Experience with Building AI Agents: Tech Companies vs. Beginners

The authors of this article, Filip Tichý (Partner at Grant Thornton Slovakia) and Jakub Chudík (Co-Founder at Assetario), take you through the world of artificial intelligence in the AI Breakfast series. This article was written without the use of AI.

Although the authors of this article, Jakub and Filip, are both big AI enthusiasts, their abilities to build AI agents are not the same at all. The process of building AI agents can look fundamentally different depending on whether it’s carried out by a tech company with an in-house AI team (Jakub) or a “regular” company that doesn’t yet have a dedicated AI department and is just getting started with serious AI (Filip). In the first case, it typically involves the systematic development of multiple agents in parallel, built on a data infrastructure, development tools, and an experienced team. In the second case, it’s more about learning through the first specific use case, where methodology, perseverance, and a good understanding of the business process are just as important as the technology itself. This year, Filip’s audit team developed its first in-house agent, which gave us a good opportunity to compare agent development between a tech AI company and beginners.

Comparison of the Approaches of a Tech Company and a “Regular” Company

What the Comparison Reveals

A technology company has a natural advantage, that it already has a development environment, data infrastructure, and people who are accustomed to working with code, data, and iterative development. Building AI agents is therefore a more scalable process: the company can develop multiple agents in parallel, continuously refine them, and integrate them with existing systems.

A regular company without an in-house AI department starts from a different position. The first agent is not just a technology project but also a learning process. The team must adopt a mindset focused on automation, properly define workflows, understand the limitations of the tools, and learn to systematically test and correct outputs. This is precisely why the overall development time may be longer, even though the tools themselves seem accessible and simple at first glance.

The biggest difference becomes apparent during the tuning phase. In a tech company, this is an expected part of the development cycle. In a regular company, tuning can be surprisingly challenging, because fixing an agent isn’t “vibe-coding,” but rather methodical, detailed, and often time-consuming work. It requires patience, precise feedback, documentation of errors, and ongoing adjustments to the solution’s logic.

Key Takeaways for Beginners

1. Beginners definitely can build agents in-house, as long as you have:

• curious and persistent people (tuning and debugging agents isn’t as fun as vibe-coding, but rather methodical and time-consuming work)
• project management and a structure for the entire process (including milestones, clear roles, etc.)
• provided you’re on the Cloud. If your company is still running on a legacy on-premise environment,  implementing AI agents is significantly more complicated. If you have both your user data and applications on the Cloud, implementing AI agents into your business processes will be almost natural.

2. It is essential to invest more time and attention into planning, defining the framework, and establishing the workflow. Right from the start, consider all implications, theoretical situations, and scenarios. This will save time spent correcting and reworking a poorly designed concept. Involving people from the relevant business process is absolutely essential.
3. Start as soon as possible; you’ll test it out on the first agent and learn as you go. Don’t be afraid to seek outside help—even limited assistance can be sufficient. You’ll be surprised at just how much your colleagues can handle.