Masthead · The desk
About
Agentic Firehose is how an engineer gets proficient at architecture judgment plus AI orchestration: wrap models into real systems, including old ones. Not autocomplete. Catch up is the four-skill spine. Starting a project is the Monday order, including wrap into the old app.
The job → the map
Recruiter language, then this desk
Recruiter job
Ng skill · this desk
01Layer AI into a 40-year platform. Classify / score / anomaly — the old system of record still runs the company. The wrap is the product.
Catch up 01 + new /project 07. Make an AI thing that actually works. Then wire it into the existing app. Skip 07 only if this is greenfield.
02Architecture + full-stack ownership. Tradeoffs, tests, security, and the spec. You steer the tools instead of vibe-coding the wrong stack.
Catch up 02 and 04. Software engineering fundamentals. Shaping the build — decide what is in the spec.
03Beyond AI assistance. Not autocomplete. Not a copilot overlay. An agent with hands, a ticket, and a repo this week.
Catch up 03, Claude, /project 01. Using coding agents. Home-stack cookbooks. Pick a coding agent and close a real ticket.
04Multi-agent pipelines. Loops, graphs, harnesses. Work that must survive a crash, and more than one pair of hands.
Catch up 01 harness/graphs, /project 03–04. Give it hands, then bound them. Graph only if the work must resume.
Independence
The name is a domain, not a product of Anthropic. This site is independent and not affiliated with Anthropic, OpenAI, Google, or any other lab.
Readers
Forward-deployed engineers. Software engineers. Anyone filing work against Ng’s four-skill map — wrap, eval, ground, ship.
Cut
Train models, VPCs. Sheets canvas, chip keynotes, AGI timelines, training runs, and networking desks stay off this board.
How the desk works
- Live items are fetched on the server from public RSS, the GitHub API, and the X desk file the site bot writes on a weekday schedule. Cached about 20 minutes.
- Editorial items are a curated file in this repo — splash ranking and systems classics. They fill gaps. Live X is the desk dump, not placeholder tweets.
- Tags on each headline are the six sub-skills from Ng’s “Building and deploying AI applications” letter (The Batch, 21 Aug 2026). Click a tag to narrow the board. The four-skill map is the Catch up spine.
- The live filter prefers RAG/grounding, evals-as-acceptance, harness/MCP, coding agents, and shipping the app (cost, latency, cloud). Labs is cookbooks for building AI apps: context, harness, graphs, cloud — not a model-drop ticker and not a networking desk. Repos are clone-to-ship. Architecture notes, pretrain launches, VPC walkthroughs, and high-star “agent platform” cards stay off.
- There is no signup and no paywall on this cut. The board is the site. Starting a project is not Catch up: it is eight steps and a URL, not a course catalog.
Live RSS desks
- Google DeepMind — labs
- Hugging Face — labs
- OpenAI — labs
- Google AI Blog — labs
- Anthropic News — labs
- Anthropic Engineering — systems
- LangChain — systems
- Lilian Weng — systems
- Simon Willison — threads
- Latent Space — threads
- Interconnects — threads
- Sebastian Raschka — threads
- arXiv cs.AI — papers, keyword-filtered
X watchlist
Threads mix live posts from these accounts (originals, not replies or retweets). ClaudeComputingBot writes them to src/content/x-desk.json on a weekday schedule. Bookmark-sourced cards come only from the X folder AI Posts (2093550752241164315), not the full bookmarks archive. RSS desks still run live. No Exa, Firecrawl, or X API key.
- @karpathy — Andrej Karpathy
- @simonw — Simon Willison
- @swyx — swyx
- @AnatoliKopadze — Anatoli Kopadze
- @KirkDBorne — Kirk Borne
- @AnnatarXBT — Annatar.md
- @virgilxbt — virgilxbt
- @hrswatigupta — Swati Gupta
- @natolambert — Nathan Lambert
- @rasbt — Sebastian Raschka
- @AndrewYNg — Andrew Ng
- @bcherny — Boris Cherny
- @LunarResearcher — Lunar
- @hanakoxbt — Hanako
- @avrldotdev — avrl
- @techNmak — Tech with Mak
- @sairahul1 — Rahul
- @suraj_sharma14 — Suraj Sharma
- @AlexFinn — Alex Finn
- @callanxai — Callan
- @Ai_Vaidehi — Vaidehi
- @jobescape_ai — Jobescape
- @rubenhassid — Ruben Hassid
- @eng_khairallah1 — Khairallah AL-Awady
- @SunNeverSetsX — 摸鱼巨匠🔨
- @GoGoFly23 — GoFly
- @Ryrenz — Ren
- @BharukaShraddha — Shraddha Bharuka
- @DataChaz — Chaz Wargnier
- @0xMovez — Movez
- @promptparag — Parag Pawar
- @RishiUvaach — Rishi
- @arceyul — arc.
