Article directory
Article directory
Find AI Agent knowledge through business questions, practical guides and concept comparisons.
Concepts
- Turning AI / Agent concepts into stories you can remember
- Acceptable Use Policy
- Access control
- Action
- AWQ
- Activation Checkpoint
- Activate
- Adapter
- Adversarial evaluation
- Adversarial examples
- Agent communication protocol
- Agent debate
- Agent decision tracing
- Agent Identity and Authentication
- Agent loop
- Agent negotiation
- Agent operating model
- Agent orchestration
- Agent Owner
- Agent policy
- Agent routing
- Agent runtime
- Agent swarm
- Agent
- Agent-to-Agent protocol
- Agent workflow
- AI assistant
- Intelligent customer service
- AI disclosure
- AI Governance
- AI impact assessment
- AI incidents
- AI-native workflow
- AI policy
- AI risk management framework
- AI risk management
- AI safety
- AI job replacement rate
- AI use case checklist
- AI workflow
- AI Operations
- Alignment Evaluation
- Alignment
- Annual recurring revenue
- Anonymization
- Answer Relevance
- Approval flow
- Approval threshold
- Approximate nearest neighbor search
- Arize Phoenix
- Artificial Intelligence
- Assistant
- Attention mask
- Attention mechanism
- ALiBi
- Attribute-based access control
- Audit evidence
- Audit logs
- Audit logs
- Audit trail
- Auditability
- Authority boundaries
- AutoGen
- Speech recognition
- Autonomy level
- Autonomy level
- Autonomous driving
- Autoregressive decoding
- Autoscaling
- Auxiliary loss
- Backdoor attack
- Backpropagation
- Batch inference
- Beam search
- Benchmark contamination
- Evaluation benchmark
- Best-of-N sampling
- BIG-bench (Beyond the Imitation Game Benchmark)
- Dual encoder
- Bias
- BERT
- bitsandbytes
- Blackboard architecture
- Blue-green deployment
- BM25
- bf16
- Browser tools
- Budget-aware routing
- Business continuity
- Byte Pair Encoding
- BPE
- Cache
- Canary release
- Canary release
- Capability assessment
- Catastrophic forgetting
- Causal attention / autoregressive
- Chain of Thought
- Chain-of-thought training
- Chain of thought and scratchpad
- Change Advisory Board
- Change management
- Chinchilla approach
- Chroma
- Chunked Prefill
- Chunking
- Churn rate
- Continuous integration and continuous delivery
- Circuit breaker
- Source citation
- Citation
- Cloud native
- Co-creation
- C2PA
- Code Interpreter
- Compliance log
- Compliance mapping
- Computer use
- Confidence threshold
- Conformity assessment
- Connector
- Constitutional AI
- Container
- Capability contamination and data leakage
- Content filtering
- Context Engineering
- Context contamination
- Context precision
- Context pruning
- Context recall
- Context length/window
- Context window
- Continued pretraining
- Continuous batching
- CLIP
- Contrastive learning
- Control Framework
- Control everyone
- Control testing
- Controls
- Copilot
- Corpus
- Cost Attribution
- Cost tracking
- Agent team
- CrewAI
- Cross-attention
- Cross-border data transfer
- Cross-encoder
- Cross-Modal Alignment
- Cross-modal embedding
- Curriculum learning
- Customer acquisition cost
- Customer health score
- Customer lifetime value
- Customer Success
- Data augmentation
- Data connector
- Data exfiltration
- Data flywheel
- Data leakage
- Data lineage
- Data localization
- Data loss prevention
- Data Minimization
- Data mixing
- Data parallelism
- Data poisoning
- Data privacy and PII masking
- Data processing agreement
- Data protection impact assessment
- Data residency
- Data retention policy
- Data usage rights
- Dataset datasheet
- Dead letter queue
- Decentralized multi-agent
- Decoding stage
- Decoder-only
- Decoder
- Decode
- Deep learning
- Automated deflection rate
- Delivery acceptance
- Denial-of-service attack
- Denoising objective
- Dense retrieval
- Deep pruning
- Developer experience
- Differential privacy
- Diffusion language models
