- In 2026 new AI agencies are flooding the market — training and certification waves plus no-code tools have drastically lowered the barrier to entry. A certificate says little about delivery depth.
- Templates are not the problem. For standard tasks they are the fastest route. The problem is squeezing every task into the same template — without saying so.
- Seven signals give the template seller away: from the demo that is ready before the first conversation to the solution that lives in the agency’s account instead of yours.
- The strongest screening question: “What do you want to know about our business before you write a proposal?” Whoever asks nothing is selling something that existed before your problem did.
First, so this piece is read in the right light: we are an AI agency ourselves, so we are a party to this. What follows is not a study but distilled project experience — including conversations with businesses that had already been through another agency before us. Those conversations follow a pattern. And you can use that pattern before you spend money.
Why there are suddenly so many AI agencies in 2026
Two developments have converged. First, no-code tools have massively lowered the technical barrier: anyone who can click a workflow together can now offer “AI automation”. Second, a whole training market has grown up around the AI boom — courses, certificates, agency coaching — continuously producing new providers who start with the same playbook: same positioning, same demos, same templates.
In itself, that is neutral. More providers mean more choice and pressure on prices. But it also means the signals people used to read competence from — a professional website, a confident pitch, fluent jargon — are now part of the starter kit. Mastering the buzzwords is a long way from having delivered a project. So you need better screening criteria than first impressions.
Template or custom work — both have their place
One thing this piece explicitly does not say: that templates are bad. For clearly bounded standard tasks — appointment booking, simple enquiry triage, recurring text blocks — a proven template is often the right call: live faster, cheaper, battle-tested. If your problem is a standard problem, custom work would be a waste of money.
The problem starts where an agency only has templates — and therefore bends every problem until it fits one. Then the metal fabricator gets the same automation as the tax firm, your business’s edge cases fall off the table, and three months in a human is once again reworking the cases “the system can’t do”. The distinction that matters is not template versus custom — it is whether someone honestly tells you which one your case is.
Seven signals you are only buying a template
- 1. The demo is ready before anyone has talked about your business. What you are being shown was shown to three others yesterday. A demo is not proof of ability — it is sales material.
- 2. The proposal lists tool names instead of outcomes. “Chatbot”, “workflow automation”, “AI agent” — but nowhere does it say which process will run how much faster afterwards, or how you will measure it.
- 3. Nobody asks about your data, systems and edge cases. Whoever doesn’t want to know what your invoices look like, which special cases exist and which system the result must land in isn’t planning your solution — they are planning the rollout of their template.
- 4. Data protection only comes up when you raise it. Where your data is processed is not a formality with AI — it is an architecture decision. If GDPR and data residency don’t come from the agency unprompted, that is a warning sign.
- 5. Everything is clicked together, nothing is built. No-code is legitimate — until the first requirement the kit doesn’t cover. Ask directly: “What do you do when there is no ready-made block for it?” Whoever dodges that cannot build.
- 6. A retainer with no defined outcome. A monthly fee for “support and optimisation” with no measurable deliverable is a subscription to hope. Ongoing support is sensible — with defined content.
- 7. The solution lives in the agency’s account. Workflows, prompts and access sit with the agency, documentation doesn’t exist. If you part ways, you start from zero. That is not ownership, that is dependency.
No single signal is a deal-breaker on its own. Two or three together are a pattern.
The questions to ask every AI agency
You don’t need to be an AI expert to test for substance. Five questions are enough — and the reaction often says more than the answer:
| Question | What a good answer looks like |
|---|---|
| “What do you want to know about our business before you write a proposal?” | Concrete questions about process, data, systems, edge cases — no instant proposal. |
| “Tell us about a project that didn’t work.” | An honest story with cause and consequence. Whoever has never failed hasn’t delivered much. |
| “What happens to cases the AI can’t decide confidently?” | A clear human review step (human-in-the-loop) — not “that practically never happens”. |
| “Where does our data run, and who owns the solution in the end?” | Precise answers on processing location and EU options; handover into your accounts incl. documentation. |
| “How do we measure after three months whether it paid off?” | A metric from your business (hours, cycle time, error rate) — not “you’ll notice”. |
Screening questions: Digital Maker — the reaction often counts more than the answer
How to recognise a real partner
The positive picture, conversely, is simple to describe. A partner who means it starts with your process, not with their toolbox. They will sometimes tell you an automation is not worth it — and would rather start with a small, measurable first step than the grand rollout. They treat data protection as an architecture question, plan for edge cases from day one, and build so the solution belongs to you and stays maintainable. What that looks like in practice is described in how we work and, for a concrete pattern, in agentic workflows explained — and why so many mid-market AI projects never reach production is shown in our analysis of the DIHK and Bitkom figures.
And yes: sometimes the honest recommendation is a template. If your problem is a standard problem, your agency should say so — and not charge you for custom work. Honesty in both directions is the real quality marker. Feel free to put the five screening questions to us directly.
Sources and context
This piece is deliberately not a market study but a view from practice: it is based on our project experience as an AI agency and on conversations with mid-sized businesses that had previously worked with other providers. We are a provider ourselves and therefore a party — but the signals and screening questions are deliberately phrased so you can apply them to any provider, ourselves included. No specific companies are assessed.
FAQ: choosing an AI agency
How do I recognise a good AI agency?
By what happens before the proposal. A good agency first wants to understand your process, your data and your edge cases — and its proposal describes a measurable outcome, not a list of tools. It also raises data protection and data residency on its own, has a plan for error cases (human-in-the-loop), and clearly settles who owns the solution after the project.
Are template solutions from AI agencies bad by definition?
No. For standard problems — appointment booking, simple enquiry triage, recurring text blocks — a proven template is often the fastest and cheapest route. It becomes a problem when an agency squeezes every task into the same template without saying so. The real distinction is not template versus custom work, but honesty about which one your case needs.
What should an AI agency want to know before making an offer?
Your actual process (who does what today, how long it takes), your systems (ERP, CRM, DMS, mailboxes), your data situation (where it lives, how sensitive it is) and your edge cases. Anyone who writes a proposal without asking is selling a solution that existed before your problem did.
Who should own the AI solution after the project?
You. Workflows, prompts, configuration and access should live in your accounts or your infrastructure, or be handed over cleanly — including documentation. If the solution lives solely in the agency’s account, you don’t own an asset, you hold a subscription with termination risk. Settle this contractually before the project starts.
Why are there suddenly so many AI agencies in 2026?
Two reasons: no-code tools have drastically lowered the barrier to entry, and a whole training and certification market keeps producing new providers who start with the same playbook. That is not bad in itself — but it means a certificate or a polished website says little about actual delivery depth. Which makes the right screening questions all the more important.
Ask us the five screening questions — we’ll gladly answer them.
In a discovery call we look at one concrete process from your business and tell you honestly whether automation is worth it — and whether your case needs a template or real work. Four eyes, thirty minutes, no slides.