Every technique, by level
A broad selection of practical language-model techniques, grouped by autonomy level. These pages are sourced explanations with illustrative runs, not measured performance reports. The selection is maintained as the field changes; it is not an exhaustive list of AI methods.
Conventional software
Use ordinary code, search, forms, or a task-specific statistical model when they solve the problem. No generative model is required; classical machine learning can belong here too.
Direct prompting
Give the model instructions and receive a response. A conversation repeats this interaction, with a person directing each turn. Prompting, structured output, reasoning, and multimodal inputs can all fit this pattern.
Images, audio and video
SourcedGiving the model images, audio, video and documents, and getting them back.
Added context
Add relevant documents, retrieved passages, or stored information to the current request. This supplies context beyond the model’s training without retraining it. Missing, stale, or misleading material can still produce a poor answer.
Retrieval-augmented generation (RAG)
MeasuredSearching your documents and giving the results to the model.
Knowledge graphs and GraphRAG
SourcedStoring facts as entities and relations, for questions that span several documents.
Workflows
Connect model calls through predefined steps, branches, checks, and retries. A model can classify an input or evaluate a result to route the workflow; software still defines the available paths.
Write and check
SourcedOne prompt writes, another checks, and the loop repeats until the check passes.
Tool use
The model can request a search, calculation, code execution, or an action in another application. Software enforces permissions, performs the action, and returns the result. Tool use alone does not create an ongoing agent loop.
Agent loops
The model uses the goal and observed results to choose an action, revise its approach, or finish. Software executes tools and enforces permissions, approvals, and stopping limits. A run can stop because it is complete, blocked, or out of budget.
The agent harness
SourcedEverything around the model in an agent: the loop, tools, context handling, permissions, caps and sandbox.
Teams of Agents
Agents divide, coordinate, or review work across separate contexts. A coordinator can combine their findings, and the agents may use the same model or different models. Coordination adds overhead, and separate reviewers can still make correlated mistakes.
Always-on agents
Saved state, schedules, and events let an agent start or resume work without a fresh chat message each time. The model need not run continuously, and a dedicated computer or agent team is optional. Permissions, human approvals, monitoring, and stop controls still apply.
Topics at every level
Five topics cut across all the levels. Each has its own set of pages.
Evaluation frameworks
SourcedThe tools that run test sets and graders for you, and what to check before trusting their numbers.
Fine-tuning and adapters
SourcedTraining a model further on your own examples, in full or with small adapters such as LoRA.
Synthetic data
SourcedUsing a model to write training or test examples, and checking them before they are used.
Guardrails
SourcedChecks on what goes into a model and what comes out, and the limits of those checks.
Red teaming
SourcedAttacking your own system on purpose, before someone else does, and turning what you find into tests.
Observability
SourcedRecording what each run did, so a bad result can be traced to the step that caused it.
AI gateways
SourcedOne entry point in front of several model providers, for keys, routing, limits, fallback and logs.
Cost optimization
SourcedSpending fewer tokens and less time for the same result: caching, batching, smaller models, shorter context.
Running models locally
SourcedRunning open-weight models on your own hardware: what fits, quantization, and what you give up.
Briefing: saying what you want
SourcedSaying what you want clearly enough that the model does not have to guess.
Deciding what to hand over
SourcedDeciding which parts of a task to hand to a model and which to keep.
Calibrating trust
SourcedLearning, from results over time, how much to rely on a model without checking.
One idea across several levels
A thread is a word people use for two different techniques. Its page says which is which and where each one lives.
Who approves what
SourcedA person never leaves; what they hold changes from the answer, to the action, to the rules the actions run under.
Checking the work
SourcedEvery level has a way to be wrong that the level below could not be, and a check that costs less than the mistake.
What the model sees
SourcedA model knows what it was trained on and what is in this request; everything else is a choice somebody made.