- Diffusion models
- Diffusion Transformer
- Digital employee
- Direct Preference Optimization
- Directed acyclic graph
- Disaster recovery
- Distillation
- Distilled reasoning model
- Distribution shift
- Document parsing
- Document storage
- Double quantization
- Draft model
- DSPy
- Durable execution
- Dynamic batching
- Early departure
- Embed
- Embed
- Emergent behavior
- Encoder-decoder
- Encoder
- Enterprise sales
- Entity extraction
- Episodic memory
- Error budget
- Upgrade path
- EU AI Act
- Evaluation suite
- Evaluation
- Evaluation framework
- Exception handling
- Unauthorized agency
- Executor
- Exit plan
- Expert routing
- Explainability
- FAISS
- Fact-checking
- Fairness
- Faithfulness
- Rollback
- Feature flag
- Feature Store
- Federated learning
- Feedforward network
- User feedback loop
- Few-shot prompting
- File search
- Fine-tuning
- Cloud cost operations
- Flash Attention
- Foundation model
- Full fine-tuning
- FSDP
- Function calling
- General-purpose AI
- Generalization
- GPT
- GGML
- GitOps
- Golden dataset
- Golden path
- Governance, Risk & Compliance
- GGUF
- GPTQ
- Degradation
- Gradient descent
- Canary release
- Graph of Thoughts
- Graph-enhanced retrieval generation
- Greedy decoding
- Gross profit
- Fact Anchoring
- Group chat orchestration
- GRPO
- Grouped Query Attention
- Safety guardrails
- GUI agents
- Hallucination detection
- Hallucination risk
- Hallucination
- Handoff
- Harmful content filtering
- Haystack
- Helicone
- Hierarchical multi-agent
- Hierarchical Navigable Small World Graph
- High-risk AI systems
- HELM (Holistic Evaluation of Language Models)
- Hot reload
- Human evaluation
- Human in the loop
- Human in the loop
- Human in the loop
- Humans out of the loop
- Human-in-the-loop oversight
- Human oversight
- HumanEval (code generation evaluation)
- Hybrid multi-agent
- Hybrid pricing
- Hybrid search
- Hyperparameters
- Hypothetical document embeddings
- Ideal Customer Profile
- Idempotency
- In-context learning
- Incident Management
- Incident response
- Indirect prompt injection
- Inference endpoint
- Inference engine
- Inference
- Information Security Management System
- Infrastructure as Code
- IA3
- Instruction data
- Instruction fine-tuning
- Intent recognition
- Internal Audit
- Internal developer platform
- ISO 27001
- Jailbreak
- Jamba
- Task queue
- Tasks to be completed
- JSON mode
- Operator fusion
- KV Cache
- Knowledge base Q&A
- Knowledge distillation
- Knowledge graph
- Kubernetes
- KV cache
- Cache offloading
- Quantized KV cache
- LangChain Expression Language
- LangChain
- Langfuse
- LangGraph
- LangSmith
- Language model
- Large Language Model
- Latency monitoring
- Latency
- Latent Diffusion Model
- Layer normalization
- Layer-wise learning rates
- Layout analysis
- Linear attention
- LiteLLM
- LlamaIndex
- Large model applications
- LLM evaluation
- LLM evaluation platform
- LLM observability
- Large language model operations
- Localized deployment
- Logits
- Long-term memory
- Retrospective decoding
- Low-Rank Adaptation
- Loss function
- Low-Rank Adaptation
- Machine learning
- Mamba
- Managed service
- Masked language modeling
- MMLU (Massive Multitask Language Understanding)
- Maximum Marginal Relevance
- Medusa
- Membership inference attack
- Memory compression
- Memory
- Merge weights
- Measurement
- Milvus
- Minimum viable product
- Mixed precision training
- MoD
- Mixture of Experts
- Machine Learning Operations
- Model approval process
- Model Card
- Model Context Protocol
- Model drift
- Model fallback
- Model gateway
- Model inventory
- Model inversion attack
- Model monitoring
- Model poisoning
- Model Pruning
- Model registry
- Model risk assessment
- Model risk management
- Model routing
- Model safety evaluation
- Model serving
- Model validation
- Model version management
- Model
