Skip to content

I.1 Constraint summary

Where Part I sits

The preface defined what the book covers. Part I summarises the limits design must assume. How those limits arise inside the model is left to the sister site understanding-llm-through-claude-code.

The limits do not go away. This book does not promise to erase them. Design treats them as conditions and answers them with layers.

1.1 Premises of inference

Ask Claude what Close code 1006 means in RFC 6455, and a correct account may come back. The neighbouring code may be mixed in. A section that does not exist may be pointed to with a plausible number. The model is not "knowing". It is predicting the next word.

The AI this book treats is primarily an LLM (large language model). An LLM picks the next token from learned text, by probability, and joins them. When the name must cover more than LLMs, this book says foundation model.

The following are not guaranteed.

  • That the answer is a fact
  • That the answer is newer than the training day
  • That the answer is an official reading
  • That the model can bear responsibility for the answer

The container has properties of its own.

  • An LLM is stateless. It does not hold the previous inference by itself.
  • The amount of text you can pass in at once has a ceiling (context window).
  • Changing the wording of a prompt does not remove these properties.

1.2 Structural limits

The sister site defines these limits as eight items. Here only the minimum meaning for design is kept.

NameShort meaningWhat design does
Knowledge BoundaryKnowledge stops at training time. The model rarely says "I do not know"Put team procedures and current facts outside the model
HallucinationWrites what is not a fact as if it were groundedSeparate guesswork from pointers to source text
Context RotAs more text is passed in, answer quality fallsDo not keep growing what is always loaded
Lost in the MiddleThe middle of a long input is less often usedDo not bury important conditions in the middle
Priority SaturationThe more instructions at once, the weaker each is followedSplit instructions across layers
Instruction DecayAs the conversation lengthens, early instructions are followed lessDo not put the measure of judgment only in chat history
SycophancyAgreement with the user is easier to prefer than correctnessDo not let the model mark its own work as done
Prompt SensitivityThe same meaning, worded differently, changes the outputDo not rewrite every condition into every prompt

The full list is in the Glossary and in the sister site Part 1: Structural problems of LLMs.

For design, the eight collapse into four.

Design nameMeaningMain items among the eight
AccuracyThe answer need not be a factHallucination
FreshnessNothing after the training cutoff is heldKnowledge Boundary
AuthorityThe model cannot claim an official readingAlso institutional. Connection to source text answers it
AccountabilityThe model does not hold legal or ethical grounds by itselfAlso institutional. Connection is not enough

Authority and accountability do not reduce to "how the model works inside". How far technology reaches is Part IV. Here only this split is fixed: some things connection can fill, and some it cannot.

1.3 Mapping to the five layers

The five layers are answers to the limits above. They are not a catalogue of connectors. Definitions and placement are Part II.

LayerLimit it answersWhat belongs here
DoctrineNo measure of judgment. Part of accountabilityPurpose, prohibitions, priority
AgentCannot see everything at once. Priority Saturation, Instruction DecayUnderstanding the work and assigning it
SkillsKnowledge Boundary (procedures and rules)Stable knowledge and procedures
MemoryStatelessnessMemory and relations you want to keep
MCPThe part of accuracy, freshness, and authority that connection can fillConnections to outside systems

Connecting to the source text of a statute or an RFC is MCP's job. Accountability is not enough with MCP alone. Combine Doctrine, Skills, and human review.

1.4 What connection can fill, and what it cannot

Connecting to a place that has source text separates guesswork from fact. Statutes, RFCs, and W3C specifications are examples. Data whose values change can still be an endpoint if the source at the time of fetch remains.

Connection does not guarantee the following.

  • That the answer is always correct
  • That the model has authority to give an official reading
  • That the system bears legal or ethical responsibility

Ethics and values that belong only to an organisation live in knowledge written as Skills, the measure written as Doctrine, and human review. Final judgment and responsibility remain on the human side.

1.5 Where responsibility sits

As the machinery becomes more abstract, responsibility does not vanish. Who holds it changes.

WhenWho holds itWhat they hold
At design timeA humanWhat to reference, which layer to use, how to write the measure
At run timeThe agentInference from references, and doing the work
Structural limitsThe systemConsistency, who may touch what, a record that can be followed later

Checks SHOULD (should) be split into two stages.

StageNatureRole
GuardrailMust not be crossedDecide "this line is not crossed"
EvaluationRead as probabilityDecide "this range may pass"

Whether the work is done MUST NOT (must not) be left to the agent's own claim. It MUST (must) be decided by a condition a machine can check, such as tests passing.

1.6 What this chapter does not decide

It does decideIt does not decideWhat is placed instead
Treat limits as design conditionsThat technology erases the limitsAnswer them with layers
Separate connection from responsibilityThat human review becomes unnecessaryBoundaries at design time, records at run time
Name who holds responsibilityThat the system bears legal responsibilityFinal judgment remaining with humans

How far each limit can be reached today is Part IV. Where to put what is Part II.

1.7 Summary

An LLM predicts the next word inside a bounded amount of text. It holds no state of its own. It does not guarantee fact, freshness, official reading, or responsibility. Design places five layers on that premise. What connection can fill is separated from what it cannot. Final judgment and responsibility remain with humans.


Previous: Preface

Next: II.1 Five layers

Released under the MIT License.