The question is not whether, but what

The profession is measurably adopting generative AI, and the productivity gains are real. For practice in BiH the real risk is less about being late than about adopting a tool that confidently guesses.

---

Adoption is happening quietly, but it is happening. In a few years, generative AI has moved into the daily work of legal professionals, most visibly in Europe.

For a lawyer in BiH, the question is no longer whether the wave is real, but how to meet it while the name beneath the answer stays safe.

The shift is already measured

The numbers bear this out. A 2024 Wolters Kluwer survey of legal professionals across the US and nine European countries found that 68% of law-firm respondents use generative AI at least weekly. The report title is itself a question: “Seizing the future or falling behind?”

In the United Kingdom the shift is sharper. According to a LexisNexis survey from September 2024, the share of lawyers using generative AI for work rose from 11% in July 2023 to 41% a year later, while those with no plans to adopt fell from 61% to 15%.

What adoption buys

What adoption buys is time. In the 2024 Thomson Reuters Institute report “Future of Professionals”, professionals project that generative AI will save them about four hours a week within the coming year, and up to twelve hours a week within five years. These are their own projections, not a measured result.

A controlled experiment gives a firmer picture. In a 2023 Harvard and BCG study (“Navigating the Jagged Technological Frontier”), consultants using GPT-4 completed 12.2% more tasks, worked 25.1% faster, and produced work rated about 40% higher in quality. The subjects were consultants, not lawyers; this is evidence about generative AI and knowledge work generally.

The trap in adopting fast

The same study carries a second half that is cited less often. On tasks outside the model's competence, the consultants who leaned on it did about 19 percentage points worse. The authors call this the “jagged frontier”: within the tool's reach the gain is large, just outside it the loss is real, and the line is not obvious.

In law, that boundary is most dangerous where the tool looks most confident. A 2024 Stanford RegLab study found that even purpose-built legal-research tools, despite “hallucination-free” marketing, still hallucinate on roughly one in six to one in three queries: more than 17% for one, more than 34% for another. A hallucination here means a wrong statement of the law, or a correct one citing a source that does not support it.

Not whether, but what

From this, the question inverts. Adoption inside a tool's real competence is a large, measured gain; pushing a confidently guessing tool past it is a measured loss. So the question is not only whether to adopt, but what: a tool that shows its work and knows its limits, or one that guesses.

In Bosnia and Herzegovina that choice carries added weight. The law runs across several levels: state, entity, cantonal, Brčko. Layers that overlap, where a cantonal provision can quietly derogate one you were sure was in force. A confidently guessing tool does not save time here; it moves risk onto the name beneath the answer.

Amiro is built for the other side of that choice. It answers only from the actual text of BiH law, and every answer opens the exact provision it came from, yours to check before you rely on it. When there is no grounded answer, it says so, rather than offer a plausible guess. No citation, no answer.

Adoption as caution, not courage

The message is not to hurry. The profession is already moving, and the direction is measured. But there are two ways to fall behind: by waiting, and by adopting a tool that confidently guesses.

The first delay is at least visible. The second surfaces only once a name already stands beneath a wrong answer. So the measure is not how fast a tool answers, but whether it shows its source and admits when it has none. That is a choice you can check yourself, and one worth making calmly.