- Model file configuration
- Multi-agent consensus
- Multi-agent orchestration
- Multi-agent reflection
- Multi-agent system
- Multi-head attention
- Multi-hop retrieval
- Multimodal jailbreak
- Multi-query attention
- Multi-task learning
- Multi-tenant
- Multimodal models
- Multimodal RAG
- Multimodal
- Net revenue retention
- Neural networks
- Next-word prediction
- NF4
- Observability
- Ollama
- Onboarding
- Online inference
- Ontology
- OpenAI Assistants API
- OpenAI Responses API
- OpenClaw
- opencode
- OpenTelemetry
- Operational metrics
- OCR
- Orchestration
- Outcome-based pricing
- Overfitting
- OWASP LLM Top 10
- Paged attention
- Paged attention
- Parameter-efficient fine-tuning
- Parameters
- Parent-child chunking
- Round-by-round evaluation
- Permission boundaries
- Permission model
- Perplexity
- User profile
- Personal data
- Personally identifiable information
- pgvector
- Pilot
- Pinecone
- Pipeline parallelism
- Planner
- Planning
- Platform Engineering
- Plugin
- Policy as code
- Policy engine
- Positional encoding
- Post-training quantization
- Post-mortem review
- Pretraining
- Preference data
- Preference optimization
- Prefill
- Prefill
- Prefix caching
- Prefix Fine-Tuning
- Pretraining
- Principle of least privilege
- Private deployment
- Product-led growth
- Product-market fit
- Production drift detection
- Productized services
- Professional services
- Program-aided language model
- Prompt caching
- Prompt engineering
- Prompt injection
- Jailbreak
- Prompt management
- Prompt sensitivity
- Prompt fine-tuning
- Prompt version management
- Prompt
- Proof of concept
- Proof of origin
- Proximal Policy Optimization
- Pseudonymization
- Purpose limitation
- Qdrant
- Quantized low-rank adaptation
- Quality assurance
- Quantization-aware training
- Quantization
- Quantized low-rank adaptation
- Query engine
- Query rewriting
- Quota
- Rate limiting
- Reasoning-Action Paradigm
- Real-time agent
- Recovery point objective
- Recovery time objective
- Recursive retrieval
- Red team testing
- Benchmark case
- Reinforcement fine-tuning
- AI feedback reinforcement learning
- Reinforcement learning from human feedback
- Reinforcement Learning
- Verifiable reward reinforcement learning
- rejection sampling
- Relation extraction
- Representation learning
- Reranker
- Reranking
- Residual connections
- Residual risk
- Responsible AI
- Retrieval-Augmented Generation
- Retrieval-Augmented Generation
- Retrieval precision
- Retrieval recall
- Reward model
- Right of regret and revocation mechanisms
- Risk acceptance
- Risk appetite
- Risk assessment
- Risk owner
- Risk Register
- Reinforcement learning from human feedback
- Role-based access control
- Role-playing multi-agent
- Rollback plan
- Rollback
- Root cause analysis
- RMSNorm
- Rotary position encoding
- Router
- Run steps
- Run
- Runtime
- Software as a Service
- Safety fine-tuning
- Sales Agent
- Sales-led growth
- Sampling review
- Sampling
- Sandbox
- Scaling laws
- Scenario-based implementation
- Draft area
- Find the corresponding concept and contrast page by keyword
- Seat-based pricing
- Secrets management
- Secure aggregation
- Security incident response
- Security questionnaire
- Separation of duties
- Self-attention
- Self-consistency
- Self-service
- Self-supervised learning
- Semantic cache
- Semantic chunking
- Semantic function
- Semantic Kernel
- Semantic Memory
- Semantic routing
- Semantic search
- Semantic similarity
- Sensitive personal data
- SentencePiece
- Service Level Agreement
- Service Level Objective
- SOC 2
- Service as software
- Shadow AI
- Signature
- Skills
- Sliding window attention
- Sliding Window Chunking
- Slot filling
- SOC 2 Type I
- SOC 2 Type II
- Span
- Sparse retrieval
- Sparsity
- Speculative decoding
- Speculative editing
- Speculative prefix tree
- Speculative sampling
- State diagram
