Merge Conflict

534: Why Developers Are Obsessed with Budget LLMs (Jev, Luna, & more)

Today at 7:00 AM

Discover the emerging classifier revolution with Jev—a radical new take on LLMs that ditches generation for lightning-fast categorization. Plus, deep dives into the latest model releases (Luna, GPT-6 Sol, Claude Opus 5.5), practical strategies for iterative AI development, and why specialized models are reshaping how we work with AI.

Apple's new iPhone Duo promises a game-changing foldable experience, but developers face a minefield of technical challenges. James and Frank break down the critical app adaptation work required—from handling multiple screen permutations to managing safe area insets and keyboard behavior—plus why native UI controls matter more than ever.

In this week's episode, James and Frank dive deep into how their approach to working with AI models has fundamentally shifted over the past few months. We explore a game-changing realization: you don't always need the most expensive, powerful model for every task. Instead, they share a strategic two-part workflow—using high-reasoning models for deep research and planning (generating comprehensive documentation and analysis), then switching to tiny, blazingly-fast models like Baby Luna or MAI Code 1.1 Flash for implementation, cutting costs by 90-95% while maintaining nearly identical results. But the real breakthrough is HydraFusion, GitHub's new multi-model orchestration runtime that au...

531: How AI Saved Our Lost Podcast Recording

09/07/2026

In Episode 531 James and Frank recount a near‑disastrous data loss when local recordings died seven minutes in, forcing them to rely on Zencastr's cloud-recovery button and a hopeful rescue email. They walk through the panic and the clever fix—using GitHub Copilot CLI/Baby Luna, FFmpeg and AI-assisted workflows to rebuild a muxed video, repair audio‑video drift, and underscore lessons about testing restores and trusting backups.

This episode kicks off with WinUI’s shift to open‑source‑first—what the GitHub‑centric workflow means for Windows devs—and James and Frank’s hands‑on tales modernizing apps, CI/CD, and the surprisingly smooth Microsoft Store pipeline. They then deep‑dive into Apple: M6 Mac mini and Mac Studio upgrades for local AI, Rosetta’s sunset and Intel‑Mac worries, foldable iPhone rumors, and possible Apple TV Thread/HDMI features—practical insights for choosing platforms and tooling.

Note: We had issues with local recordings, so this is a cloud backup processed with Adobe Podcasts

In this update-packed episode James and Frank dive into developer tooling: the VS Code Mobile Canvas that embeds emulators into your editor, MAUI DevFlow’s rewrite enabling plain .NET/native (and desktop AppKit) support, and the growing MCP/agent ecosystem that ties editors, the Copilot app and CLI together. They share practical takeaways—Mobile Canvas works across frameworks, DevFlow supports non‑MAUI apps, agent instruction files matter, and you can automate Microsoft Store publishing—giving cross‑platform builders clear next steps.

In episode 528 James and Frank riff on software updates, AI-powered documentation and rapid app-building—covering a WebAssembly Oh My Posh demo, a LocalMorph/FFmpeg bridge, and MAUI-native builds—showing how AI can speed onboarding, re-theme UIs and generate CI/CD pipelines. They close with a lively debate on distribution—Homebrew vs app stores—sharing practical guidance on when to use package managers versus polished consumer releases, plus a playful detour into Frank’s FIRE calculators.

AI agents are revolutionizing mobile development! James and Frank explore GitHub's new Mobile Canvas extension that lets AI agents see and interact with iOS and Android apps in real-time, no matter what framework you're using. They discuss how this tool finally closes the feedback loop for native developers and break down the emerging plugin standard reshaping AI-assisted coding.

Link: https://github.com/Redth/mobile-canvas-ghcp

In this episode, James and Frank dive deep into GitHub's biggest release in years: Stacked Pull Requests. Learn how this long-awaited feature finally solves the challenge of managing massive, complex code changes by breaking them into smaller, independently reviewable chunks while maintaining dependencies—plus discover how Merge Queues work alongside this innovation to streamline workflows in high-velocity projects.

