Choose a starting point
Foundations
Learn the AI basics, then follow the timeline. Start here if the concepts are new.
Strategy
Choose a use case, compare costs and value, and build a phased plan with review gates.
Practice
Try a worked first task with fictional data. Continue to the technical deep dive for validation methods.
Evidence
Check source passages, reported results, and corrections before quoting a claim in a decision.
About
Read about the author and editorial approach, suggest a correction, or explore further reading.
Foundations · AI basics · Everyone · ~17 min
Submissions, loss runs, policy wording, claims files, and broker correspondence are natural starting points for AI-assisted document work. The aim is a useful draft or structured record that a person can check against the source.
What this technology is, in brief
Large language models (LLMs), the technology behind ChatGPT, Claude, and Gemini, are software systems trained on enormous volumes of text that can read, summarize, draft, extract, classify, and reason over language. They can produce useful first drafts and structured records, but fluent output can contain errors. Review consequential work against its source material.
They are not databases, search engines, or calculators. The model generates an answer; connected search and retrieval tools can supply source material. That is why it can draft a coverage summary in seconds, and also why it can invent a plausible-sounding policy exclusion that doesn't exist. This failure mode is called hallucination, and managing it is the central discipline of using LLMs well.
Give the model relevant context
Context in, quality out. Give the model the task, relevant source material, and the standard its answer must meet. Context is one lever you control; model capability, retrieval quality, and verification also matter. Adding more text can make relevant information harder to find, as Anthropic's context-engineering guidance explains.
That is the overarching idea of this site, because the ways of giving a general model access to context have multiplied. Not long ago it meant pasting text into a chat box. Now a model can be handed your documents, templates, and data; loaded with reusable Agent Skills that package your procedures and domain knowledge; and connected to tools through the Model Context Protocol (MCP) or a provider's own browsing and search to look things up while it works: read a live web page, pull a filing, query a database, call an internal system, then reason over what it found. Tools can supply information published after training. Retrieved material can still be wrong, outdated, or contain malicious instructions, so check its source and relevance.
Make the assignment explicit. State the task, supply relevant documents, describe the expected output, and specify what the model should do when information is missing. Review the result before using it in a decision.
Context in, quality out
Distinguish information learned during training from material supplied for a task and material retrieved through tools. Each needs different checks.
Levels of context engineering
Choose a level according to how often the task recurs, how much the answer matters, and what the team can maintain.
-
1
Paste it in The submission or the wording, dropped into the conversation. Reach: one question, one person, once.
-
2
Structure it A reusable prompt carrying the format, a worked example, and the standard the output must meet. Reach: the same task, repeatably, across a team.
-
3
Retrieve it RAG over your guidelines and appetite guides, or a connected tool that fetches from outside them, so the answer is drawn from a named source and cites it. Reach: answers that stay current as the guidelines change, and are checkable.
-
4
Package it A versioned skill: the procedure, its scripts, and its checks, loaded only when the task calls for them. Reach: an organizational procedure that can be tested, corrected once, and reused everywhere.
Terminology, one line each
Definitions before the sections that use them. The three carrying the most weight below are context engineering, skill, and agent.
| Term | Plain-English meaning |
|---|---|
| LLM / foundation model | A general-purpose AI model trained on vast text; the "engine" (e.g., GPT, Claude, Gemini). Left to itself it answers the question in front of it and stops |
| Generative AI | AI that creates content (text, images, code) rather than just scoring or classifying |
| Prompt | The instruction you give the model; prompt quality drives output quality |
| Hallucination | Confident, fluent output that is factually wrong or fabricated |
| Context window | The amount of input and output a model can handle in one interaction; limits vary by model |
| RAG | Retrieval-augmented generation: finding relevant source material and providing it to a model when it generates an answer |
| Fine-tuning | Additional training on selected examples to adapt a model for a task or behavior |
| Agent | The same engine given a goal, tools, and a loop: it plans, acts, reads what comes back, and continues until the job is done or a limit stops it. What separates an agent from a general model is that scaffolding, not a stronger engine |
| Agentic | A workflow pattern in which an agent plans, acts, observes results, and continues within defined boundaries |
| Artifacts | Finished outputs an agent produces as standalone files or apps (documents, charts, webpages) rather than disposable chat replies; they can be saved, shared, and revised |
| Tool use | Letting the model call something outside itself while it works: a web search, a database, a script, an internal system |
| MCP | Model Context Protocol: the open standard for connecting a model to those tools and data sources, so the same connector works across products |
| Context engineering | Designing what the system sees: instructions, source material, examples, state, and retrieved evidence |
| Skill | A packaged, versioned procedure the model loads when a task calls for it: the steps, the scripts they run, and the checks that must pass |
| Human-in-the-loop | A named person approves consequential output or reviews exceptions surfaced by the harness |
| Guardrails | Technical and policy controls that constrain what the AI can see, say, and do |
| Token | A unit of text or other input processed by a model; text tokens can be words, word fragments, or punctuation |
| Machine learning | Methods that learn patterns from data; includes predictive models and generative models |
Follow the evidence · worked example
From a question to an answer you can check
Retrieval supplies source material; review checks the answer against it. This fictional example shows the process, using authored sample responses and an invented guideline.