- State machine
- State space model
- Step success rate
- Streaming responses
- Structured output
- Structured output
- Structured pruning
- Sub-processor
- Supervised Fine-Tuning
- Supervised Learning
- Supervisor agent
- Supply chain risk
- SWE-bench
- SwiGLU
- Synthetic data
- System card
- System prompt
- Table extraction
- tamper-evident log
- Target module selection
- Task decomposition
- Task success rate
- Technical documentation
- Prompt optimizer
- Temperature
- Tenant isolation
- Tensor parallelism
- Test set
- Test-time compute
- Text generation inference
- Speech synthesis
- T5
- Third-party risk management
- Conversation thread
- Three lines of defense
- Throughput
- Time to First Token
- Time to value
- Token budget
- Token cost
- Token usage monitoring
- Token
- Tokenization
- Tokenizer
- Tokens per second
- Tool call accuracy
- Tool call latency
- Tool calling
- Tool output contamination
- Tool registry
- Tool selection
- Tool calling
- Top-k routing
- Top-k sampling
- Top-p sampling
- Tracking records
- Traceability
- Call chain tracing
- Training data exclusion
- Training set
- Trajectory evaluation
- Transformer
- Transparency
- Tree of Thoughts
- Tree search
- Triple
- TruLens
- Trustworthy AI
- Noise predictor/UNet
- Unit economics model
- Unstructured Pruning
- Unsupervised Learning
- Usage-based pricing
- Usage policy
- User Proxy Agent
- Validation set
- Vector database ecosystem
- Vector database
- Supplier due diligence
- Supplier evaluation
- Vendor Lock-In
- Version control
- Video understanding
- Vision-language model
- Vision Transformer
- vLLM
- Vocabulary
- Voice agent
- Watermark
- Weaviate
- Web agents
- Weight tying
- Weights & Biases Weave
- Width pruning
- Word embeddings
- Workflow engine
- Workflow
- Working memory
- ZeRO
- Zero-shot prompting
Guides
- If an enterprise wants to apply AI to the business, which problem should it start with?
- What's the difference between AI Agents and workflows? How should enterprises choose for their projects?
- Which scenario should an enterprise AI Agent start with? Criteria and first steps
- How can large models more easily understand a professional services firm?
- What's the difference between an enterprise knowledge base and RAG? Why do answers still come out wrong when all the materials are there?
- How can sales, pre-sales and delivery share the same business context?
Insights
- How does AI in the enterprise move from "can answer" to "can take on the work"?
- In complex sales, how do you find, develop, test and maintain a Champion?
- What does each of the six MEDDIC sales steps need to prove?
- How do individual Agents gradually form organization-level collaboration?
- Why do professional Agents need object-based memory rather than just vector retrieval?
Comparisons
- Easily confused concepts, laid out on the same table and explained clearly
- Access control and permission models
- Access control and the principle of least privilege: one is the lock on the door, the other tells you how many keys the lock should have
- Adversarial evaluation and red team testing: one uses tools to attack automatically, the other asks people to manually think up devious tricks
- How the agent operates and who is accountable for the agent
- Agent orchestration and multi-agent orchestration
- Agent orchestration and orchestration
- The one that commands multiple agents and the execution flow
- Agent owner and risk owner: one manages how the car is driven, the other manages who pays if something goes wrong
- Assigning tasks to the right agents and coordinating multiple agents
- Agent runtime and OpenClaw
- The one that manages agent lifecycles and the engine that runs flowcharts
- Agent runtime and workflows
- A Self-Organizing Group vs. a Team with Divided Roles
- What is the difference between an Agent and an Agent Loop?
