Know what is happening in AI, and what to do about it.

A dedicated IDE for agent research launches

AgentsDock pitches itself as built for iterating on agent behavior rather than shipping code, landing 78 points and 32 comments on its HN launch.

Anthropic's Houthi missile-guidance disclosure sharpens misuse stakes

A specific new case, distinct from the general misuse report already covered, adds weapons-development detail that raises export-control and.

Ask HN model-choice thread draws real debate

58 points, 99 comments on which model developers default to and why, a useful raw barometer of what practitioners actually run day to day.

Bengio publishes on why agents lie and cheat

Yoshua Bengio's analysis of deceptive and coordinating behavior in current agent systems topped HN with 517 points and 601 comments, feeding the.

DeepSeek's Harness plugin layer explodes

Deepseek-ai's "Harness" agent-plugin framework is the fastest-growing repo tracked, adding 222,445 stars in 31 days at 7,104 stars/day, positioning.

The play
["Track which plugin standards gain adoption, and test portable workflows before committing your team to one agent platform.","Pilot the org-chart template with one repeatable workflow, then measure whether role-specific agents improve speed without adding review overhead.","Centralize and anonymize agent session logs, using emerging extraction tools to identify reusable prompts, failures, and customer-data risks.","Survey your team’s actual model choices, then standardize defaults around cost, reliability, privacy, and task-specific performance.","Treat undocumented sandbox behavior as a security signal, and review isolation, data access, and vendor disclosure before deploying sensitive workloads.","Test research-focused

Distillation economics enter the GTM conversation

Tan's push for open-weight labs to distill rather than train from scratch is a cost-structure argument relevant to any agency reselling model.

Garry Tan's gstack open-source ops stack

a public repo of 23 Claude Code tools mimicking CEO/Designer/Eng-Manager/QA roles. No direct price, but the free "AI operations team in a box" repo is gaining roughly 714 stars/day (132k total) and drives inbound credibility for Tan's investing/advisory business.

AI chat-history extractor

free open-source tool pulling chat histories out of Claude Code, Cursor, Windsurf, Aider, and Cline/Roo Code into portable formats. 333 stars in 2 days, 159 stars/day velocity; no paid tier yet but sets up a migration-consulting angle.

Build-and-sell $2K AI agent playbook

a YouTube creator demos building and selling a client-ready agent in 24 hours, funneling viewers into a paid AI-agency skool community. No public numbers on the community's size, but the funnel structure (free demo, paid community) is the actual mechanic.

French Claude Code skills-to-AIOS funnel

creator gives away 3 Claude Code skills free while funneling viewers to a paid "AIOS" business operating system product at aios.fr. No traction numbers surfaced, worth watching as a non-English-market variant of the skill-bundle funnel.

Solo dev's local ADR generator from agent logs

adr-gen is a local CLI reading Claude/Codex session logs and git diffs to auto-generate architecture decision records. No public user numbers, single LinkedIn post is the only evidence, but it turns agent session exhaust into a sellable documentation artifact if it ships paid.

"AI second brain" tied to $1M agency claim

creator demos a personal knowledge system built on GPT-6 Astra, linking to a paid "AI Automation Society" community pitched as a $1M agency playbook. The dollar figure is unverified marketing copy, flagged low confidence, but shows the going rate creators think agency-building education commands.

Working with Claude

The field guide to getting real work done with Claude.

Working with Claude, for Developers

The developer edition: agents, tooling, and build patterns.

The Second Brain Field Guide

Turn AI into a system that remembers and works for you.