ATAILA Newsroom · Budapest · 2026-08-16
The news is not the 11 months. It is the 30 hours.
The Hungarian news site index.hu reports a practical case from Hungary: AI-directed agent teams took on work in two businesses that would have required months under traditional work organisation. We do not dispute the numbers — the pattern is familiar from our own deliveries. But the news is not where most people look for it. The interesting figure is not the eleven months; it is the thirty hours that replaced them.
The article we are responding to
„Tizenegy hónapnyi munkát váltott ki 30 óra alatt a mesterséges intelligencia”
index.hu · 2026-08-12
What index.hu reports
According to the article, Tamás Tóth, lead expert at AI Workshop, put agent teams directed by artificial intelligence to work in two of his businesses. The figures as published:
The tasks covered four areas: software modernisation, marketing, administration and reporting. The article also names the conditions for success — according to Tóth, using AI effectively requires “precise task definition, appropriate company rules and keeping data security in mind”. The report is based on a statement by Cube-Élmény Kft.
We agree with almost all of it
What the article describes matches our own experience: where a company has a few well-defined, repetitive, expensive workflows, a well-configured agent team today really does get through them orders of magnitude faster than a traditionally organised team. That is no longer a promise; it is practice.
The agreement ends at the next sentence: the thirty hours are not part of the good news. The thirty hours are the bad news.
Because those are no ordinary thirty hours. They belong to an expert who knows how to decompose a business goal into agents, what permissions an agent gets, where the data boundary runs, what may be copied out of the company and what may not — and how you know the result is correct. There are few such people in Hungary, and you cannot manufacture more of them just because more companies want agents.
It is worth looking at the three success conditions the article names, because none of them is an AI problem:
- Precise task definition — that is specification and architecture.
- Appropriate company rules — that is governance, approvals, audit logs.
- Data security — that is infrastructure and a data boundary.
All three are platform questions. The agents solve the easier half of the problem. The harder half is that these three have to be rebuilt for every single project — by hand, by an expert, in thirty hours. We call this the provisioning gap: the model is ready, the operable system around it is not. Our earlier response was about the same gap, with different data.
What ATAILA does with the 30 hours
Our goal is not to dispute the result. Our goal is that the thirty hours stop being an expert’s privilege and become the platform’s default.
The article's finding
AI needs precise task definition.
Our answer
ATAILA Factory makes exactly that description machine-readable. You describe once, in a short manifest, what you want — and the factory builds the repository, the secrets, the DNS, the machines, the ingress routes and the monitoring from it. Task definition stops being knowledge living in one expert’s head and becomes a versioned, re-runnable declaration. That is the difference between someone setting it up once and the company being able to repeat it tomorrow.
ATAILA Factory →The article's finding
Appropriate company rules are required.
Our answer
In the Release Manager every component travels its own path, but through the same four stages — SANDBOX, DEV, UAT, PROD — and writing to production is gated on owner approval. The approvals, the separation of environments and the access logs are themselves the compliance evidence. In our own words: compliance here is not an audit; it is how delivery operates.
The article's finding
Data security must be kept in mind.
Our answer
ATAILA Cloud is a sovereign, EU-resident cloud, sized for production loads, with an identifiable operator. The private AI runs on our hardware, not behind an American API. And releasing a build never touches your data: copying data between environments is a separate, approved operation. When an agent reads contracts, patient records or financial material, this is not a detail — it is the decision itself.
ATAILA Cloud →What the article does not list among the success conditions: the thirty hours themselves were expert work. That is the expert seat ATAILA Studio delivers — a thin client on your desk, a secure tunnel to the platform, a managed cloud workstation and a private agentic AI coding fleet on EU GPUs. Plug it in, build; everything else is our job. We do not rent out GPUs and we do not hand over a project: we deliver the running system, operated end to end.
This is not theoretical: our customer stories include a certification body that spent two years and wrote off roughly HUF 20 million on two failed builds — first a traditional integrator, then a low-code platform — before getting a working, modern system on ATAILA in about two months, with four real environments: SANDBOX, DEV, UAT and PROD. At the customer’s request we publish the reference anonymously. The pattern is the same as in the article: what was missing was not intent, but the base layer to stand on.
What we do not claim
We do not claim eleven months of work disappears at every company. The numbers in the article are about Tamás Tóth’s two businesses, in the hands of an expert who knows what he is doing. It is a valid result, but not an industry benchmark — and we will not use it as one.
Nor do we claim agents replace professional judgement. If a workflow is not worth automating, or the data is not actually there for it, we say so in the first conversation, not after the second invoice. The platform cuts the time and risk of the setup — it does not decide for you what is worth entrusting to an agent in the first place.
Finally: we do not claim our path is zero effort. We claim the effort is spent once, as a declaration, and is repeatable afterwards — instead of costing thirty scarce expert hours again for every new application. Your code and your data stay yours throughout, exit route included.
Let’s look at your thirty hours
If you have that one confidential workflow you have long wanted to hand to agents — the one you would never paste into public AI — tell us about it. We will work out together what your thirty hours amount to, and how much of them the platform does for you. If the answer is that it is not worth it, we will say that too.
Source: Tizenegy hónapnyi munkát váltott ki 30 óra alatt a mesterséges intelligencia — index.hu, 2026-08-12
← Back to the Newsroom Press inquiries: contact us