{"type":"rich","version":"1.0","provider_name":"Transistor","provider_url":"https://transistor.fm","author_name":"Code and Cognition","title":"Agents, Workflows, and LLM Productivity in 2025","html":"","width":"100%","height":180,"duration":2276,"description":"Show NotesIn this episode, we cover:00:00 – Agents vs. Workflows\nWhat’s new in LLM-powered agents, and how they differ from traditional, on-rails workflows.01:30 – Pattern Libraries and Agent Frameworks\nExploring tools like Amazon Bedrock and visualizing agentic workflows.03:00 – LLM as Judge\nUsing higher-cost models to validate output from cheaper models. A clever way to balance cost and quality.04:30 – Real-Time Evaluation and Prompt Engineering\nHow teams can move from static success criteria to live evaluation systems.05:45 – System Prompt Design\nLessons learned from designing effective system prompts and using them the right way in production.08:00 – Tool Use and LLM Decision Making\nTeaching LLMs to use internal tools to answer queries—think calendar lookups, database queries, and more.11:30 – End-to-End Testing with Natural Language\nA new generation of testing libraries using LLMs and Playwright to turn plain English into functional tests.12:45 – Linear vs. Branching Workflows in Data Pilot\nHow AI can define and adapt its own steps in data analysis and synthesis.15:00 – Multi-Source LLM Querying\nMerging inputs from time tracking, Slack, GitHub, Jira, and more to create richer outputs.18:30 – Human-in-the-Loop Patterns\nDesigning workflows where humans review and approve AI-generated outputs before final delivery.20:30 – Cursor and AI-Accelerated Development\nHow Olio engineers use Cursor for faster iteration while managing tech debt and consistency.23:00 – Heuristics for Managing AI-Created Code\nTips for writing better system prompts, rules files, and maintaining consistent output across codebases.27:00 – Infrastructure Throttling and LLM Scaling Challenges\nBehind-the-scenes look at orchestrating LLM queries at scale, managing Lambda concurrency, and avoiding AWS throttles.30:00 – LLM Identity & Masquerading Challenges\nWhat it takes to make your AI chatbot actually believe it’s your brand—and how Claude and GPT-4 stack up.33:00 – Looking Ahead to 2025\nThe team...","thumbnail_url":"https://img.transistorcdn.com/FibZYR2Jc4bAsvUYB_faLieYy39Memr1SvaXdzAp7Qc/rs:fill:0:0:1/w:400/h:400/q:60/mb:500000/aHR0cHM6Ly9pbWct/dXBsb2FkLXByb2R1/Y3Rpb24udHJhbnNp/c3Rvci5mbS82ODFl/ZTE4MWM5YjQyMzM2/MzhkNjViNzUxYmFl/YmIyYi5qcGc.webp","thumbnail_width":300,"thumbnail_height":300}