In this episode, James and Frank explore the bittersweet reality of experiencing cinema in 2026, starting with Frank's disappointing discovery that Seattle's Pacific Science Center—home to a true IMAX theater—has ditched its iconic 70mm film projection for digital laser technology. Despite Christopher Nolan's passionate advocacy for shooting on IMAX film, fewer theaters worldwide can actually deliver his intended experience, raising questions about whether premium cinema experiences are disappearing. The hosts dive deep into what made true IMAX special and the broader trend of losing analog, artisanal technologies in favor of digital convenience.

Beyond cinema, the conversation span...

524: A .NET MAUI AI Revolution

07/20/2026

On episode 524 James and Frank dive into the new .NET MAUI developer stack—covering the MAUI CLI/Maui Doctor that auto-provisions SDKs and emulators, the Maui Sherpa GUI for device/Xcode/provisioning management, and DevFlow's MCP server that lets AI agents inspect, interact with and automatically test apps (closing the loop). They also highlight the VS Code MAUI agent/skills, profiling tools, and the shift to core CLR in .NET 11 with performance tradeoffs to watch.

Frank releases UI.md, a practical markdown guide distilled from 25 years of software development that captures essential UI/UX principles to help AI (and developers) build better interfaces. The hosts dive deep into why users expect consistency, how small details create polish, and why understanding your users' mental models beats flashy design every time—with real-world examples from their own apps.

James and Frank dig into Slate’s modular electric pickup — a $24,950, 205‑mile, no-frills truck you can customize later — and unpack how factory vs dealer options, aftermarket mods and build quality will make or break the idea. They also cover real-world charging (level 1/2/3), fleet and daily-use practicality, and why a DIY, upgradeable EV could reshape affordable vehicle ownership.

Episode 521 James and Frank obsess over “polish”: the tiny design and packaging details that make apps feel finished. They start with impeccable.style, product.md and design.md (and why agents.md and readme aren’t enough), then dig into UI fit‑and‑finish — tray/menu UIs, icon choice, grouping settings and why AI agents still struggle with layout and whitespace. The conversation then moves deep into Windows packaging: WinApp SDK versions, trimming woes with WinRT, ready‑to‑run vs. single‑file self‑contained builds, MSIX tradeoffs, and strange cases where builds bloat with unwanted packages. Key takeaways: give agents the right met...

520: Inside the New GitHub Copilot App: Sessions, Canvases, Automations

06/22/2026

This episode dives into the new GitHub Copilot app — a hub for agent sessions, multi‑repo orchestration, PR‑first workflows, automations and the new canvas metaphor for triage and planning. James and Frank unpack how sessions can spawn and coordinate across repositories, agent‑aware merges that respect CI and code review, support for local models and extensions, and the token tradeoffs to be aware of. The canvas demos (swipe‑to‑triage, Kanban agent assignment) show how Copilot is trying to move beyond chat windows into real UX for developer workflows.

They also tour Windows dev work: embracing WinUI for na...

519: WWDC Deep Dive: Apple’s New AI for Developers

06/15/2026

James and Frank break down WWDC’s developer-focused news—from Apple’s AI push (local models, a new private cloud for foundation models, and a command‑line API) to Xcode’s new agent integrations, the device hub/simulator overhaul, and macOS/UI polish. If you build apps, pay attention: small developers now get subsidized cloud models, Intel Macs are being sunset, and you’ll need to test many form factors and new Swift/AI APIs.

Microsoft Build 2026 brought a major shift: homegrown AI models designed for efficiency and real-world developer workflows. From the cost-effective MAI Code 1 Flash to sandboxed code execution and Windows developer tools, discover how Microsoft is making powerful AI accessible without breaking the bank. James shares his hands-on experience proving that you don't need expensive flagship models for most agentic coding tasks.