- Agents and Agent runtime
- Agents and AI assistants
- Agents that make decisions and autonomous driving that guards boundaries
- What is the difference between an Agent and a Copilot
- Agents and digital employees
- Agents and large model applications
- Agents and Runtime
- Agents and sales agents
- Specific communication protocols and general communication specifications
- AI-assisted workflows and AI-native workflows
- AI impact assessment and data protection impact assessment: one asks whom AI will harm, the other asks how data will leak
- After something goes wrong and before it goes wrong
- AI incident and safety incident response
- The rules for how companies use AI and the rules for how agents do work
- Those who handle incidents and those who set the rules
- AI risk management and AI risk management frameworks
- AI risk management and model risk management
- What is the difference between AI safety and AI governance
- AI job replacement rate and automation deflection rate
- Those who test rules and those who test skills
- Alignment evaluation and model safety evaluation
- ARR and NRR: one measures annual rent, the other measures whether the bucket is leaking
- Anonymization and pseudonymization
- Recording what happened and recording how to investigate it
- Audit trail and audit evidence: one leaves footprints, the other shows what the footprints can prove
- Auditability and audit evidence: one is the structure that lets people inspect the system, the other is the slip of paper found during inspection
- Authority boundaries and autonomy levels
- Boundaries of authority and boundaries of permission
- One scale mark and one ruler
- A self-driving system and its levels of autonomy
- Benchmarks and benchmark contamination: one is the ruler in your hand, the other is the ruler someone has already seen in advance
- One problem set and one exam
- Evaluation benchmarks and golden datasets: one is a full set of exam papers plus scoring rules, the other is a book of standard answers
- Dual encoders and cross encoders
- The entire old-to-new switch and small-scale trial rollout
- The ones that go to web pages to fetch things and the ones that operate web pages for me
- Browser tools and web agents
- View cost routing and full-dimension routing
- Canary release and blue-green deployment
- Start with a small batch and gradually roll out to users
- Put one car in first to see versus releasing section by section in proportion
- Chain of thought and scratchpad
- Pushing one path all the way through vs. evaluating multiple paths at once
- Chunking and parent-child chunking: one is cutting tofu, the other is cutting the tofu while still keeping the original plate of food
- Chunking and retrieval-augmented generation
- Cutting materials and finding materials
- Chunking and semantic chunking: one cuts long text into segments, the other cuts by meaning rather than word count
- The big family of methods that split long text into chunks, and the one among them that slides along at a fixed step size
- Churn rate and net revenue retention
- Circuit breaker and degradation: one cuts the power, the other turns off the lights and opens the window
- How is a customer service assistant built with opencode different from opencode itself?
- Compliance logs and audit logs
- Computer use and GUI agents
- The engineering of managing the information environment versus writing a single instruction
- Context engineering and retrieval-augmented generation
- Context contamination and data poisoning
- Whether what is fed into the context is accurate and complete
- Context precision and faithfulness: one measures retrieval relevance, the other measures whether the answer relies on evidence
- What to delete and how much it can hold
- Context recall and faithfulness
- Context windows and context engineering
- Context window and context length: one governs what can fit, the other governs how much can fit
- Do it as it comes versus wait until enough has accumulated
- It sounds like one thing, but open it up and there are two sets of switches
- Handrails and the entire safety management manual
- Cross-encoders and rerankers: one is a measuring tool, the other lines up the measurement results
- Customer acquisition cost and customer lifetime value
- Reading the pulse today and reviewing the medical records at year-end
- Customer success and customer health scores: one runs the whole journey, the other records the readings
- Data exfiltration and data loss prevention: one is an incident that has already happened, the other is a wall that prevents incidents
- Collect only what is needed and use it only for its original purpose
- Data privacy, PII masking and anonymization
- Mask sensitive fields and make statistical results hard to reverse-engineer into individuals
- Data Processing Agreement and Data Usage Rights
- Data processing agreement and sub-processors—one is the contract itself, the other is the parties listed in the contract
- Data residency and cross-border data transfer
- Where data lives and where data must stay
- Data retention policy and data minimization: one governs how long to keep data, the other governs not taking too much.