- @DavidOndrej1 — David Ondrej
- @polydao — Mr. Buzzoni
- @undefinedKi — Yarchi
- @reiraxbt — reiraxbt
- @SofiaSici — Sofía Sicilia
- @VibeMarketer_ — J.B.
- @beamnxw — beamnxw ./
- @LearnWithBrij — Brij Pandey
- @gippp69 — Gipp
- @seekjourney — 极客杰尼
- @Sumanth_077 — Sumanth
- @JayBisen473370 — Jay Bisen
- @hilbertspaess — Jacob Coxon
- @deanwperkins — Dean W. Perkins
- @0xCodila — codila
- @AdarshChetan — Adarsh Chetan
- @Mahaximus_ — Mahax
- @fabrice_mayrand — Fab
- @claudeskills101 — Mr. Claude
- @txbrraa — tobarra
- @_jaydeepkarale — Jaydeep Karale
- @v1lrok — v1lrok
- @alex_prompter — alex_prompter
- @ecommartinez — ecommartinez
- @0xCodez — 0xCodez
- @AvinashSingh_20 — AvinashSingh_20
- @EngMoElgaraihy — EngMoElgaraihy
- @dkare1009 — dkare1009
- @hasantoxr — hasantoxr
- @iiiichigo_chan — iiiichigo_chan
- @rvaniaaaa — rvaniaaaa
Repo watchlist
These are fetched live from GitHub and mixed with a search for recently updated agent/harness repos.
- upstash/context7
- farion1231/cc-switch
- deepmemteam/deepmem-claude-plugin
- slowave-ai/slowave
- addyosmani/agent-skills
- coleam00/Archon
- mksglu/context-mode
- zachsaw/graphify-csharp
- affaan-m/ECC
- anomalyco/opencode
- earendil-works/pi
- stablyai/orca
- ChromeDevTools/chrome-devtools-mcp
- MemPalace/mempalace
- DeusData/codebase-memory-mcp
- bytedance/deer-flow
- deepseek-ai/deepseek-harness
- Graphify-Labs/graphify
- thedotmack/claude-mem
- unclecode/crawl4ai
- ucsandman/declick
- OtoDock/oto-dock
- diegosouzapw/OmniRoute
- mixpeek/amux
- yc-software/qm
- Untrivial-ai/agent-orchestrator
- Waishnav/devspace
- chaitanyagiri/munder-difflin
- vanillagreencom/kendex
- omartelo/lich
- blackwell-systems/mcp-assert
- Staatsgeheim/MathKernel
- okf-memory/okf-agent-memory
- redhat-et/ripwire
- huggingface/funes
- Kiln-AI/Kiln
- vshulcz/deja-vu
- Undertone0809/rudder
- google-deepmind/amplio
- Rul1an/assay
- mezmo/aura
- uipath/coder_eval
- langfuse/langfuse
- comet-ml/opik
- VectifyAI/PageIndex
- getzep/graphiti
- strands-agents/harness-sdk
- dbt-labs/dbt-mcp
- Jwuthri/Tracely-ai
- github/github-mcp-server
- cobusgreyling/loop-engineering
- xai-org/grok-build
- truefoundry/trueforge
- headroomlabs-ai/headroom
- JordyZomer/lemmalog
- HowProgrammingWorks/Agent
- CopilotKit/OpenBot
- browser-use/macos-harness
- furkankly/zoetrope
- iAmCorey/Wake
- run-llama/llama_index
- deepset-ai/haystack
- modelcontextprotocol/python-sdk
- stanfordnlp/dspy
- explodinggradients/ragas
- promptfoo/promptfoo
- langchain-ai/langgraph
- openai/codex
- OpenHands/OpenHands
- huggingface/smolagents
- browser-use/browser-use
- pydantic/pydantic-ai
- google/adk-python
- langchain-ai/deepagents
- letta-ai/letta
- osaurus-ai/osaurus
- vercel/ai
- openai/openai-agents-python