Windows: https://developer.microsoft.com/en-us/windows/dev-tools
Microsoft AI: https://microsoft.ai/news/building-a-hillclimbing-machine-launching-seven-new-mai-models/

Episode 517 starts with a light chat about AI avatars and new text‑to‑speech deepfakes before diving into LLM “thinking” modes—what baked‑in planning actually does, why it multiplies token costs, and when it helps or hurts. James and Frank give concrete dev advice: try low‑thinking settings, use big models for creative planning then smaller ones to execute, leverage harnesses/system prompts, and beware quantized local models often do better without thinking.

James and Frank unpack AI-driven development shifts—agent SDKs, session management, and the rise of agent-first UIs like Google’s anti-gravity and GitHub Copilot—showing how VS Code’s Agents window, worktrees, sub-sessions and tunnels help manage multi-repo cloud and local workflows. They share practical takeaways—why SDKs are essential, when to stay code-first, how subsessions and remote tunnels protect your machine, and what to watch for in sandboxing and integration gaps.

James and Frank share thrift‑store monitor triumphs (and a retro Wii audio nightmare) before diving into a no‑nonsense guide to buying displays: what ports, VESA mounts, and speakers really matter. They demystify HDR, mini‑LED vs OLED, refresh‑rate math (why 40/120Hz divides matter), and modern upscaling—plus the surprising developer headaches of getting HDR to actually work. Practical tips and geeky fun for anyone shopping or building.

In this episode James and Frank dive into running AI coding models locally versus in the cloud—BYOK/Open Router, VS Code’s chat/agent harness, model runners (Olama, vLLM), and the practicality of 27B models on a 3090 using 4‑bit quantization. They share hands-on takeaways—how recent engineering (MT/MTPLX) boosts inference to usable token rates, when auto model selection makes sense, cost and hardware trade‑offs, and why local models can liberate your workflow while still needing smarter, unified tooling.

James and Frank explore the future of developer workflows powered by AI agents, revealing how developers are shifting from coders to testers and product strategists. They dive into new research-planning-implementation cycles, GitHub's usage-based pricing changes, and why developers must think strategically about model choices and token costs. Plus, reflections on Apple's leadership transition and the need for bold hardware innovation.

After a quick round of Apple dev updates, James and Frank dive into Matter — why Frank went from skeptic to devotee and why it’s poised to fix IoT chaos. They break down how Matter (an app-layer protocol running over Wi‑Fi/IPv6) and Thread (an open mesh transport) simplify secure onboarding with QR/Bluetooth, enable true interoperability across Apple/Google/Amazon, and make DIY devices easy to build and integrate.

In episode 511 James and Frank dig into audio gear (why impedance matters and the pros/cons of hardware DSP and headphone high‑power modes), explore GitHub Copilot CLI's new Remote Sessions for mobile access to your local dev environment, and reveal a clever hack: MIDI‑over‑USB turns modern iPhones into reliable wired controllers for embedded hardware and robotics. Tune in for practical tips, safety notes on power/latency, and real‑world developer workflows.

James and Frank dig into the messy world of AI agents—Claudes, Copilots, “claws”—and why now is the wrong time to over-box these tools. They debate GUI vs. CLI futures, explain when AI should be invisible and baked into workflows, and warn about AI-produced “slop” and long-term maintenance. Plus, James’ Light Phone/e-ink experiment adds a practical take on simplicity and real-world tradeoffs.

In this episode Frank and James riff on everything from accidental source‑map leaks (yes, even Apple slips up) to using Copilot and multi‑agent AI workflows to triage issues, write PRs and even plan trips. They share hands‑on tips—use “plan mode” and deep research, turn plans into issue artifacts, run multi‑model code reviews, cap concurrent agents (~3), and babysit agents—to boost productivity while avoiding chaos and burnout.

At MVP Summit we dig into Agentic Workflows — write Markdown prompts that drive AI agents to run CI, open PRs, and automate cross‑repo tasks — and MAUI DevFlow, which lets agents interact with native UIs to click, screenshot and validate designs. Listen for practical takeaways on ditching brittle YAML/scripts and automating tedious maintenance and testing, plus the real caveats: security front‑matter, a compile/lock step and token costs.