- Peer consultation versus superior command
- Automation deflection rate and intelligent customer service
- Dense retrieval and semantic search
- What is found by vectors and what is found by keywords
- Digital employees and software as a service
- Disaster recovery and business continuity: one keeps machines alive, the other keeps the business from stopping
- Tasks can resume from checkpoints and queue for execution
- Embedding and embedding
- Turning content into numbers and a warehouse to store those numbers
- The big one that turns everything into vectors, and the old master within it who turns only words into vectors
- A two-module architecture that reads everything before writing, and a single-module architecture that writes while only looking at the preceding context
- Enterprise sales and sales-led growth
- The EU AI Act and high-risk AI systems: one is the whole law, the other is a category marked out within it
- Evaluation and benchmarks
- Evaluation and LLM evaluation: one is the ruler, the other is using that ruler to measure a specific model
- Whether the answer follows the material and whether it addresses the question
- Rollback and model rollback
- Few-shot prompting and zero-shot prompting
- Fine-tuning and instruction fine-tuning: one is major surgery, the other is changing the way you converse
- Fine-tuning and supervised fine-tuning
- Full fine-tuning vs. parameter-efficient fine-tuning: one rewrites the whole book, the other sticks on a note
- General-purpose AI and foundation models: one depends on how regulators classify them, the other on how technical teams train them
- A perfect answer set annotated item by item by experts and a test sample set used for batch spot checks
- Governance frameworks and security certifications
- Graph-enhanced retrieval generation and knowledge graphs
- Those found by relationship graph and those found by multiple data forms
- Calculating the whole gross margin versus breaking it down to individual transactions
- Fact grounding and source citation: one checks truth, one checks provenance
- Factual grounding and faithfulness—one governs telling the truth, the other governs not making things up
- Agents that operate general interfaces and agents that operate web pages
- Hallucination detection and fact verification: one checks whether the model is making things up, the other checks whether the answer is correct
- Detecting hallucinations and the risk of evaluation hallucinations
- A real person sits down to judge quality versus using another large model to score
- Human in the loop and human on the loop
- Human-in-the-loop and approval flows: one is real-time correction, the other is sign-off in advance
- Hybrid search and dense retrieval
- Hybrid search and sparse retrieval
- Ideal Customer Profile and User Persona: one targets the buyer, the other targets the user
- In-context learning and few-shot prompting: one relies on the model to induce patterns itself, the other relies on you providing examples
- The general inference framework that gets models running, and the specific implementation built to accelerate text generation
- Inference engines and vLLM: one sets the standard, one handles the volume
- Infrastructure as code and GitOps: one writes machines into files, the other lets files drive machines automatically
- ISO 27001 and the information security management system: one is an exam standard, the other is daily homework
- Jobs to be done and user personas: one asks what to do, the other asks who does it
- KV cache and KV cache: the same thing called two different names in papers and in engineering docs
- KV Cache vs. prefix caching: one is the inference engine remembering every step, the other is the serving layer reusing someone else's memory
- Kubernetes and containers: one manages how goods are placed, the other manages how goods are packed
- LangGraph and workflow engines: one draws the storyboard, the other runs the theater
- Language models and large language models: one is a species, the other is the famous variant within that species
- Large language models and foundation models: one lives in the language channel, the other can sleep anywhere
- Latency and throughput
- On-premises deployment and cloud native
- Localized deployment and private deployment: two sets of books installed on the same machine
- Low-rank adaptation and low-rank adaptation: two ID cards for the same term
- Managed services and professional services
- Model Context Protocol and Agent Communication Protocol, with one end connecting models and tools, and the other end connecting agents and agents
- Swapping in a backup when the model fails vs. still making do when the system fails
- Continuously monitor model performance and detect production environment drift
- Model poisoning and data poisoning: one poisons the factory, the other adulterates the raw materials
- Model risk management and model risk assessment: one is an ongoing mechanism, the other is a one-time judgment
- Model routing and Agent routing: one sends the question to the right model, the other assigns the task to the right role
- Model serving and inference endpoints
- Model serving and inference engines
- The one managing the deployment entry point and the inference running the compute
- The process of reaching consensus and the process of debating each other
