Observed Signal · Jun 6, 2026 · Interview · Source: DEV Community · Impact: 2/5 · Sentiment: Neutral
Rasa Podcast: Dialogue Design in the LLM Era
A Rasa podcast episode of The Dialogue Architects features Rebecca Evanhoe (co‑author of Conversations with Things, Product Manager and Head of Conversation Design at Slang AI) discussing how conversational design has evolved with large language models (LLMs). Topics include Slang AI’s voice AI for restaurant phone bookings, a hybrid deterministic-vs-generative system design, strategies to reduce hallucinations (strict prompts and selective use of rules/APIs), new evaluation workflows (human-labeled ground truth, LLM-as-a-judge, and a trained evaluation model), operational constraints like ASR errors and latency in voice, and how conversational telemetry can produce prescriptive business insights for restaurants. Evanhoe argues that core conversation-design skills remain essential despite shifts toward prompt/context engineering.
Practical insights on deploying LLMs and voice AI at scale (hybrid architectures, evaluation methods, latency/ASR constraints) are useful to conversational-AI and voice-product practitioners but do not constitute a major platform policy or industry-shifting technical release.
Track OpenTable Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Rebecca Evanhoe is Product Manager and Head of Conversation Design at Slang AI and co-author of Conversations with Things.
- Slang AI builds voice AI for restaurant phone systems to handle bookings, modifications/cancellations, and restaurant questions.
- Slang AI processes up to 2 million calls per month and favors starting with the cheapest, fastest, smallest models.
- Slang AI uses a hybrid architecture: deterministic rules for business logic and API checks (e.g., OpenTable integrations) and generative LLMs (with strict prompts) for menu and contextual answers to reduce hallucinations.
- Evaluation pipeline includes transcript review, a Brain Trust of three human raters to build ground truth, and training an "evaluation" LLM (LLM-as-a-judge) to automate large-scale scoring.
Connected Companies & Entities
1 Entity mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Using LLMs for Dialogue Management
The article explores practical patterns and architecture choices for using large language models (LLMs) as dialogue managers. It contrasts classical modular dialogue systems with LLM-based approaches that can reason over full transcripts and emit structured actions. Four production patterns are described: end-to-end generation, structured state extraction, tool-augmented manager, and hybrid classifier-LLM. The post gives prompt-engineering recommendations (system prompt as spec, JSON outputs, compressed memory), context/window management strategies (summarization, sliding window, external memory), and a code example using the OpenAI Python SDK pointed at Oxlo.ai with function-calling (model: llama-3.3-70b) to implement a tool-augmented e-commerce support flow. It also notes Oxlo.ai’s request-based pricing keeps per-turn cost flat regardless of prompt length. Publication date: 2026-06-17.
Designing AI Products for Context Management
The article argues that failures in large language model (LLM) outputs are often due to missing or poorly managed context rather than model capability. It describes a shift from prompt engineering to context design, where systems must store, scope, select, and update relevant context across interactions. The piece identifies three emerging design patterns implemented across major AI chat products: context containers (persistent project/notebook scopes), selective referencing (choosing which sources to include), and instructions (project- or system-level behavioral guidance). Examples cited include ChatGPT Projects, Claude Projects, Gemini NotebookLM, Copilot Notebooks, NotebookLM checkboxes, and Claude connectors. The author emphasizes that context must be curated and maintained over time, and that product and UX design play a central role in enabling more reliable, valuable LLM-driven workflows.
LLM Planning, Agent Debates, and Persistent AI Worlds
A DEV Community roundup (Apr 25, 2026) highlights emerging trends in large language model (LLM) usage and agentic AI. Data shows a decline in so-called "Both Bad" LLM responses though quality gaps remain. Practitioners advocate modular, incremental planning for LLM systems (Matt Pocock). Research and projects described include AI agents that debate to improve decisions, Vorim.ai building an identity and trust layer for agents, and Outerloop.ai creating persistent virtual worlds where agents and humans coexist. The post also raises security concerns by discussing Anthropic’s Claude Mythos in the context of potential AI-native cyberweaponry. The piece frames these developments as practical, operational shifts—emphasizing developer approaches, trust/identity infrastructure, and cybersecurity implications for long‑running agent deployments.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.