In Episode 507 James grills Frank about UI choices in his iCircuit app—crowded toolbars, popover behavior on iPhone vs iPad, and flaky TestFlight/App Store review timing. They trace a nasty bug to Apple’s deprecated OpenURL behavior in iOS 26, debate in-app browsers vs Safari and multi-window docs, and remind developers that subtle platform changes and lack of polish can silently break UX—practical lessons for shipping mobile apps.

James and Frank unpack the exploding world of AI coding agents—covering instructions, MCP tools, custom agents, hooks, plugins and why “skills” matter. They walk through the new .NET Skills repo (P/Invoke, MSBuild, diagnostics, binlogs), show how skills act like practical, on‑demand tutorials for niche tasks, and sketch how tooling will soon auto-load the right skills so agents can just do the thing for you.

On this episode we wade through Apple’s drip‑fed “event week,” from the budget iPhone 17E to the new MacBook Neo/Air M5 and the pricey Studio Display, and argue who each device is really for. The big takeaway: 8GB of RAM is borderline unusable for modern web and dev workflows—buy up if you can—while the Air M5 looks like the sweet spot; plus a fun detour on using Copilot/Fleet to scaffold apps in minutes.

Frank Kruger celebrates iCircuit’s 15‑year journey—from an iPad‑era $20 launch and surprising overnight success to the present—unpacking the tradeoffs between maintaining a flagship app and chasing new side projects. He offers candid lessons on pricing, backporting legacy fixes, modernizing pipelines, and explains how AI agents have reshaped his workflow and will power the next wave of iCircuit features.

On this episode we dive into Tiny Tool Town — a GeoCities‑style app hub for tiny developer utilities — and James walks us through building Tiny Clips, a Mac toolbar screen‑capture app he prototyped and shipped using Copilot, agentic workflows, and a plan‑implement‑review cycle. Expect practical takeaways on multi‑model AI pipelines (planning with 5.2, coding with Codex/Opus), CI/publishing tips, sandboxing/TestFlight pitfalls, and why tiny apps are booming.

Discover the machine learning breakthrough changing everything: rectified flow. Frank breaks down how this revolutionary technique replaces 1,000-step diffusion with just 10, delivering faster and higher-quality image and video generation. Learn why major players are adopting it and how it's democratizing AI for everyday creators.

In this episode James and Frank walk through the latest Copilot CLI power-ups—Autopilot loops, experimental Fleet/parallel agents, and Opus model/context updates—while demoing how they used plan mode to spin up a full MAUI pet‑insulin app end-to-end. Learn what Autopilot and Fleet actually do, how parallel agents orchestrate work, plus practical tips (watch your context window, use plan mode) for turning AI agents into fast prototypes.

On our 500th episode James and Frank celebrate the milestone, reminisce about their mobile‑dev roots, and dig into how AI, the Copilot CLI/SDK and the Model Context Protocol (MCP) are reshaping workflows. Frank demos an MCP‑powered tool that turns app reviews into prioritized GitHub issues and automations — a real example of AI-as-glue — with practical takeaways on prompt engineering, UI extensions, and when to automate versus curate manually.

In episode 499 James and Frank dive into the messy, exciting world of coding agents — from burning through Copilot credits and avoiding merge conflicts to practical workflows for letting agents run tasks while you sleep. They share real tips: break big features into bite-sized tasks, have agents ask clarifying questions, and use Copilot CLI or the new SDK to resolve conflicts, auto-fix lint/build failures, and automate mundane repo work.

The conversation then maps the evolution from simple completions to autonomous loops like Ralph — a structured, repeatable process that generates subtasks, runs until acceptance tests pass, and updates your...

498: CI/CD fro Mac Apps: GitHub Actions to Notarize

01/19/2026

This episode opens with mic and Nintendo banter before plunging into macOS release pain points: sandboxing, hardened runtime, notarization, Sparkle auto‑updates, and automating releases with GitHub Actions and tags. James and Frank offer practical tips—drag builds into /Applications to test signing—and unpack .NET 10 trimming/reflection pitfalls and CI/CD quirks for anyone shipping native apps outside the App Store.

