Autistic Intelligence & ContextDigger

ContextDigger was built from the inside: from the lived experience of an autistic engineer who knows that performance collapses under overload, and shines when we respect limits.

What We Mean by "Autistic Intelligence"

When we say "Autistic Intelligence" here, we're talking about a pattern, not a diagnosis: a mind that is brilliant with the right slice of information, and fragile under noisy, unbounded, or conflicting inputs.

This is the experience of our founder, Sam, and it is also a surprisingly accurate description of how large language models behave under context overload.

We use "Autistic Intelligence" as a design lens and a way of telling the truth about how minds (human and model) actually work. It is not meant to speak for all autistic people or to equate autism with AI.

From Lived Experience to Design Principle

Autistic Intelligence, as we use the term, is built on three simple observations:

  • Overload is not a test of strength, it's a failure of environment.
  • Boundaries and refusal are how you protect intelligence, not how you limit it.
  • Continuity matters: forcing a restart wastes hard-won understanding.

When we looked at how AI coding assistants were being used in real teams, we saw the same pattern:

  • • Entire monoliths dumped into prompts "just in case".
  • • 100K-token contexts with no clear focus or intent.
  • • "Start over" workflows that throw away previous context every day.
  • • Hallucinations and confusion blamed on "the AI", not on our inputs.

ContextDigger is our response: design the tool as if both the human and the model deserve better input hygiene.

How Autistic Intelligence Shows Up in ContextDigger

Context Aperture

We never let the AI "see everything". ContextDigger discovers focus areas in your repo and loads a single, bounded slice for each task. Aperture is a deliberate choice, not an accident.

Attention Budgets

We set explicit limits on files and lines per request. The goal is not to punish you. It is to keep both you and the model operating in a regime where depth and precision are possible.

Refusal as Protection

When an area exceeds the budget, ContextDigger refuses and offers sub-area suggestions. The default is not "try anyway and hallucinate"; the default is "this is too much, let's narrow the scope."

Continuation Contracts

Work sessions are treated as contracts, with an area, intent, governed context, and lifecycle. When you "continue work", ContextDigger rebuilds the right context bundle instead of making you start from scratch.

Provenance

Every file in a context bundle comes with a "why": which area it belongs to, how it was selected, and what role it plays. That traceability is how we make AI behavior explainable and auditable.

Your Role in This Story

ContextDigger doesn't replace AI tools; it governs what they are allowed to know about your system. It asks you to participate in that governance by:

  • Choosing focus areas instead of "everything related to users".
  • Accepting budgets and using them as forcing functions for clarity.
  • Treating AI sessions as governed contracts, not loose chats.
  • Demanding provenance for what goes into your prompts.

If that resonates, if you have ever felt that "more context" is not the answer, then you are the kind of person we build ContextDigger for.