About this episode
Watch the https://bolt.new Reward Ceremony Live!July 26th 10AM PST on https://www.twitch.tv/theprimeagen#sponsored https://balls.yoga📌 Chapters:00:00:00 - balls.yoga, drama & intro00:02:00 - Building OpenCode: terminal AI agents00:03:00 - Bolt.new Reward Ceremony - July 26 10AM PST00:03:18 - Agent architecture basics explained00:04:08 - What is an AI Agent?00:05:00 - Prompt engineering and tool calling00:06:10 - OpenCode UX and mobile plans00:07:27 - Why OpenCode runs agents locally00:08:34 - Vision for mobile remote dev workflows00:09:15 - The reality of build opencode00:10:10 - Claude vs Opus vs others00:13:32 - Tool usage & LSP integration00:14:20 - LSP tool feedback loop00:17:00 - Claude is best at calling tools00:18:10 - Prompt tuning limitations and expectations00:22:20 - Modeling agent loops with tool calls00:24:40 - Why building a real agent isn’t just weekend work00:26:20 - Security tradeoffs in early development00:28:30 - Loop management & lossy compression00:29:40 - Session design & managing context00:30:40 - Parallel sessions and subagents00:33:20 - Adam banned from Twitch by his cofounder00:34:20 - OpenCode & Terminal.Shop development connection00:35:00 - What surprised them building an agent00:36:30 - Why TUIs are hard compared to web00:37:30 - Layering, pixel constraints, UI struggles00:38:30 - Final laughs & wrap-up00:39:00 - Plug for OpenCode: sst.dev/opencode00:40:00 - Outro: still no clue how agents work, but fun ride⸻Topics Covered:• AI agents in terminal environments• OpenCode’s CLI-first agent approach• Loop-based tool calling in LLMs• Integrating LSP feedback into agents• Prompt engineering & model tool awareness• Evaluating agents: benchmarks & metrics• Local vs remote dev environments• Designing for low-friction agent UX• Safety, permission models & sandboxing• The tradeoffs of TUIs vs web UIs• Session design, compression & memory limits• Humor, Twitch drama & voice AI characters