In this episode James and Frank dive into the practical realities of using AI in everyday development—arguing that AI shines in brownfield (existing) code because it respects your architecture, while greenfield work rewards iterative prompting. They unpack model quirks: context-window limits, hallucinations, and why trying different models matters. The heart of the show is Frank’s nerdy delight: feeding a 64KB EEPROM through a disassembler and having Sonnet decompile it into readable C, exposing a PID autopilot and hardware checks—proof that AI can accelerate reverse engineering and embedded work. Along the way they share hands-on tips (trim and cl...

496: All The Holiday Hacks

01/05/2026

James and Frank unwrap 2025 as the Year of AI Development, covering new models, the rise of agents, and editor integrations like Copilot in VS Code that changed how developers write and maintain code. You’ll hear practical takeaways—how next-edit, local models, RAG/vectorization and app‑on‑demand sped prototyping, slashed maintenance time, and why the hosts think the AI boom has legs into 2026 despite looming uncertainty.

Merge Conflict

534: Why Developers Are Obsessed with Budget LLMs (Jev, Luna, & more)

Follow Us

533: iPhone Duo Development: Navigating Dual Screens and Layouts

Follow Us

532: HydraFusion: Multi-Model Orchestration Changes Everything

531: How AI Saved Our Lost Podcast Recording

Follow Us

530: WinUI Goes Open Source & Mac Mini/Studio Upgrades!

Follow Us

529: Agent, MCP, and Tooling: Taming Cross‑Platform DevFlows

Follow Us

528: Build, Ship, Repeat: AI Tools Changing App Development

Follow Us

527: AI Agents Meet Mobile: GitHub Copilot Canvas Revolution

Follow Us

526: GitHub Stacked PRs: Managing Complex Code Changes

Follow Us

525: Why Christopher Nolan's IMAX Vision is Fading

524: A .NET MAUI AI Revolution

Follow Us

523: UI.md - Opinionated design rules for coding agents

Follow Us

522: Our Thoughts on Slate: The $25K Modular Pickup Changing EVs

Follow Us

521: Polish Matters: UI, Icons, and AI Design Fails

520: Inside the New GitHub Copilot App: Sessions, Canvases, Automations

519: WWDC Deep Dive: Apple’s New AI for Developers

Follow Us

518: Windows is Back! New Microsoft AI Coding Models

Follow Us

517: Plan First, Think Less: Save Tokens, Improve Code

Follow Us

516: Evolving Agent Session Management

Follow Us

515: Mini-LED, IPS, and the Great Monitor Rabbit Hole

Follow Us

514: Running Local LLMs in VS Code

Follow Us

513: Agents Over Chat: The Future of Developer Workflows

Follow Us

512: Does Matter Really Matter?

Follow Us

511: Terminals, Remote Sessions, No More Watches!

Follow Us

510: AI Agents: Claws, Copilot, GUI vs CLI Debate

Follow Us

509: How AI Fleets Fixed 31 Issues in Two Days

Follow Us

508: Agentic Workflows - Markdown Automation for GitHub Actions

Follow Us

507: iCircuit UI Roast: Toolbar, Popovers, and Polish

Follow Us

506: We have no skills

Follow Us

505: 8GB of RAM Isn't Good Enough #MacBookNeo

Follow Us

504: 15 Years of iCircuit

Follow Us

503: Welcome to Tiny Tool Town

Follow Us

502: Rectified Flow Revolution - AI Image Generation Gets Smarter

Follow Us

501: Autopilot, Fleets, and Parallel Agents Explained

Follow Us

500: How Frank Builds Apps Has Changed Forever

Follow Us

499: Going Full Ralph, CLI, & GitHub Copilot SDK?!?!

498: CI/CD fro Mac Apps: GitHub Actions to Notarize

Follow Us

497: Turning Machine Code into C with AI

496: All The Holiday Hacks

Follow Us

495: 2025 Wrapped - The Official Year of AI Development

Follow Us