Insights
Notes on shipping AI and software
Longer pieces from the Zheat team: crawlable on this site, not locked inside a LinkedIn feed.
Perpetual application: Grok Bot hits Composer agents, Rocky and No More Slop keep the gates
A perpetual application is a live product that keeps itself in production after launch. Pluggable gates (Appzi, Clarity, Ahrefs). Grok Bot organizes. Cursor Composer builds the local agents. Rocky and No More Slop keep the work fast and clean.
Read →Why I gave Grok Bot a Routine bot (and left features to humans)
Field notes on organizing a Grok Bot team for Order of Battle: a dedicated Routine bot, a PO that reads Appzi, a Developer that ships fixes — plus Rocky and Humaniseur behind the scenes.
Read →
Parallel agents are a false good idea
Parallel agents felt like rolling dice. I could not follow, could not answer, then had to test a feature bigger than I would have shipped. Going local forced a slower, clearer rhythm.
Read →Your quota is exhausted. Local models, hybrid use.
Token prices fall, consumption rises, quotas die faster. I moved toward local models on a high-RAM machine. Hybrid: keep cloud for the hard parts. You do not need a god model for a CRUD.
Read →Agents, same recipe, different ingredients
An agent is an LLM that can call tools. Same skeleton every time: model, prompt, MCP, sub-agents. YAML to instantiate them, a UI so the client creates them, no single provider lock-in.
Read →Humaniseur v5: style tells vs watermarks
Humaniseur v5 spec. Default rewrite is light (simple, even if incomplete). full is forensic. Two scores. Cannot certify Anthropic.
Read →mcp-raganything is not a RAG product
mcp-raganything wraps RAG-Anything as two MCP tools, index_corpus and ask_corpus. A socket for Claude, Codex, or Cursor. Not tenancy, rights, or production RAG.
Read →Humaniseur v4: bilingual AI prose cleanup
Humaniseur v4 white paper: 53 EN + 12 FR patterns, 0-100 AI density scoring, voices, French track, CLI validation. Open source.
Read →No More Slop: dual-score AI code cleanup
No More Slop white paper: 22 regex patterns, structural scoring, cost breakdown, style calibration, Rocky escalation. Open source.
Read →Rocky white paper: agentic engineering MCP
Rocky white paper: devkit gateway, handbook agents vs skills, architecture profiles, senior workflow, measured token savings.
Read →robots.txt decides, llms.txt suggests
llms.txt is not robots.txt for AI. One file answers crawl permission. The other is a reading list. Start with robots and a clean public graph.
Read →Rocky: free pre-PR MCP quality gate
Rocky is a free pre-PR quality gate for AI coding tools. Same model, about 65% fewer tokens. Works with Cursor, Claude Code, VS Code.
Read →Rocky MCP value report: tokens and cost
Same feature, same model: about 255k to 89k tokens (~65%), discovery down 89%. Side-by-side results without vs with Rocky MCP.
Read →Agentic coding: token and cost breakdown
One gateway vs many MCP tools, graphify before grep, pre-PR gate vs rework loops. Rocky keeps handbook overhead around 4-6k tokens.
Read →Why Rocky splits agents and skills
Agents are the short front door. Skills are deep reference loaded only when needed. Keeps Rocky sessions around 4-6k handbook tokens.
Read →Is RAG right for a messy implant catalog?
Stock sizes in tables. Preference notes and FAQs in chunks. Keep inventory and prose out of one pipe. We standardise on Xberg.
Read →Rocky profiles: detect before you impose
Rocky detects the repo before imposing structure: Next.js App Router, hexagonal, layered React SPA, or node-api-only. Match the playbook to the codebase.
Read →Rocky senior workflow: discover to PR
End-to-end pipeline Rocky reinforces: discover, match conventions, implement small diffs, verify locally, pre-PR gate until ready, then open the PR.
Read →I rebuilt an agent for the fourth time
Agent architectures converged: model, tools, MCP, sub-agents, observability. I wrapped it in YAML, then bigger players shipped better tools.
Read →Agents and RAG are no longer the hard part
After shipping agents for large accounts and startups, the pain was not implementation. It was cost, permissions, and observability.
Read →The ChatGPT that said yes. The API that said no.
A jewelry try-on looked perfect in ChatGPT. Same model on the API failed. We shipped a three-model pipeline that never rewrites the customer.
Read →The AI harness war is over: infra and privacy
LangChain, Claude Managed Agents, Agno AgentOS: same architecture. The harness is a commodity. The fight moved to infra and privacy.
Read →MCP is dead? Why a contract still helps custom APIs
Perplexity walked away from MCP and obituaries started. Token waste is real, but a contract still helps when LLMs talk to custom APIs.
Read →If AI wrecks your codebase, fix the frame
Blame the hammer, or fix the frame. AI amplifies what already exists. Rigorous standards accelerate. Missing standards accelerate debt.
Read →The RAG paradox: buy or build carefully
Standardise RAG and long-term memory instead of rewriting from scratch. Cognee looks ready for industrialisation: vectors, graph, MCP.
Read →