- Observability and LLM observability: one is monitoring for the whole building, the other is a camera in the elevator
- Onboarding and time to value: one paves the road, the other measures how far you walked to get there
- Online inference and batch inference
- What is the difference between opencode and OpenClaw
- opencode and Agent runtime
- What is the difference between opencode and OpenClaw
- opencode and the sales agent: one is the blacksmith's shop, the other is the blacksmith
- Operational metrics and task success rate: one looks at whether the machine runs well, the other at whether the machine does the right work
- Pure results-based payment and hybrid pricing
- Preserve hierarchical relationships and split by semantics
- How permissions are allocated and which line must never be crossed
- Data that can identify people and data that requires special protection
- Information that can identify an individual, and broader personal data
- Pilot and co-creation: are you here to test or to build together
- Planner and executor: one draws the roadmap, one presses the accelerator
- Post-training quantization and quantization-aware training: one is losing weight after the fact, the other is controlling weight from childhood
- Pre-training and fine-tuning: one lays the foundation, the other renovates the room
- In the prefill and decode phases, one takes in the question and computes full attention once, while the other moves forward one token at a time as it emits text
- In the prefill and decode phases, a whole batch of input is swallowed in one gulp, then spat out one character at a time
- Private deployment and data residency: one governs who touches the machines, the other governs where data sleeps
- Growing through the product itself versus driving deals through sales
- Product-market fit and time to value: one verifies whether you are doing the right thing, the other measures how long it takes to see results
- Priced delivery-type and long-term managed-type services
- Prompt caching and prefix caching: one stores the whole order, the other reuses the shared header
- Prompt engineering and prompt management
- Prompt injection and indirect prompt injection
- Prompt injection and jailbreaking
- Prompt leakage and jailbreaking: one is peeking at the script, the other is getting the actor drunk
- Prompt management and prompt version management: one governs process, the other governs history
- Prompt tuning vs. fine-tuning: one teaches the model a secret code, the other gives the model a new brain
- Adding a layer on the surface versus moving a little at every layer
- Prompts and prompt engineering: each brick is material, the full blueprint is engineering
- Prompts and system prompts
- Proving it can be done and the minimum deliverable product
- Proof of concept and pilot
- Quantized low-rank adaptation and quantized low-rank adaptation: two spellings, one technology, one set of management rules
- Quality assurance and delivery acceptance: one governs whether the process is right, the other governs whether the result is accepted
- Quantization and post-training quantization
- What is the difference between RAG and a knowledge base
- ReAct vs. planning: one reads the road while running, the other draws the map before setting out
- The reasoning-action paradigm and tool calling: one governs how to think, the other governs how to use
- Recovery Time Objective and Recovery Point Objective
- Benchmark Cases and Pilots: One Proves Others Have Done It Successfully, the Other Proves It Can Succeed for You
- Reranking and avoiding results that are too similar
- Reranking and reranker: one describes what was done, the other describes what was used to do it
- Retrieval-augmented generation and graph retrieval-augmented generation
- Retrieval-augmented generation and knowledge base question answering
- Retrieval-augmented generation and knowledge graphs: one finds materials from the warehouse, the other has already drawn the relationship map in advance
- Two names for retrieval-augmented generation: the same concept, two sets of applicable definitions and criteria
- Retrieval precision and context precision
- Retrieval precision and retrieval recall
- Retrieval recall and context recall
- Signing off to take on risk, and the risk that remains after handling
- Risk appetite and risk acceptance: one decides how big the waves must be before going to sea, the other says whether this boat can sail this time
- Risk assessment and model risk assessment: one scans the whole house, the other specifically checks AI
- The process of assessing risk and the ledger that records it
- Risk register and risk owner: one writes risks down, the other makes sure someone owns them
- Granting permissions by role vs. granting permissions by attribute
- Rollback and rollback plan
- When a program runs versus when an agent runs
- Sandbox and permission boundaries: one builds the isolation room, the other writes who can enter
- Seat-based pricing and usage-based pricing
- Making suppliers fill out forms versus actually vetting suppliers
- Self-service and product-led growth
- Segmenting by Meaning vs. Segmenting by Sliding Window
- Semantic routing and Agent routing: one routes by meaning, the other assigns by capability
- Service level objectives and error budgets
- Service level objectives and service level agreements
- A physical exam ticket and a photo kept on file
- Service as software and productized services