- AskDefine the decision and its limits
- RetrieveFind the relevant source passage
- DraftConnect each claim to the source
- VerifyCheck the meaning and missing facts
01 · A precise assignment
The question
Does this coastal property submission need referral under our current guideline?
The instruction
Use the approved guideline. Cite the relevant clause. If a required fact is missing, say what to obtain.
Without the guideline, a fluent answer cannot establish your company's rule.
02 · Put the rule and the account facts together
Fictional guideline · approved version · §4
Properties within five miles of the coast require referral to the coastal underwriting team.
Fictional submission · location schedule
Distance to coast: not supplied.
A retrieved clause helps, but the account fact needed to apply it is still missing.
03 · A draft with a visible boundary
Illustrative response · not a live AI result
I cannot determine whether referral is required. Section 4 requires referral within five miles of the coast, but the submission does not give that distance. Obtain the location evidence before applying the rule.
The answer names the rule, cites its location, and keeps the unresolved fact visible.
04 · Verify before using the answer
Compare with the source
Does §4 exist in the current approved version? Does it say five miles, and does it require referral?Check wording and version
Resolve the exception
The distance is still unknown. Route the file to a person to obtain the missing evidence; do not infer eligibility.Keep the decision with its owner
A citation is a route back to evidence. It does not by itself prove that the answer is correct.
Teaching example only; the five-mile rule is invented. Retrieval can miss clauses, and generated answers can misread or miscite the material they receive.
Make recurring procedures reusable
A skill packages instructions, scripts, and resources for reuse. The host loads its instructions when relevant; the file persists between runs. That gives a team a procedure it can version and improve, but does not guarantee that the model will select it or follow it correctly. Test each revision before making it the team's default.
Skills mix reasoning with determinism
In a workflow, code determines the sequence; in an agent, the model can choose the next action. Both can combine model calls, scripts, checks, and human review. Anthropic's workflow and agent patterns describe this distinction. A skill makes a procedure reusable; the surrounding system must enforce its permissions and release checks.
Two ways to encode a procedure
Code can fix the route through a task, or a model can choose the next step. Either design needs an explicit path for exceptions.
Deterministic pipeline
Design an exception path: reject an unsupported input or send it for review.
Agent with a skill
Check the model's choices as well as its output. Versioned scripts can make fixed calculations repeatable.
What it does well, and where it fails
Five useful tasks and five failure modes to check
Match the control to the failure. Put calculations in tested code, evaluate model judgments, and keep a named reviewer for consequential output.
Useful starting points
- Reading and summarizing large volumes of documents
- Drafting, where the model produces the first version and a person produces the final one
- Extracting structured data from unstructured text
- Explaining, translating, and reformatting information
- Writing and reviewing code
Failure modes
- Calculation errorsGenerated answers can contain arithmetic errors. Review the method and use tested calculation tools. Use tested code
- They are non-deterministicThe same question can return a slightly different answer, which is what makes reproducibility and audit hard. Test repeated runs
- They fabricateA fluent answer can contain invented facts. Require source passages and compare each claim with them; a valid output format does not establish factual accuracy. Evaluate and monitor
- They can be manipulatedPrompt injection: instructions hidden in a document the model reads, written to be obeyed rather than summarized. Least-privilege tools and separating instructions from untrusted content limit what a successful injection can reach. Evaluate and monitor
- Biased outcomesModel output can reflect bias in data and system design. Test relevant groups and review the applicable requirements in Governance. Evaluate and monitor
These are controls to test, not guarantees. Code can repeat a wrong calculation, and a checker can miss a false claim. Measure errors on representative work before reducing human review.
Build checks into the workflow
Separate fixed calculations from model judgment, and verify both:
- Put arithmetic and data manipulation in tested code. Review the formula, inputs, units, and edge cases. Keep the code and data needed to reproduce the result.
- Evaluate variation in model output. A versioned skill fixes the procedure you supply, but model decisions can still vary. Run representative tasks repeatedly, record errors, and enforce failed-check stops in the surrounding system.
For fabrication, compare the answer with its sources. For prompt injection, restrict tool access and enforce approval before consequential actions. For bias, test outcomes across relevant groups. Sample accepted outputs as well as flagged exceptions so you can see what the checks miss.
Agent evaluation guidance describes repeated trials and complementary code, model, and human graders. Use these methods to assess the whole workflow; packaging instructions is only one part of it.
Put progress in context
A model release, a carrier deployment, and a regulatory deadline establish different things. Compare each with the decision you need to make: whether a tool can perform the task, whether a team can operate it, and which requirements apply.
The timeline separates technology milestones, insurance adoption, and regulation. It also distinguishes documented events from future scenarios. Use that context when considering the integration phases.
Continue from the foundations
Next, read the timeline to put the concepts in historical context. Then explore insurance use cases and build a strategy, or try the worked exercise. Before using company data, check your approved tools and the rules in Governance.