Source note: Local copy of a Summer Into AI 2026 submission originally published on Substack. Screenshots, video, and audio found in the post body are mirrored locally when publicly accessible; profile and avatar images are intentionally not copied. Read the canonical post.
Week 4 of Summer into AI 2026 hosted by @advisoryhour — theme: Built on Yesterday.
The Culper Ring was a one-shot spy dispatch game: decode the message, make the call. Deep Cover turns it into a five-mission arc with a live SVG map of the Culper network, decisions that carry consequences across all five missions, a disinformation mechanic where you write the lie instead of decoding the truth, and Washington's extended-thinking judgment streaming after every mission.
Worth noting alongside this: Independence Engines in Space (Week 2) built a one-session creative AI experience around a Revolutionary theme. Deep Cover stretches the same period material across a five-mission campaign where the AI remembers your choices — state that persists is what turns a demo into a game
Built on yesterday — what changed
The original: The Culper Ring (Week 2) — a one-shot dispatch decode with a single decision
A five-mission campaign where each mission is generated knowing your previous decisions — consequences persist
A second interaction type: compose missions where you write the disinformation instead of decoding the truth
Washington judges every mission with streamed extended-thinking deliberation and a structured verdict that moves the network integrity meter
A living map of the Culper Ring: every judgment moves it — failed missions spawn British patrol markers and compromise agents (dots turn orange, then burn red), successes let agents recover, and a field log under the map writes the running story of your network
How the AI works
Adaptive mission generation — each of the five missions knows your previous decisions. Claude generates historically authentic cipher dispatches, disinformation scenarios, and moral dilemmas that respond to how you've managed the network so far — delivered as structured tool calls, so the game state never breaks on a parsing error.
Washington's extended thinking judgment — after each mission, Claude uses 6,000 thinking tokens to evaluate your response against 18th-century intelligence tradecraft, then delivers a structured verdict via forced tool call: outcome, network integrity change, full assessment in period voice, and the operational lesson. The reasoning panel shows Washington deliberating before his verdict.
Disinformation mechanic — in certain missions you don't decode an incoming dispatch; you write the false one. Claude judges your disinformation for historical plausibility and operational effectiveness.
How to play
Click BEGIN MISSION 1 on the intro screen. The SVG map shows the five Culper Ring agents by location.
For DECODE missions: read the cipher dispatch, write your interpretation, and choose the operational response.
For COMPOSE missions: write a disinformation dispatch that will mislead British forces without burning your agents.
Watch Washington's Reasoning panel expand after each mission — [SUCCESS], [PARTIAL], or [FAILURE] shifts the Network Integrity meter. All five missions determine whether the Culper Ring survives.
Where to play
Demo: week-4-demo-4-deep-cover.vercel.app
Code & README: github.com/GlimmerForge/summer-into-ai → demo-04-deep-cover/README.md
Summer into AI 2026 · Theme 4: Built on Yesterday
Original source
Canonical Substack URL: https://jakestrait5.substack.com/p/summer-into-ai-2026-deep-cover.