- Service as software and software as a service
- SOC 2 Type I and SOC 2 Type II
- One class of methods and one specific formula
- The fast-guess, slow-verify method and the small model responsible for fast guessing
- First practice against answers, then practice by scoring against preferences
- Tamper resistance and auditability, two different ledgers
- Task success rate and step success rate
- Adjust the entire distribution and keep only the part where probability is concentrated
- Time to first token and tokens per second
- How much is spent on models and how much remains after direct costs are deducted
- Token cost and usage-based pricing
- See how many tokens the model consumed and how much these calls actually cost
- Choosing the right tool and completing the task are two different metrics
- Whether tool calls are accurate and whether tool calls are fast
- Tool calling and function calling
- Tool calling and tool registry
- Tool invocation and tool selection
- Tool calling and tool use
- Tool output contamination and indirect prompt injection
- Tool registry and Model Context Protocol
- Things used to do work and things that do work for me
- Fixed selection of a few versus selection by cumulative probability
- A traceable link from design to implementation and an operational record that shows what happened at every step
- Training data exclusion and data usage rights: one door only governs what may not enter, the other governs what may be done once it is in
- Evaluate the entire task path and evaluate each round of response
- Trajectory evaluation and step success rate
- Transformer and attention mechanisms: one is the whole factory, the other is the most critical machine in the factory
- Usage-based pricing and metering
- Usage-based pricing and outcome-based pricing
- Usage Policy and Acceptable Use Policy: one governs how it may be used, the other governs what is prohibited
- Supplier due diligence and third-party risk management
- Vendor evaluation and vendor due diligence—one is about choosing whom to use, the other is about whether you checked thoroughly enough before using them
- Vendor lock-in and exit plan
- Workflows with fixed steps versus workflows where the agent chooses the path
- Workflows and workflow engines
Other public content
- First clarify the work you are doing, then decide which part the Agent should take on.
- Boss Agent: Make every opportunity's next step traceable with evidence
- The method already exists; the key is putting it into your daily judgment checklist
- Every key moment in a complete sales journey
- Technical approval is in—so why is procurement still stalled?
- Champion says "I'll help push it" — are they really pushing?
- The customer says "let's start talking price" — does that mean the deal is secure?
- After the demo the customer says "not bad" — are they really validating?
- While you sleep, who's checking your opportunities?
- You arrive at the office in the morning with dozens of customers — which one do you push first today?
- The customer said they were "interested" — what did this meeting actually confirm?
- Why do your opportunities always get pushed to "next week"?
- The client says "I want an AI system" — did this meeting align on scope?
- Meeting a customer this afternoon, where do I start looking through materials?
- After a visit, do you still remember what the client said and what they promised?
- Weekly review meeting: besides "gut feel," what else is on the table?
- 199 RMB AI practice community group
- First let one person get the work running smoothly, Then let agent collaboration connect the team.
- Put existing materials to use in the next professional delivery.
- 30 self-check questions: which part is most worth systematizing right now?
- Your competitors have already started Recommended by large models. Now, it's time to let the Agent Start bringing you business opportunities.
- From recording the sales process, to driving sales results.
- From free consultation to long-term managed operations, how to choose
- How Andy Ren Xu works
- Knowledge management system
- OpenClaw and business systems
- Pre-sales materials system
- Sales Harness
- 9 typical problems in B2B sales management
- Direction Intervention · Is the Customer Worth Pursuing
- Progress quality intervention · MEDDIC six-element assessment
- Fake progress detection · Surface heat vs. real progress
- Next Action · Five-Element Precise Action
- Pace intervention · countdown to key milestones
- Lost-deal review · Find the signals you should have spotted earlier next time
- Manager Review · First get the review materials in order
- Pipeline overview · Real health monitoring
- New sales acceleration · From novice to ready to sell
- Technical architecture and dependencies
- 3 types of ToB service business path examples
- AI transformation for consultants and trainers
- Turn interviews into evidence, and turn proposals into action.
- From facts to fulfillment, make the evidence behind legal work and its handoffs clearer.
- Be better prepared for every customer conversation, with stronger evidence to move each opportunity forward.
- Connect course exercises to the business, and bring learning back to the workplace.
- Exhibition service providers: don't miss the chance to reconnect with old clients
- Get one real piece of work running first, then decide on AI Agent investment