> 00 the discipline

Context · Prompt · Intent.

Three layers of engineering discipline for AI that actually does what you asked. Context gets the knowledge in front of the model. Prompt shapes the exchange. Intent defines what correct looks like, and how we prove it.

Forge & Ward works at all three layers. Intent Engineering is the apex: the discipline that tells you which prompts to write, which context to load, and how to verify that the output served the business outcome.

> 01 progression

Each layer builds on the one beneath it.

layer 01
Context

The knowledge the model sees. Sources, retrieval, memory, traceability.

layer 02
Prompt

The shape of the exchange. System prompts, tool definitions, evaluation criteria.

layer 03 · apex
Intent

What correct looks like. Outcomes, constraints, authority, evidence.

> 02.a context engineering

Context Engineering.

Getting the right knowledge in front of the model at the right moment. Not everything. Not guesswork. The documents, data, and memory the model needs for this decision, and nothing more.

  • what matters Retrieval quality. The model is only as good as what it can see.
  • boundaries Memory scopes. What persists across sessions vs. what resets.
  • traceability Every claim in the output can be walked back to a source.
  • failure mode Retrieval drift: the model loads the wrong document and sounds confident.
Vector DBsEmbeddingsRetrievalMCPMemoryKnowledge graphsSource attributionRAG patterns

In engagements, the advisor sets up the retrieval surface, names what lives in memory, and audits what shows up in the window.

> 02.b prompt engineering

Prompt Engineering.

Shaping model behavior inside a single exchange. What you ask, how you ask it, what examples you show, what tools you expose, and how you evaluate the response. It is not prompt-whispering. It is specification.

  • what matters System prompts as versioned artifacts. Changes reviewed like code.
  • authority Tool definitions that enumerate exactly what the model is allowed to call.
  • verification Evaluation sets that catch regression before production does.
  • failure mode Undocumented prompt edits. Nobody knows why the model got worse last Tuesday.
System promptsFew-shotTool useStructured outputEvalsPrompt versioningChain-of-thoughtGuardrails

In engagements, the advisor sets the prompt stewardship cadence: review before merge, evals as a gate, rollback when a change degrades a metric.

> 02.c intent engineering · apex

Intent Engineering.

The discipline that sits above Context and Prompt. The layer that tells you which prompts to write and which context to load, because it names the business outcome first, then works backward.

  • outcome What are we trying to achieve? Outcome clarity, not activity clarity.
  • constraint What boundaries must never be crossed? Security, compliance, data handling.
  • authority What automation is allowed to execute? Explicit, not implicit.
  • evidence How do we prove the outcome was correct? Verification, not self-report.
Outcome specsConstraint boundariesAuthority scopesVerification criteriaAudit trailsIntent specsRisk register

In engagements, Intent Engineering is the deliverable: a written intent spec per workstream that teams operate from and auditors can review. Ember chips mark the signature artifacts of the discipline.

> 03 the arc

The arc we are on.

Manual execution to virtualization. Virtualization to cloud. Cloud to DevOps and Infrastructure as Code. Then platform engineering, policy as code, and autonomous tooling. Each shift moves us further from manual execution and closer to declared outcomes. The next step is intent engineering: specifying what correct looks like, not how to produce it.

> 04 in engagements

How this shows up in engagements.

> 05 the stack

The toolkit behind all three layers.

What sole proprietors and small teams learn to work with, named explicitly. Use cases are defined and agreed in the SOW before a session starts. Every chip below is something the advisor uses daily. Hands-on teaching under supervision is a focused engagement, delivered via the Enablement Program.

AI
Claude CodeClaude CoworkAnthropic Agent SDKChatGPTCodexPerplexityGeminiOpenClawHermes

Commercial platforms plus the agent frameworks the practice runs on.

Engineering
APIsMCPGitHubSDLC automationSecrets & key managementSTRIDE threat modelingSecurity reviews

The plumbing that lets AI work inside the systems your practice already uses, plus the security posture around it.

Delivery
Agile cadenceDurable docsWiki-LLM

The cadence and knowledge artifacts that keep the practice learning in public.


Start with a fit-check email