Building reliable, useful systems with LLMs and agents: what goes into context, how knowledge is retrieved, and how to make agent behavior trustworthy.
Concepts
- image-generation: controllable image models
- ai-red-teaming: breaking models before attackers do
- world-models: models of the physical world
- ai-infrastructure: AI factories and GPU networking
- multimodal-llms: vision LLMs in production
- agent-architectures: matching agent structure to the job
- ai-cybersecurity: AI’s impact on security testing and defense
- agentic-vulnerability-response: agentic CVE-to-protection pipeline in minutes, not days
- hebrew-llms: building Hebrew language models
- continual-learning: learning after deployment
- llm-pretraining: pre-training craft
- llm-reasoning: how models reason
- ai-moats: defensibility beyond the model
- solo-builders: building and shipping products without VC
- customer-pain-research: finding paid-software workarounds as business ideas
- enterprise-ai-adoption: enterprise AI go-to-market
- ai-growth-marketing: AI-driven growth and marketing
- workflow-automation: no-code automations and agents
- agent-memory: agent memory and its governance
- multi-agent-orchestration: orchestrating specialist agents
- small-language-models: when to train small models
- voice-agents: building phone/voice agents
- ai-rd-rollout: org-wide AI adoption
- autonomous-agents-outlook: speculation on autonomy and AGI
- ai-engineering-metrics: measuring coding-agent impact
- ai-sre: AI for production reliability
- llm-pipelines: when a pipeline beats an agent
- llm-cost-optimization: measuring (FinOps attribution) and reducing LLM spend
- ai-pricing: credits, buckets, and pools for pricing AI products
- ai-sdlc: AI across the software lifecycle
- agent-ready-products: make products discoverable and usable by agents
- coding-agent-workflow: day-to-day coding-agent practices
- agent-security: secure agents against prompt injection and misuse
- personal-ai-assistants: assistants for operational work
- skill-engineering: building and sharing agent skills
- decision-classifiers: cheap, calibrated decisions at scale
- proactive-ai: make AI capabilities discoverable
- future-of-software-engineering: how engineering roles shift
- ai-gateway: one layer for routing, cost, security, trust
- llm-evals: know it works, keep it working
- rag: retrieval-augmented generation
- agent-workspaces: shared human+agent workspaces
- second-brain: also Knowledge Management
- llm-inference: GPUs, parallelism, KV cache, and price per token
- model-selection: right-size the model; narrow the task
- open-weight-models: open models as a cost lever
- ai-hype: filter hype; build durable value
- harness-engineering: environment that lets agents run long and correctly
- agent-ready-codebase: make the repo operable by agents
- test-driven-development: frozen failing tests keep agents honest
- context-engineering: context makes the difference; keep it lean and current
- wiki-retrieval: finding the right page; multiple guesses, ranking, RAG/hybrid
- ai-guardrails: enforce rules technically, not by instruction
- ai-verification: read outputs, demand proof, check sources
- llm-wiki: Karpathy’s raw/wiki/schema pattern
- open-knowledge-format: Google’s universal knowledge-file format