Issue #10 ·
Five things from the last day worth your attention if you build with AI. No fluff.
Five things worth your attention if you build with AI, about 3 minutes. Join free and get 13 ready-to-paste Starter Prompts for coding agents, vibe coding and game dev right away.
GAME DEV
At the Unreal Fest Tokyo keynote, Epic said UE6 will drop Actors, Blueprints and the C++ Gameplay Framework when it reaches Early Access, moving gameplay code to Verse and Scene Graph. C++ stays for the low-level engine, UE5 projects can keep going, and Epic plans AI-assisted conversion to Verse.
Why builders care: If your game logic lives in Blueprints, start planning for Verse now.
MODEL
LongCat-2.5 is a natively multimodal MoE with 1.6 trillion total parameters and about 48 billion active per token, trained with large-scale RL on long coding tasks. Meituan's own table puts it at 88.4 on Terminal Bench v2.1, up from 70.8 for LongCat-2.0, using a Claude Code harness.
Why builders care: Another agentic coding model worth testing; treat the scores as vendor-reported.
CODING
With no open-source engine to build on, Levels says Claude Code disassembled the shareware demo into about 57,000 x86 functions, translated them into roughly 100,000 lines of C, compiled that to WebAssembly and rewrote the Direct3D calls for WebGL 2. He says it took about 1 million output tokens.
Why builders care: A vivid data point on what agents can do with legacy binaries.
AGENTS
Send one MP3 or WAV between 5 seconds and 5 minutes plus plain-language questions, and you get back probabilities instead of a transcript: yes or no, multiple choice, scores on a scale, or time spans. Pricing is $0.47 per million input tokens, with a 1 million token trial credit.
Why builders care: A cheap way to test voice agents, like flagging calls where the agent talks over the caller.
REPO
FlashProve produces cryptographic proofs that a reply came from a committed set of weights, without revealing those weights. It runs full-size Llama 4 Maverick on one H100 at about 110 tokens a second of prefill, and anyone can verify a proof on an ordinary CPU in seconds. It's an Apache 2.0 research artifact with a 1,024-token context, and it isn't maintained.
Why builders care: If customers ask you to prove which model answered, this is a working reference.
That's the issue. Know a builder who'd use this? Share it on X or forward them this page. New here? Subscribe free for tomorrow's.
Five things worth your attention in AI, agents, vibe coding and game dev, every weekday. Plus 13 free Starter Prompts and the AI Builder Starter Stack PDF when you join.
Want the bigger picture? Read our guides to AI coding agents, vibe coding tools, AI game dev tools, agent frameworks, MCP servers and web scraping tools, or all guides.
