What is Agentic AI and why should I care?
The concept of Agentic AI (or autonomous agents) represents the next step in the evolution of large language models. While common models, such as standard ChatGPT or Claude, primarily function as advanced text assistants that answer queries, Agentic AI can perform autonomous actions. An agent has tools at its disposal – it can browse the web, work with databases, send emails, or edit documents within defined parameters.
The difference is fundamental: While a chatbot will write you a recipe for a bundt cake, an AI agent can choose a recipe for you, check stock levels in an online store, add items to a cart, and schedule the purchase in your calendar. In the context of city administration, this means an agent can automatically process building permit applications, check compliance with local regulations, and then initiate the approval process in the relevant office.
Boston's Strategy: Fighting "Vendor Lock-in"
The main reason Boston is advocating an open-source approach is to protect against so-called vendor lock-in (dependence on a single supplier). By implementing closed (proprietary) systems from companies like Microsoft, Google, or OpenAI, city governments risk becoming inextricably dependent on their pricing policies, licenses, and technological changes.
According to reports from StateScoop, Boston wants to create an ecosystem where code is publicly available, auditable, and easily customizable. This brings several key advantages:
- Transparency: Citizens and oversight bodies can verify how algorithms make decisions.
- Security: Open code allows the community to find and fix bugs faster.
- Interoperability: Tools can be easily integrated with existing systems of other cities.
Comparison: Open-source Agents vs. Closed Models
When deciding on AI implementation in an organization, a designer faces a choice between commercial giants and open frameworks. The following table summarizes the main differences:
| Feature | Proprietary AI (e.g., GPT-4, Gemini) | Open-source Agentic Frameworks (e.g., CrewAI, LangGraph) |
|---|---|---|
| Transparency | Low (black box) | High (full access to code) |
| Cost | Subscription / Tokens (in USD/EUR) | Free software (infrastructure costs) |
| Customization options | Limited (only via API) | Unlimited (full control over logic) |
| Privacy | Data passes through provider's cloud | Option to run on own servers (On-premise) |
In terms of benchmarks, closed models like GPT-4o still lead in pure intelligence and complex reasoning. However, in the area of specialized tasks and workflow automation, open-source frameworks, which use these models as a "brain" but add a layer of agentic logic on top, are starting to show higher reliability and lower error rates in specific processes (such as document management).
Impact on the Czech Republic and Europe
This initiative from Boston has direct relevance for Czech public administration and the private sector. In Europe, the EU AI Act is coming into force, which is the world's first comprehensive regulation of artificial intelligence. This law places strict demands on the transparency and security of systems classified as "high-risk" (e.g., in education, employment, or infrastructure management).
For Czech municipalities that may start experimenting with AI, Boston's model is very inspiring. The use of open-source tools could help maintain digital sovereignty. Instead of sending sensitive data of Czech citizens to American clouds, it would be possible to operate agentic systems on local infrastructure, which is in line with strict GDPR rules.
Currently, no specific "Czech agentic model" is available, but development teams within the EU are working on local LLMs (Large Language Models) that can be integrated into these agentic structures. For the Czech market, it is crucial that these tools support Czech linguistic specifics (declension, syntax), which open-source solutions allow for much easier fine-tuning than closed systems.
Price and Availability
It is important to realize that "open-source" does not mean "free" in an absolute sense. While the software itself is free (no licensing fees), operating Agentic AI requires:
- Computational power: Costs for GPUs (graphics processing units) in the cloud (e.g., Azure, AWS) or local hardware.
- Development costs: The need to have engineers who can configure and maintain these systems.
- Integration: Costs for connecting agents with existing municipal office databases.
For smaller Czech companies or municipalities, the best start may be to use free tier available APIs (e.g., from OpenAI or Anthropic) in combination with open-source frameworks like CrewAI, which minimizes initial infrastructure investment.
Is it safe to let AI agents work with sensitive office data?
If an open-source approach is used, as in Boston, security is higher because the system can be operated in an isolated environment (on-premise) without sending data to external clouds. However, proper configuration and audit of rules are key.
Can a small Czech office realistically use technology from Boston?
Yes, technologically it is possible. The biggest barrier is not software, but human capacity for management and integration. The recommended approach is to start with simple automations (e.g., email sorting) that do not require complex infrastructure.
Will these agentic tools understand Czech in the future?
Yes, language comprehension ability depends on the underlying LLM (base model). If we use a model capable of Czech (like GPT-4 or Llama 3) as the agent's "brain," the agent will be able to work with Czech text without problems.