The most common debug nightmare in multi-agent systems is 'the answer is wrong, but I don't know which agent did it.' Three layers of observability are essential: per-agent token metering, execution traces, and real-time cost dashboards.
Multi-agent orchestration splits into three camps: scripted determinism (LangGraph, Claude Code Workflow) is predictable but rigid, model-driven (Codex, Devin) is flexible but unpredictable, and hybrid (Windsurf 2.0) acts as a command center integrating multiple agents. The choice depends on how much predictability you need.
Multi-agent security risks aren't just amplified single-agent risks — inter-agent communication is itself an attack surface. A compromised sub-agent can pass malicious instructions to the parent through its return value. Core defense: treat agent output as untrusted data.
Three routes to commercial document parsing: specialized parsers (Cohere Parse at $1.50/k pages, LlamaParse Agentic Plus at 90.2% on ParseBench), Big Three cloud prebuilts (Azure/Google/AWS for structured field extraction), and general-purpose VLMs (Fable 5.1 scores 78.92 on ParseBench and crushes specialized parsers on charts, but costs 3–16× more and hallucinates). At 100K pages/month, plain OCR runs ~$150 across providers; add tables and AWS jumps to $1,500, Claude Sonnet 5 to $900. The first question isn't 'which is most accurate' — it's 'do you need transcription or comprehension?'
WebMCP is a browser-native W3C standard proposed by Google and Microsoft. It lets web pages expose structured tools to AI agents via document.modelContext.registerTool() — no backend, no HTTP/SSE transport. Chrome 149 Origin Trial is live.
The finale reads Week 11: Monday upgrades instruction-waiting reactive assistants into proactive agents that observe, infer, and act first via the GUM paper, while Wednesday folds multimodal systems, long-running agents, and production observability into three open problems. Ends with a pre-Demo-Day checklist and a one-line map of all 11 posts.
Week 2 connects model internals to retrieval engineering: decoding, attention, training, inference and context budgets in Lecture 02; chunking, indexes, sparse/dense retrieval, reranking and agentic RAG in Lecture 03. Separate reading digests examine workflow selection and the RAG and ColBERT experiments.
Lecture 04 follows tool use from training and API retrieval to function calling, MCP and reliability evaluation. Lecture 05 compares DSPy, LangGraph, harnesses, Temporal and observability. Separate MCP and DSPy reading digests explain the protocol and optimization mechanisms; HW1 remains a hand-built harness.
Week 4 connects the ReAct loop and MemGPT memory hierarchy to Lecture 06/07: composing design patterns, keeping context focused without losing commitments, choosing memory architectures, and evaluating correct use. An original email-assistant example tests whether compaction preserves an approval constraint.
Lecture 08 explains orchestration, communication, task ledgers and worktree isolation. Lecture 09 starts with data and evaluation, then compares context optimization, SFT, RL and distillation. Use a traceable email example to decide whether collaboration or optimization addresses the actual bottleneck.
Week 6 connects Terminal-Bench outcome verification with Shankar’s data flywheel: check what the agent actually changed, turn reviewed failures into development examples, and keep independent test tasks separate. HW1 is due October 30.
Week 7 is midterm checkpoint week: data selection on Monday, evaluation and benchmark design on Wednesday. Zhu et al. teach you not to be fooled by your own scores, SWE-smith scales software-engineering tasks to 50,000 instances, and you close by drafting a first 4-tuple for HW2.
Week 8 builds model judges with MT-Bench and Anthropic's eval guide on Monday, then faces production leakage with PrivacyLens and four guardrails on Wednesday. The paper video is due Friday, and this week's deliverable is one working judge score plus one permission check.
Week nine turns to coding agents: SWE-agent shows interface is performance, OpenHands packs sandbox plus benchmarks into one general base, and the second homework is due Friday — ship one working bug-fix exam this week.
Start with the opening slides on the agent loop, components, and engineering challenges, then examine BAIR's compound systems, Ng's work patterns, and Si's execution-feedback experiments. Distinguish the lecture, required reading, and additional readings.
A close look at three open-source projects training LLMs from scratch in the Chinese community — baby-llama2-chinese (218M, 63.4B tokens), ChatLM-mini-Chinese (0.2B T5, 10.23M dialogues), and Steel-LLM (1.12B, 1T tokens, 8 months) — comparing corpus strategy, tokenizer decisions, and community ecosystem. Honest evaluation included: baby-llama2 scored a bottom-ranking 21 in MiniMind's side-by-side test; ChatLM has the strongest knowledge (62) but weak coding.
Four coding agent TUIs all follow the same two rules: tools maintain constant height (one-line summary, never jumping from 0 to N lines) and never auto-collapse (only user-initiated expand/collapse). looplane violates both.
OpenELM uses layer-wise scaling to shift parameters toward layers near the output; with 1.08B parameters and 1.5T tokens it beats OLMo 1.2B (+2.36% on the LLM360 average) despite OLMo training on 3T tokens. MiniCPM trains multimodal small models from scratch with a three-stage unfreezing recipe (Resampler first, vision encoder next, everything unfrozen last); MiniCPM-V 4.5 reaches sub-30B SOTA on VideoMME with only 8B parameters, and 4-bit quantization squeezes fp16's 16–17GB memory footprint down to about 5GB for phones.
A tour of Karpathy's three teaching repos: nanoGPT (2022, ~300 lines each to reproduce GPT-2 124M), llm.c (pure C/CUDA training), and nanochat (2025-10, one speedrun.sh from tokenizer to WebUI). $100 and 4 hours on 8×H100 buys a chatty model; GPT-2-grade capability is now down to about 2 hours and $48.
LitGPT (Lightning AI, ~13,600 stars, Apache 2.0) rewrites 20+ mainstream LLMs — Llama 3, Qwen2.5, Phi 4 — from scratch as single-file, no-abstraction implementations, with a full pretrain / finetune / evaluate / serve CLI. TinyLlama (1.1B parameters, 3T tokens) was trained on this codebase. This post breaks down how it differs from MiniMind, how to actually use it, and where it stops.