DSLs Enable Reliable Use of LLMs

2026-07-14 · Source: Martin Fowler · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, long

Summary

Domain-Specific Languages (DSLs) play a critical role in making Large Language Models (LLMs) more reliable for code generation. Published on 14 July 2026, the article explains that while LLMs generate code rapidly, they require clear boundaries to ensure precise output. DSLs provide this harness by offering constrained syntax, reducing output variation, and enabling deterministic validation. Examples like Tickloom, a semantic model and DSL for distributed systems, demonstrate how LLMs can act as both design partners for developing abstractions and natural language interfaces for established DSLs. This approach allows LLMs to generate verifiable code, with the generated DSL code itself becoming the primary, maintainable source of truth, rather than the initial prompt.

Key takeaway

For AI Engineers and Software Architects building reliable systems with LLM-generated code, relying solely on high-level prompts is insufficient; instead, invest in developing Domain-Specific Languages (DSLs) or robust semantic models. This approach ensures LLMs generate precise, verifiable code, reducing hallucination and simplifying maintenance. You should treat DSLs as the primary source of truth, leveraging LLMs as design partners and natural-language interfaces to these constrained systems.

Key insights

DSLs provide LLMs with constrained syntax and validation, making code generation reliable and maintainable.

Principles

• DSLs constrain LLM output, boosting reliability.
• Design emerges iteratively, LLMs accelerate discovery.
• DSL-generated code serves as the primary source of truth.

Method

LLMs operate in two phases: first as a brainstorming partner for designing abstractions/DSLs, then as a natural-language interface to the established DSL for reliable generation.

In practice

• Use PlantUML or SQL with LLMs for reliable generation.
• Build internal DSLs for complex system testing scenarios.
• Employ semantic models to ground LLM protocol logic.

Topics

• Domain-Specific Languages
• Large Language Models
• Code Generation
• Distributed Systems Testing
• Semantic Models
• Software Reliability

Code references

• unmeshjoshi/MADSPlantUMLSteps
• unmeshjoshi/tickloom

Best for: AI Engineer, Software Engineer, AI Architect

Related on AIssential

• A Roadmap for Tamed Interactions with Large Language Models — Software Engineering principles can formalize LLM interactions, ensuring reliable and structured outputs through a dedicated scripting language.
• Can LLMs Write Correct TLA+ Specifications? Evaluating Natural-Language-to-TLA+ Generation — LLMs struggle significantly with semantic correctness in TLA+ specification generation, requiring expert oversight.
• Planning in the LLM Era: Building for Reliability and Efficiency — LLMs are shifting from direct plan generation to construction-time generation of verifiable, efficient symbolic planners.
• Building AI with AI — LLMs should be used to build systems, not as systems, reinforcing code's critical role.
• 4 Things Everyone Must Know About LLMs and Agentic AI — Effective LLM and agentic AI development prioritizes model selection, task simplification, context management, and structural prompting over complex prompts.
• An Empirical Study of LLM-Generated Specifications for VeriFast — LLMs can generate specifications for separation logic verifiers, but their domain-specific knowledge limitations lead to modest verification success.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Martin Fowler.