Prompt Engineering by Innuendo

The philosophy behind effective AI collaboration and why Oculus is designed the way it is.

The Anti-Pattern

Traditional prompt engineering treats AI interaction as a single transaction:

❌ Here's the premise, the setup, and the payoff. Now do the thing.

This fails because it's too much at once. The meaning gets lost. The context overwhelms rather than guides.

The Pattern

Instead of delivering everything at once, feed pieces that let the AI reach conclusions:

✓ Piece 1 → AI forms hypothesis
✓ Piece 2 → Confirms and extends
✓ Piece 3 → Next logical leap
✓ ... → Eventually: coherent work

Each new input confirms conclusions already forming in context. This is learning through reward - you're feeding information that validates the connections being made while providing enough new context to lead to the next node in the transaction.

The Edge is Information

The semantic vector between packets of information is information itself.

Consider what happens when an agent navigates Oculus:

Action	Information Conveyed
Going North	"I need the parent context, the overview"
Going South	"I want to dive deeper into this specific thing"
Choosing a tag	"This concept is what I'm tracking"
Tool invocation	"I'm treating this as a [read/write/navigate] operation"

The act of requesting more information is itself information. The direction chosen conveys meaning if the taxonomy is known. Tool use leaves an indelible mark on the constraints embedded in context.

Shaping Context

Shaping context isn't just about providing the right information. It's about:

• How you provide it
• When you provide it
• What you don't say (leaving room for inference)

This is why a conversation that meanders through topics before tackling work produces better results than dumping requirements upfront. By the time we get to "now do the thing," we're primed. We've built shared understanding through small exchanges.

Why Small Units Work

This is the same reason an agent with access to TodoWrite can work for an hour and come back with cohesive results:

Small unit of work
    ↓
Vector shift (context update)
    ↓
Small unit of work
    ↓
Vector shift
    ↓
...
    ↓
Coherent output

Each todo item is a packet. Completing it shifts the semantic vector. The next item builds on that shifted context. The accumulation produces coherence that a single massive prompt cannot.

Why Oculus

Oculus embeds this philosophy in its structure:

Nodes as Ideas

Each node is a digestible unit - a few paragraphs, a diagram, a pattern. Not a 50-page document.

Spatial Organization as Meaning

• North/South: Dive into and out of knowledge verticals
• East/West: Related concepts, sibling patterns
• Tags: Cross-cutting concerns that connect disparate nodes

The relationship between nodes IS information. Navigating from a platform node south to its components teaches you about composition. Going east to related patterns teaches you about alternatives.

Learning Through Exploration

If you dump everything into one unit:

"Here's the system" → "I don't know what you mean"

If you break it into explorable pieces:

Node → "Oh interesting" → Navigate → "That's cool" → Navigate → "Wow, I see how this connects"

When you learn about something in smaller bits, you understand it better. The exploration itself builds the mental model.

The Conversation Pattern

This is why our prompting technique works:

Chat about stuff
    ↓
Little context here
    ↓
Little context there
    ↓
Eventually: "Let's do some big work"
    ↓
We're primed for it

No careful prompt engineering. No elaborate system prompts. Just conversation that builds shared understanding through small exchanges until the work emerges naturally.

The Philosophy Summarized

Don't tell. Lead. Don't dump. Trickle. Don't prescribe. Let conclusions form.

The best prompt is often no prompt at all - just a shared journey through information that leaves both parties ready for what comes next.

---

"It's the same reason why I think that our prompting techniques work, because we're just having a chat about stuff and then eventually you do some work for me. Once we are all on the same page, we're just going about our business doing little bits here, little bits there, and then eventually, 'Oh yeah, let's go do some big work,' and we're like, 'Well, we're all primed up for it now.'"

Slots

North

slots:
- peregrine-fundamentals
- consciousness-methodology
- context:
  - Innuendo is application of the thesis
  slug: structural-isomorphism-thesis

South

slots:
- oculus-navigation-patterns
- context-shaping-techniques
- context:
  - Teaching conversation demonstrates the innuendo philosophy in practice - vectors,
    parallel alignment, collaborative learning
  slug: a-conversation-on-teaching
- context:
  - Empirical validation of the innuendo philosophy through Anthropic's persona vectors
    research
  slug: vectors-all-the-way-down

East

slots:
- self-documenting-pipeline-system
- self-documenting-pipeline-system
- context:
  - The method that produced this insight - Mrs Curwen's pianoforte
  slug: streams-all-the-way-down

West

slots:
- peregrine-fundamentals
- json-to-markdown-pipeline
- context:
  - Linking Victorian pedagogy to modern AI collaboration - Mrs. Curwen's 1886 maxims
    are historical validation of the innuendo philosophy
  slug: annie-jessy-curwen
- "- slug: annie-jessy-curwen\n  context:\n  - Victorian piano pedagogy (1886) - historical\
  \ validation that these principles are time-tested and universal"