with one license per person and a fixed monthly fee, as with Microsoft 365 or Google Workspace. AI is moving this toward usage-based pricing, where the cost depends on how much a model works on a project. That is new for agencies, because tool costs can now be linked to a specific project instead of disappearing into a rate. Where these costs are significant, we show them instead of hiding them in the hourly rate. They are either included in a fixed price with a stated usage band, or passed through with a stated margin. We expect usage-based pricing to become an even larger part of our tool stack.
What we tell our clients is that the price reflects the result and our responsibility for it. We are open about how we work so they can see we have nothing to hide.
For us, pricing the result and being open about how we work go hand in hand.
Some clients don’t want their code or content to go anywhere near a cloud AI model, whether because of data protection, internal policies, or simply principle. We respect that and offer it as a choice:
Option | How we work | What it means for the client |
| Cloud AI | Business-tier AI tools with a data processing agreement | Fastest delivery, lowest price |
| Local AI | Open-weight models running on our own or the client’s infrastructure | No data leaves the controlled environment; somewhat slower and currently pricier |
| No AI | Classic craftsmanship, as we’ve done for 20+ years | Different calculation and rate; longer timelines |
The result may be the same in all three options, but the constraints, the effort, and the delivery cost differ, and our calculation reflects that. Choosing not to use AI is a legitimate decision, but it comes at a cost.
Many clients, especially in the public sector, are bound by procurement rules that ask for hourly rates. At first this looks like a dead end, but the rules are often more flexible than the habits that have formed around them.
In Germany, the EVB-IT Erstellung contract can structure custom software development as a Werkvertrag under §§ 631 ff. BGB, which means the contractor owes a result rather than hours.
Procurement law offers a second route. Under § 43(2) UVgO below the EU thresholds, and § 58(2) VgV above them, a contracting authority may set a fixed price or fixed costs and award the contract purely on quality criteria.
Our advice to public clients is to talk to us before the tender is written, and to define results and quality instead of blocks of hours. Where procurement still requires an hourly rate, that rate increasingly has to reflect the expertise, review, and responsibility that go into each hour, not just the time spent producing the work.
Once AI is part of the workflow, it is also part of the data flow. We treat AI providers like any other sub-processor:
The EU AI Act adds another layer. Since 2 August 2026, the transparency obligations under Article 50 have applied to certain AI systems and AI-generated content. We wrote about this in AI Content in TYPO3: Labelling Needs Accountability. These obligations don’t require AI-assisted source code to be labeled in general. Our own rule is simple: AI can generate or propose a change, but a person at b13 signs off on everything we ship and is accountable for it.
The tools change every few months, and our pricing will keep changing with them where possible. The general direction is clear to us, though.
People have been declaring the billable hour dead for decades, and it is still around. It won’t disappear overnight. Some industries will move faster than others, and public procurement will probably take longer because rules and habits change slowly. Some clients, and some agencies, may never switch at all, and we are fine with that. Different models will exist side by side for quite a while.
Our advice to freelancers and agencies is not to wait until clients ask for a discount on your efficiency. Start the conversation yourself, be open about how you work, and price what you deliver.
If you’re a client, ask your agency how they use AI, how they handle your data, and who signs off on the result. You will learn more from those answers than from any hourly rate.
AI hasn’t made our work less valuable. If anything, it has made it easier to see where the value was all along.
The shift isn’t only about pricing. It changes our own work, too: colleagues who loved writing code for years now spend more of their time reviewing, guiding, and taking responsibility for what an agent produces, and that’s an adjustment we’re still making together.
How are you handling this, with your clients and within your own team?
Whether you’re a freelancer, an agency, or a client, we’d like to hear from you.
This blog post offers a practical perspective and does not constitute legal advice.