How I AI

A harness is code that wraps around an AI agent to make it more effective. The speaker demonstrates how a harness can automatically investigate production issues by analyzing Sentry warnings, identifying root causes, determining impact, and recommending whether to create tickets or apply fixes.

The speaker discusses the shift from traditional agent prompting to agent management, highlighting the limitations of local Kanban board approaches and advocating for cloud-based VPS solutions with multiple communication channels to effectively manage autonomous AI agents across projects.

A custom AI harness is code wrapped around an AI agent to make it more effective for specific workflows. The speaker demonstrates building a Sentry bug-debugging harness using Claude SDK with a terminal UI, showing how structured constraints, custom prompts, and specific tool adapters enable agents to handle complex tasks more efficiently than general-purpose AI tools.

Alessio Finelli demonstrates how he uses OpenAI's Symphony framework combined with Linear and Codex to run autonomous coding agents from his phone, managing both software engineering tasks and a Pokémon card trading business through cloud-based VPS infrastructure instead of local machines.

The speaker discusses discrepancies between their subjective evaluation of model performance and automated benchmark results, noting that different judges (Opus, 4A, 5.5) show varying levels of generosity and bias. They conclude that personal judgment matters significantly and plan to incorporate more subjective taste into evaluation metrics while retiring saturated benchmark tasks.

The speaker provides task-specific model recommendations across different use cases, suggesting GPT 5.5 for PRDs, Sonnet 4.6 for prototyping and casual interaction, and Opus 4.8 or Sonnet 5 for codebase work. Model selection varies based on complexity, with Opus 4.8 excelling at dense UI design and Sonnet suitable for simpler implementations.

The speaker introduces 'The How I AI Bench,' a new set of human and AI-graded benchmarks designed to evaluate language models on practical tasks like writing PRDs, solving bugs, and designing systems. They test Claude Sonnet 3.5 against these benchmarks and note that while it scores lower than some specialized benchmarks (69% on Agentic Coding SweetBench Pro, 82% on Terminal Bench 2.1), the difference may not be noticeable in real-world usage.

Katie transitioned from designer to top-performing engineer, ranking in the 94th percentile for code throughput across the entire R&D organization. Her success stemmed from technical curiosity combined with supportive engineer mentors who reviewed her code and helped her improve her craft.

A team successfully shipped a project in 10 weeks by eliminating traditional project management structures entirely—no meetings, Jira, documentation, or text communication. Instead, they relied solely on a 24/7 Zoom room where team members could work synchronously and asynchronously as needed.

The speaker discusses how Claude's Co-worker feature helps parents automate tedious online administrative tasks, freeing up time for more meaningful interactions with their children. By handling tasks like returns and help emails, AI removes low-value busywork rather than replacing genuine human experiences.

A parent describes using Claude as a household management and shopping assistant to find high-quality, naturally-made products from reputable brands. They created a project in Claude with specific brand criteria and used it to organize notes and vet brands. A key benefit highlighted was Claude surfacing that a previously reputable brand had declined in quality after a corporate takeover.

Nicole Ruiz demonstrates how she built a Claude project to automate high-quality shopping decisions for her family, using curated brand lists and purchasing criteria to filter out cheap, poorly-made products. She also shows how Claude Computer Use helps her draft return emails by pulling order details directly from her Gmail. The system is designed to reduce the mental overhead of conscious consumption so she can spend more time with her children.

The host demonstrates using Google Flow, a generative AI creative suite, to create a hype video for the 'How I AI' podcast. Using a fish-eye lens avatar of themselves, they generate a storyboard with cinematic shots including keyboard close-ups, office wide shots, and a humorous chair spin with a digital heads-up display overlay.

A person recounts shipping an app to the app store without learning to code, relying primarily on basic computer skills like copying, pasting, and file labeling. She candidly admits she still doesn't understand the underlying software infrastructure, including the deployment platform Railway, even after successfully completing the project.

A creator demonstrates using Google Omni to generate a hype/promo video for their podcast 'How I AI' in under 15 minutes. The resulting video includes music, narration, and a polished promotional script. The creator expresses genuine amazement at the output quality.

The speaker describes a workflow for creating anthropomorphic animal exercise videos using two AI tools. Animals are generated in Gemini, while Higgsfield is used to merge those still images with self-filmed exercise footage. The result is an animated anthropomorphic animal appearing to perform the exercises.

Claire Ho from the 'How I AI' podcast demonstrates creating an AI video avatar of herself using Google Flow and the Gemini Omni model in approximately 15 minutes. She generates a full storyboard, produces multiple AI video scenes using her avatar, and stitches them into a one-minute hype video for her podcast. Despite imperfections like inconsistent backgrounds and uncanny valley moments, she considers the result a success given the minimal time and effort involved.

A developer with a beginner's mindset used Claude to guide them through preparing a Replit app for App Store submission over a single weekend. They developed a workflow using two AI tools in tandem — Claude for strategy and Claude Code for writing code — ultimately succeeding on their second App Store submission attempt.

Bryce Ratner Keithley, a non-technical talent and recruiting professional, built and launched a fitness app called Daily Hundreds to the Apple App Store using AI vibe coding tools including Replit, Claude, and Lovable — with zero coding knowledge. The app features AI-generated anthropomorphic animal exercise videos created by combining Gemini-generated images with real workout footage processed through Higgsfield's Cling model. Her journey illustrates how beginner's mindset and AI tools are enabling non-technical people to build production-ready software.

This transcript explains when not to use the /goal command in Codex, outlining three scenarios where it is inappropriate. It also defines the three key properties that make a goal well-suited for the tool: a durable objective, an evidence-based finish line, and a multi-turn path.