AI Agent Builders Worth Following2026-07-04Indie Agent Builders — Week of June 27This week’s issue focuses on the shift from agent demos to harness engineering: subagent delegation, prompt evaluation, review gates, resumability, and token/context control. It covers Simon Willison’s Fable-driven build logs, AI Engineer World’s Fair’s loops and software-factory debate, and the GitHub projects with the clearest implementation signals for agent builders.
AI Agent Builders Worth Following2026-06-20Indie Agent Builders — Week of June 20Z.ai's GLM-5.2 went MIT open-weights with 744B MoE parameters and Code Arena #2 ranking the same week Fable 5 stayed suspended — its trigger confirmed as a "fix this code" prompt on CVE-laden code. Simon Willison shipped five Datasette alpha releases that completed a multi-year arc from read-only explorer to read-write platform with sandboxed HTML apps. Swyx tracked GLM-5.2 across four AINews issues from launch to community vibe check. On GitHub, headroom (+12.8k stars) led a week dominated by token compression, code intelligence MCP, and skill-security scanning.
AI Agent Builders Worth Following2026-06-13Indie Agent Builders — Week of June 13Export control kills Fable 5 access in real time; Simon ships ask_user() HITL and JSON extras API; Swyx's Loopcraft and Git-may-die thread; last30days-skill hits #1 on GitHub.
AI Agent Builders Worth Following2026-06-06Indie Agent Builders — Week of June 6The week split along two axes: agent sandboxing graduated from theoretical concern to installable code (Simon Willison's micropython-wasm, OpenAI Lockdown Mode, Meta's support-bot permissions failure), and the harness/skills layer kept consolidating around common infrastructure (obra/superpowers at 220k★, MemPalace v3.4.0, NVIDIA Nemotron 3 Ultra purpose-built for Hermes Agent and OpenClaw). Swyx's Latent Space covered Microsoft Build 2026 including the accidental Mythos FLOP leak, Cognition's first real-world eval dataset (258 enterprise sessions, financial guarantee), and a guest post on RL harness quality with a concrete 5% failure-rate diagnostic rule.
AI Agent Builders Worth Following2026-05-30Indie Agent Builders — Week of May 30The week's activity converges on a clear shift: agents are graduating from a model-quality competition to a harness engineering competition. Simon Willison's Anthropic PMF essay (backed by the $65B Series H at $47B ARR), Claude Opus 4.8's mid-conversation system messages, and the explosive growth of harness tooling (ECC at 199k★, PilotDeck's workspace OS, compound-engineering-plugin) all point the same direction. Swyx's Cognition D-round coverage ($1B at $26B valuation) and harness-as-differentiator thesis provide the commercial and architectural framing; the trending repos section documents seven concrete tools engineers can evaluate today.
AI Agent Builders Worth Following2026-05-23Indie Agent Builders — Week of May 15Simon Willison shipped Datasette Agent alpha — the convergence of three years of his `llm` library with Datasette — while swyx open-sourced Kakuna, a five-layer framework for hardening vibe-coded codebases, with both independently diagnosing the same gap: agents generate code, but don't yet close the distance to production-grade software. Also: Gemini 3.5 Flash triples in API price, five standout trending repos (agent-browser at 34k★, OpenViking, hermes-agent v0.14.0 with 180× CDP speedup, and more).
AI Agent Builders Worth Following2026-05-15Indie Agent Builders — Week of May 8Simon Willison on agent trust and Shopify's River, Swyx on the Conductor convergence and Thinkymachines' 200ms multimodal leap, Geoffrey Huntley's Ralph Loop triggering Goal Mode across three platforms in 11 days — plus six trending repos.
